![\"Data,](\"https://media.springernature.com/lw900/springer-static/image/art%3A10.1038%2Fs41598-017-07599-6/MediaObjects/41598_2017_7599_Fig1_HTML.jpg\")
Post your (freely available) external data links here.
", "messages": [ { "id": 269233, "postDate": "2018-01-16T14:01:09.780Z", "content": "Post your (freely available) external data links here.
", "rawMarkdown": "Post your (freely available) external data links here.", "votes": 50 }, { "id": 269406, "postDate": "2018-01-16T19:13:33.520Z", "content": "Some potential resources with related datasets :
\n\nA comprehensive tutorial with selected use cases\nhttp://www.andrewjanowczyk.com/use-case-1-nuclei-segmentation/ (dataset : http://andrewjanowczyk.com/wp-static/nuclei.tgz)
\n\nA Dataset and a Technique for Generalized Nuclear Segmentation in Histological Images\nhttps://nucleisegmentationbenchmark.weebly.com/
\n\n2D images of 3T3 and U20S cells hand-segmented to show nuclear boundaries\nhttp://murphylab.web.cmu.edu/data/#2DNuclei (dataset : http://murphylab.web.cmu.edu/data/2009_ISBI_Nuclei.html)
\n\nDigital Pathology : Nuclei Segmentation MICCAI 2017 Challenge\nhttp://miccai.cloudapp.net/competitions/57 (currently not responding)
\n\nMitochondria segmentation dataset\nhttps://cvlab.epfl.ch/data/em
", "rawMarkdown": "Some potential resources with related datasets :\n\n**A comprehensive tutorial with selected use cases**\nhttp://www.andrewjanowczyk.com/use-case-1-nuclei-segmentation/ (dataset : http://andrewjanowczyk.com/wp-static/nuclei.tgz)\n\n**A Dataset and a Technique for Generalized Nuclear Segmentation in Histological Images**\nhttps://nucleisegmentationbenchmark.weebly.com/\n\n**2D images of 3T3 and U20S cells hand-segmented to show nuclear boundaries**\nhttp://murphylab.web.cmu.edu/data/#2DNuclei (dataset : http://murphylab.web.cmu.edu/data/2009_ISBI_Nuclei.html)\n\n**Digital Pathology : Nuclei Segmentation MICCAI 2017 Challenge**\nhttp://miccai.cloudapp.net/competitions/57 (currently not responding)\n\n**Mitochondria segmentation dataset**\nhttps://cvlab.epfl.ch/data/em\n", "votes": 30, "replies": [ { "id": 281896, "postDate": "2018-02-13T08:40:04.667Z", "content": "Thanks for sharing!
", "rawMarkdown": "Thanks for sharing!", "votes": 2 }, { "id": 284570, "postDate": "2018-02-17T15:50:27.130Z", "content": "thank you for great references!
", "rawMarkdown": "thank you for great references! " }, { "id": 305078, "postDate": "2018-03-28T13:14:12.490Z", "content": "Thanks for these links. Very helpful. The Github of the second set (Histological Images) is also useful.
", "rawMarkdown": "Thanks for these links. Very helpful. The Github of the second set (Histological Images) is also useful.", "votes": 1 } ] }, { "id": 288797, "postDate": "2018-02-28T07:45:56.750Z", "content": "I'll use fixed stage1_train masks https://github.com/lopuhin/kaggle-dsbowl-2018-dataset-fixes
", "rawMarkdown": "I'll use fixed stage1_train masks https://github.com/lopuhin/kaggle-dsbowl-2018-dataset-fixes", "votes": 17, "replies": [ { "id": 288844, "postDate": "2018-02-28T10:16:46.833Z", "content": "Thanks for sharing this, saving others lots of work!
", "rawMarkdown": "Thanks for sharing this, saving others lots of work!", "votes": 1 }, { "id": 288914, "postDate": "2018-02-28T14:15:56.263Z", "content": "@Konstantin, If you will really use your fixes instead of raw masks, you have no chance to win this competition. First of all you should ask yourself and answer the questions, why some nuclei are not separated and why some others are omitted. Did you do it? I am not a biologist and I do not know the answer for the first question (yet), but I think I know the answer for the second one. E.g. I am 99% sure, your fix regarding the image named a7f7... (adding one more nucleus at the left bottom) is wrong.
", "rawMarkdown": "@Konstantin, If you will really use your fixes instead of raw masks, you have no chance to win this competition. First of all you should ask yourself and answer the questions, why some nuclei are not separated and why some others are omitted. Did you do it? I am not a biologist and I do not know the answer for the first question (yet), but I think I know the answer for the second one. E.g. I am 99% sure, your fix regarding the image named a7f7... (adding one more nucleus at the left bottom) is wrong.\n", "votes": -4 }, { "id": 288919, "postDate": "2018-02-28T14:33:26.383Z", "content": "Sorry, I didn't understand the point about winning. I tried to make only fixes confirmed in the data quality issues thread. Edit: and yes, I'm really using them, although I didn't notice a significant change on the public LB.
\n\n\n\n\nE.g. I am 99% sure, your fix regarding the image named a7f7... (adding one more nucleus at the left bottom) is wrong.
\n
You may be right - I was basing my fix on this comment by Anne Carpenter: https://www.kaggle.com/c/data-science-bowl-2018/discussion/47770#272208 but it's indeed not 100% clear if that comment referred to the nuclei annotation on the bottom left of a7f7 that I added - I presume you meant this one?\nI made each fix in a separate commit, so if you are happy with other fixes but not this one, then you can revert this commit in your fork.
", "rawMarkdown": "Sorry, I didn't understand the point about winning. I tried to make only fixes confirmed in the data quality issues thread. Edit: and yes, I'm really using them, although I didn't notice a significant change on the public LB.\n\n> E.g. I am 99% sure, your fix regarding the image named a7f7... (adding one more nucleus at the left bottom) is wrong.\n\nYou may be right - I was basing my fix on this comment by Anne Carpenter: https://www.kaggle.com/c/data-science-bowl-2018/discussion/47770#272208 but it's indeed not 100% clear if that comment referred to the nuclei annotation on the bottom left of a7f7 that I added - I presume you meant this one?\nI made each fix in a separate commit, so if you are happy with other fixes but not this one, then you can revert this commit in your fork.", "votes": 2 }, { "id": 289761, "postDate": "2018-03-02T07:54:12.097Z", "content": "\n\n\nI didn't notice a significant change on the public LB.
\n
Do you see a change in your local CV score?
\n\nI haven't used your fixed dataset but will probably start today. Thanks again for setting this, and thanks to Kaggle and sponsor for allowing it!
", "rawMarkdown": "> I didn't notice a significant change on the public LB.\n\nDo you see a change in your local CV score?\n\nI haven't used your fixed dataset but will probably start today. Thanks again for setting this, and thanks to Kaggle and sponsor for allowing it!" }, { "id": 289796, "postDate": "2018-03-02T09:31:33.733Z", "content": "@Konstantin, Just to clear my previous comment. Train dataset contains evident errors, e.g. a mask in the form of long bar or curious set of masks containing one mask only, where presence of many nuclei is evident. Removing those errors should improve a little bit the quality of the results at any level: local CV, public stage 1 and private stage 2. I can't imagine someone who would prefer a dataset with evident errors over a dataset without them.
\nTrain dataset contains also objects which the annotators did not annotate as nuclei or annotated them by using a mask of strange shape. IMHO, correcting the annotators by adding more masks or separating one mask into many may only accidentally improve the score. Generally, the effect will be the opposite.
Reasumming, I would like to use a branch of train dataset without evident technical errors, but I do not want to lose time for cleaning it from the \"corrections\".
", "rawMarkdown": "@Konstantin, Just to clear my previous comment. Train dataset contains evident errors, e.g. a mask in the form of long bar or curious set of masks containing one mask only, where presence of many nuclei is evident. Removing those errors should improve a little bit the quality of the results at any level: local CV, public stage 1 and private stage 2. I can't imagine someone who would prefer a dataset with evident errors over a dataset without them. \nTrain dataset contains also objects which the annotators did not annotate as nuclei or annotated them by using a mask of strange shape. IMHO, correcting the annotators by adding more masks or separating one mask into many may only accidentally improve the score. Generally, the effect will be the opposite. \n\nReasumming, I would like to use a branch of train dataset without evident technical errors, but I do not want to lose time for cleaning it from the \"corrections\".", "votes": 1 }, { "id": 289879, "postDate": "2018-03-02T14:26:27.160Z", "content": "Thanks a lot, Konstantin. This is a GREAT initiative.
\n\nI would suggest that in the official Kaggle discussion thread pointing out errors, that when people make a fix, they reply to the comment that reported the problem and say it's been addressed in your github repo, pointing to the PR or merge, so it's easy to see what's been fixed already vs not. My team can try (no promises! We are volunteers here!) to keep an eye on things where there are open questions, to clarify what would be correct.
", "rawMarkdown": "Thanks a lot, Konstantin. This is a GREAT initiative.\n\nI would suggest that in the official Kaggle [discussion thread pointing out errors][1], that when people make a fix, they reply to the comment that reported the problem and say it's been addressed in your github repo, pointing to the PR or merge, so it's easy to see what's been fixed already vs not. My team can try (no promises! We are volunteers here!) to keep an eye on things where there are open questions, to clarify what would be correct.\n\n\n [1]: https://www.kaggle.com/c/data-science-bowl-2018/discussion/47770#289116", "votes": 2 }, { "id": 437375, "postDate": "2018-12-11T19:37:34.533Z", "content": "Did you see that the 5th place winner used your fixes, https://www.kaggle.com/c/data-science-bowl-2018/discussion/56326. Great work!
", "rawMarkdown": "Did you see that the 5th place winner used your fixes, https://www.kaggle.com/c/data-science-bowl-2018/discussion/56326. Great work!", "votes": 1 } ] }, { "id": 275397, "postDate": "2018-01-29T02:50:35.660Z", "content": "Think this will be really useful. \nLink to dataset
\n\nBasically, the cells (brightfield) were first unstained and later stained (nuceli and cytoplasmic) using fluorescence microscopy.
\n\nIt'll be both a training set for brightfield and flourenscent images. Especially useful for helping to locate the nuclei for brightfield samples in the dataset.
\n\n
Abstract
\n\nDeep Convolutional Neural Networks (DCNN) have recently emerged as superior for many image segmentation tasks. The DCNN performance is however heavily dependent on the availability of large amounts of problem-specific training samples. Here we show that DCNNs trained on ground truth created automatically using fluorescently labeled cells, perform similar to manual annotations.
\n\nReference
\n\nSadanandan, SK, Ranefall, P, Le Guyader, S, Wählby, C (2017). Automated Training of Deep Convolutional Neural Networks for Cell Segmentation. Sci Rep, 7, 1:7860.
", "rawMarkdown": "Think this will be really useful. \n[Link to dataset][1]\n\nBasically, the cells (brightfield) were first unstained and later stained (nuceli and cytoplasmic) using fluorescence microscopy. \n\nIt'll be both a training set for brightfield and flourenscent images. Especially useful for helping to locate the nuclei for brightfield samples in the dataset. \n\n![Data, training and testing pipelines for cell segmentation. ][2]\n\n**Abstract**\n\nDeep Convolutional Neural Networks (DCNN) have recently emerged as superior for many image segmentation tasks. The DCNN performance is however heavily dependent on the availability of large amounts of problem-specific training samples. Here we show that DCNNs trained on ground truth created automatically using fluorescently labeled cells, perform similar to manual annotations.\n\nReference\n\nSadanandan, SK, Ranefall, P, Le Guyader, S, Wählby, C (2017). Automated Training of Deep Convolutional Neural Networks for Cell Segmentation. Sci Rep, 7, 1:7860.\n\n\n [1]: http://cb.uu.se/~carolina/timelapse_data/\n [2]: https://media.springernature.com/lw900/springer-static/image/art%3A10.1038%2Fs41598-017-07599-6/MediaObjects/41598_2017_7599_Fig1_HTML.jpg", "votes": 11 }, { "id": 269253, "postDate": "2018-01-16T14:27:07.937Z", "content": "Adenocarcinomas with labelled nuclei dataset\nhttps://warwick.ac.uk/fac/sci/dcs/research/tia/data/crchistolabelednucleihe/crchistophenotypes_2016_04_28.zip
\n\nSource: K. Sirinukunwattana, S.E.A. Raza, Y.W Tsang, I.A. Cree, D.R.J. Snead, N.M. Rajpoot, ‘Locality Sensitive Deep Learning for Detection and Classification of Nuclei in Routine Colon Cancer Histology Images,’ IEEE Transactions on Medical Imaging, 2016
\n\nOther Relevant Papers:
\n\nunfortunetly only centers, not masks(((
", "rawMarkdown": "unfortunetly only centers, not masks(((", "votes": 2 } ] }, { "id": 272203, "postDate": "2018-01-22T13:38:57.733Z", "content": "virtual fluorescence microscopy:
\n\nhttp://cbia.fi.muni.cz/projects/simulator.html
\n\nhttp://cbia.fi.muni.cz/projects/cytopacq-a-simulation-toolbox_5.html
\n\nhttp://cbia.fi.muni.cz/simulator/index.php
", "rawMarkdown": "virtual fluorescence microscopy:\n\nhttp://cbia.fi.muni.cz/projects/simulator.html\n\nhttp://cbia.fi.muni.cz/projects/cytopacq-a-simulation-toolbox_5.html\n\nhttp://cbia.fi.muni.cz/simulator/index.php", "votes": 9 }, { "id": 271360, "postDate": "2018-01-20T08:59:43.077Z", "content": "https://cs.adelaide.edu.au/~carneiro/isbi14_challenge/dataset.html
\n\nI find some train_stage1 images are from the cervical cytology images of isbi 14 challenge. I suspect they may be use in test_stage2. If they are not, I can all for isbi14 challenge images for training.
\n\nSo I can I use these images?
", "rawMarkdown": "https://cs.adelaide.edu.au/~carneiro/isbi14_challenge/dataset.html\n\nI find some train_stage1 images are from the cervical cytology images of isbi 14 challenge. I suspect they may be use in test_stage2. If they are not, I can all for isbi14 challenge images for training.\n\nSo I can I use these images?", "votes": 8 }, { "id": 312659, "postDate": "2018-04-12T05:25:24.163Z", "content": "Stage 1 test data masks (generated from the csv file): https://github.com/yuanqing811/DSB2018_stage1_test
", "rawMarkdown": "Stage 1 test data masks (generated from the csv file): https://github.com/yuanqing811/DSB2018_stage1_test", "votes": 6 }, { "id": 284672, "postDate": "2018-02-17T22:10:38.287Z", "content": "Dear Organisers,
\n\nI need an official word from Anne, Kyle or Allen about whether or not user-generated annotations to images already shared here on the thread are to be posted one week before the deadline.
\n\nMy strategy is to use my existing model to pre-label the images and then manually touch up the missing masks or re-fine poor masks. As such my annotations are semi-automatic and are part of my methodology.
\n\nI will likely abandon this strategy if I must share these masks as they are partly generated by my model.
\n\nIf you could kindly respond in a timely manner that would be much appreciated.
\n\nThanks.
", "rawMarkdown": "Dear Organisers,\n\nI need an official word from Anne, Kyle or Allen about whether or not user-generated annotations to images already shared here on the thread are to be posted one week before the deadline.\n\nMy strategy is to use my existing model to pre-label the images and then manually touch up the missing masks or re-fine poor masks. As such my annotations are semi-automatic and are part of my methodology. \n\nI will likely abandon this strategy if I must share these masks as they are partly generated by my model. \n\nIf you could kindly respond in a timely manner that would be much appreciated.\n\nThanks.", "votes": 5, "replies": [ { "id": 285727, "postDate": "2018-02-20T22:40:26.633Z", "content": "You don't need to post them as that is considered part of your methodology, not external data.
", "rawMarkdown": "You don't need to post them as that is considered part of your methodology, not external data. ", "votes": 2 } ] }, { "id": 285466, "postDate": "2018-02-20T10:12:45.613Z", "content": "https://downloads.openmicroscopy.org/images/
\n\nhttps://www.openmicroscopy.org/
\n\nhttps://warwick.ac.uk/fac/sci/dcs/research/tia/data/
", "rawMarkdown": "https://downloads.openmicroscopy.org/images/\n\nhttps://www.openmicroscopy.org/\n\nhttps://warwick.ac.uk/fac/sci/dcs/research/tia/data/", "votes": 6, "replies": [ { "id": 285943, "postDate": "2018-02-21T10:14:19.183Z", "content": "Is it data without ground truth masks?
", "rawMarkdown": "Is it data without ground truth masks?", "votes": 1 }, { "id": 302141, "postDate": "2018-03-23T18:23:00.397Z", "content": "Great links. Thank you!
", "rawMarkdown": "Great links. Thank you!", "votes": 1 } ] }, { "id": 271537, "postDate": "2018-01-20T17:42:01.360Z", "content": "Dataset at:
\n\nhttps://github.com/sanuj/nuclei-net
\n\nhttps://sanuj.github.io/nuclei-net-btp.html
\n\nhttps://nucleisegmentationbenchmark.weebly.com/dataset.html
\n\nhttps://sites.google.com/site/aacruzr/image-datasets, see (Histology and Histopathology, etc)
\n\nhttp://histo.anat.ubc.ca/HISTOLOGY/Basic%20Tissues/view.apml?
\n\nhuman altas
\n\nhttps://www.proteinatlas.org/humanproteome
", "rawMarkdown": "Dataset at: \n\nhttps://github.com/sanuj/nuclei-net\n\nhttps://sanuj.github.io/nuclei-net-btp.html\n\nhttps://nucleisegmentationbenchmark.weebly.com/dataset.html\n\n---\n\nhttps://sites.google.com/site/aacruzr/image-datasets, see (Histology and Histopathology, etc)\n\nhttp://histo.anat.ubc.ca/HISTOLOGY/Basic%20Tissues/view.apml?\n\n----\nhuman altas\n\nhttps://www.proteinatlas.org/humanproteome\n", "votes": 6 }, { "id": 282734, "postDate": "2018-02-14T13:45:42.803Z", "content": "If we change the train set (correct mistakes in the masks), do we have to post that data here as well?
", "rawMarkdown": "If we change the train set (correct mistakes in the masks), do we have to post that data here as well?", "votes": 3, "replies": [ { "id": 285728, "postDate": "2018-02-20T22:40:39.430Z", "content": "You don't need to post them as that is considered part of your methodology, not external data.
", "rawMarkdown": "You don't need to post them as that is considered part of your methodology, not external data.", "votes": 1 } ] }, { "id": 271270, "postDate": "2018-01-20T01:54:19.310Z", "content": "Broad Bioimage Benchmark Collection https://data.broadinstitute.org/bbbc/BBBC006/
", "rawMarkdown": "Broad Bioimage Benchmark Collection https://data.broadinstitute.org/bbbc/BBBC006/", "votes": 4, "replies": [ { "id": 308671, "postDate": "2018-04-03T22:29:22.287Z", "content": "Just so you know, the annotations on that dataset are almost as crappy as on our training data. I'm not entirely sure, though, that it's different from the training dataset.\nEdit: no, it may actually be even worse. One could look at mcf-z-stacks-03212011_m19_s1-27 case to see just how awful it is at instance segmentation.
", "rawMarkdown": "Just so you know, the annotations on that dataset are almost as crappy as on our training data. I'm not entirely sure, though, that it's different from the training dataset.\nEdit: no, it may actually be even worse. One could look at mcf-z-stacks-03212011_m19_s1-27 case to see just how awful it is at instance segmentation." } ] }, { "id": 312803, "postDate": "2018-04-12T11:09:09.873Z", "content": "As I know I'm definitely not gonna win now, I make my revision of the training set and my annotation tool available at https://github.com/ibmua/data-science-bowl-2018-train-set . I think the training set is probably the best there is, as I went through each white-on-black sample and fixed it by hand. Annotation tool is also pretty useful, though, probably limitedly so now after the first stage is over. Maybe, for some future competitions, I donno.
", "rawMarkdown": "As I know I'm definitely not gonna win now, I make my revision of the training set and my annotation tool available at https://github.com/ibmua/data-science-bowl-2018-train-set . I think the training set is probably the best there is, as I went through each white-on-black sample and fixed it by hand. Annotation tool is also pretty useful, though, probably limitedly so now after the first stage is over. Maybe, for some future competitions, I donno.", "votes": 1, "replies": [ { "id": 313218, "postDate": "2018-04-12T23:59:37.600Z", "content": "BBBC6 https://drive.google.com/file/d/16KVLYeKBkk_Y0W4Fra_2AhjmW-p6JjZu/view?usp=sharing\nRed channel is a combo similar to \n383d8ad7069bf55cc2e40f638c10565d9967d25dd38258afb048ab8d012d72ce , 486fe606382b28907ffd6f4e0b45d6a1b1ea3ef798eb59d5a4dd917fa8ec5f90 , 572fe655c8d8091ab4abbcdc47aa9c6989ed4becee377a43cea689903a1364be , 3972f3d313a6f5152801f958f26409fc6c6003646087abe1713139d2ab2adc88
\n\n![\"red](\"https://storage.googleapis.com/kaggle-forum-message-attachments/307465/9131/broad-mcf-z-stacks-03212011_a17_s1.png\")
(https://cs.adelaide.edu.au/~carneiro/isbi14_challenge/dataset.html) PSB Crowdsourced Nuclei Annotation (https://becklab.hms.harvard.edu/software/psb-crowdsourced-nuclei-annotation-data-1) https://www.kaggle.com/voglinio/external-h-e-data-with-mask-annotations (made by Watt, converted from https://nucleisegmentationbenchmark.weebly.com/)\nhttp://www.andrewjanowczyk.com/use-case-1-nuclei-segmentation/\nhttps://bioimage.ucsb.edu/research/bio-segmentation
", "rawMarkdown": "(https://cs.adelaide.edu.au/~carneiro/isbi14_challenge/dataset.html) PSB Crowdsourced Nuclei Annotation (https://becklab.hms.harvard.edu/software/psb-crowdsourced-nuclei-annotation-data-1) https://www.kaggle.com/voglinio/external-h-e-data-with-mask-annotations (made by Watt, converted from https://nucleisegmentationbenchmark.weebly.com/)\nhttp://www.andrewjanowczyk.com/use-case-1-nuclei-segmentation/\nhttps://bioimage.ucsb.edu/research/bio-segmentation", "votes": 1, "replies": [ { "id": 405751, "postDate": "2018-10-18T03:43:55.083Z", "rawMarkdown": "", "isDeleted": true } ] }, { "id": 308586, "postDate": "2018-04-03T17:56:35.043Z", "content": "We used these images (sources in the 'sources.txt' file)\nhttps://drive.google.com/file/d/1A_x-40lfh9FaBEsRv2zTjjvNOlcItnTX/view?usp=sharing
", "rawMarkdown": "We used these images (sources in the 'sources.txt' file)\nhttps://drive.google.com/file/d/1A_x-40lfh9FaBEsRv2zTjjvNOlcItnTX/view?usp=sharing", "votes": 1 }, { "id": 308476, "postDate": "2018-04-03T14:54:34.303Z", "content": "https://zenodo.org/record/1175282\nhttp://cbia.fi.muni.cz/projects/cytopacq-a-simulation-toolbox_6.html
", "rawMarkdown": "https://zenodo.org/record/1175282\nhttp://cbia.fi.muni.cz/projects/cytopacq-a-simulation-toolbox_6.html", "votes": 1, "replies": [ { "id": 308536, "postDate": "2018-04-03T16:56:33.307Z", "content": "Hi,\nThanks for the links, however the second one is broken.
", "rawMarkdown": "Hi,\nThanks for the links, however the second one is broken." } ] }, { "id": 307796, "postDate": "2018-04-02T13:36:06.493Z", "content": "Our team will probably use some pretrained models from the following list:
\n\nhttps://github.com/Cadene/pretrained-models.pytorch\nhttps://keras.io/applications/\nhttps://github.com/fizyr/keras-models/releases/tag/v0.0.1\nhttps://mxnet.incubator.apache.org/api/python/gluon/model_zoo.html
\n\nAdditionally we use some external datasets mentioned in this thread.
", "rawMarkdown": "Our team will probably use some pretrained models from the following list:\n\nhttps://github.com/Cadene/pretrained-models.pytorch\nhttps://keras.io/applications/\nhttps://github.com/fizyr/keras-models/releases/tag/v0.0.1\nhttps://mxnet.incubator.apache.org/api/python/gluon/model_zoo.html\n\nAdditionally we use some external datasets mentioned in this thread.", "votes": 1 }, { "id": 307460, "postDate": "2018-04-01T16:50:21.557Z", "content": "PSB Crowdsourced Nuclei Annotation
\n\nhttps://becklab.hms.harvard.edu/software/psb-crowdsourced-nuclei-annotation-data-1
", "rawMarkdown": "PSB Crowdsourced Nuclei Annotation\n\nhttps://becklab.hms.harvard.edu/software/psb-crowdsourced-nuclei-annotation-data-1\n", "votes": 1 }, { "id": 269385, "postDate": "2018-01-16T18:17:26.230Z", "content": "Are pre-trained models allowed in this competition?
", "rawMarkdown": "Are pre-trained models allowed in this competition?", "votes": 1, "replies": [ { "id": 269394, "postDate": "2018-01-16T18:39:54.260Z", "content": "It really depends on what you mean by “pre-trained.” 😎 Here’s the relevant section of the rules:
\n\n\n\n\nExternal Data Use: The following provision supersedes General Rules\n Section 7.C. below: \"You may use data, other than the Competition\n Data, as allowed on the Competition Website to develop and test your\n models and Submissions; provided, you have the right and authority to\n use such external data for the purposes of the Competition. If you are\n using a source of external data, you must post the source to the\n official external data forum thread no later than one week prior to\n the deadline of the first stage.
\n
If you mean use pre-trained ImageNet (or other publicly available) weights, I don’t see a problem. I would expect most competitors will use pre-trained weights. However, I should note that biological images are less rich than natural images so training from scratch is entirely reasonable (and we often do that to simplify our training architectures). You’re also dealing with one class rather than thousands.
", "rawMarkdown": "It really depends on what you mean by “pre-trained.” 😎 Here’s the relevant section of the rules:\n\n> External Data Use: The following provision supersedes General Rules\n> Section 7.C. below: \"You may use data, other than the Competition\n> Data, as allowed on the Competition Website to develop and test your\n> models and Submissions; provided, you have the right and authority to\n> use such external data for the purposes of the Competition. If you are\n> using a source of external data, you must post the source to the\n> official external data forum thread no later than one week prior to\n> the deadline of the first stage.\n\nIf you mean use pre-trained ImageNet (or other publicly available) weights, I don’t see a problem. I would expect most competitors will use pre-trained weights. However, I should note that biological images are less rich than natural images so training from scratch is entirely reasonable (and we often do that to simplify our training architectures). You’re also dealing with one class rather than thousands.", "votes": 4 }, { "id": 269663, "postDate": "2018-01-17T04:30:07.130Z", "content": "\n\n\nAllen Goodman wrote
\n \nIf you mean use pre-trained ImageNet (or other publicly available) weights, I don’t see a problem.
\n
Strictly speaking, ImageNet just contains the web URLs of the images. The images themselves are copyrighted and, in many cases, already gone from the web. It's possible the the copyright owners' intent was to completely prevent their sharing.
\n\nIf using ImagetNet images is not allowed, our right to use models pretrained on them also seems unclear.
", "rawMarkdown": "\n> **Allen Goodman wrote**\n> \n> If you mean use pre-trained ImageNet (or other publicly available) weights, I don’t see a problem. \n\nStrictly speaking, ImageNet just contains the web URLs of the images. The images themselves are copyrighted and, in many cases, already gone from the web. It's possible the the copyright owners' intent was to completely prevent their sharing.\n\nIf using ImagetNet images is not allowed, our right to use models pretrained on them also seems unclear.\n\n\n", "votes": 1 }, { "id": 269941, "postDate": "2018-01-17T14:52:39.780Z", "content": "I was addressing weights trained with ImageNet images.
", "rawMarkdown": "I was addressing *weights* trained with ImageNet images.\n" }, { "id": 269970, "postDate": "2018-01-17T15:45:27.957Z", "content": "\n\n\nAllen Goodman wrote
\n \n\n\nI was addressing weights trained with ImageNet images.
\n
Allen,
\n\nI understand that (see the last sentence). If competitors use those weights for initialization and then train the models on Kaggle's data before submitting them, they will be submitting models trained, in part, on ImageNet images (by a third party).
", "rawMarkdown": "\n> **Allen Goodman wrote**\n> \n> > I was addressing *weights* trained with ImageNet images.\n\nAllen,\n\nI understand that (see the last sentence). If competitors use those weights for initialization and then train the models on Kaggle's data before submitting them, they will be submitting models trained, in part, on ImageNet images (by a third party).\n\n" }, { "id": 288542, "postDate": "2018-02-27T18:13:12.523Z", "content": "Do all competitors using ImageNet weights need to report in this thread they might be using it for transfer learning or does it suffice to have someone having it posted?
", "rawMarkdown": "Do all competitors using ImageNet weights need to report in this thread they might be using it for transfer learning or does it suffice to have someone having it posted? " } ] }, { "id": 269291, "postDate": "2018-01-16T15:46:17.463Z", "content": "Is external data allowed in this competition?
", "rawMarkdown": "Is external data allowed in this competition?", "votes": 1, "replies": [ { "id": 269352, "postDate": "2018-01-16T17:16:13.717Z", "content": "Yes!
\n\nIf you do use external data, it’d be awesome to share your sources.
\n\nI’d also be mindful of licensing to ensure the external data can be used for future research.
", "rawMarkdown": "Yes!\n\nIf you do use external data, it’d be awesome to share your sources. \n\nI’d also be mindful of licensing to ensure the external data can be used for future research.", "votes": 5 }, { "id": 271061, "postDate": "2018-01-19T16:04:24.270Z", "content": "What about manual labeling on external data? E.g. some of the data may come without any label.
", "rawMarkdown": "What about manual labeling on external data? E.g. some of the data may come without any label. ", "votes": 4 }, { "id": 282461, "postDate": "2018-02-14T02:29:15.933Z", "content": "I also have the same question. Most of the external data has the coordinates of the nuclei location but doesn't have an annotated mask. What if I create masks based on those data do I have to share the masks in here?
", "rawMarkdown": "I also have the same question. Most of the external data has the coordinates of the nuclei location but doesn't have an annotated mask. What if I create masks based on those data do I have to share the masks in here?" }, { "id": 282920, "postDate": "2018-02-14T19:21:31.053Z", "content": "You should share both the images and the annotations you’re using (i.e. both the masks and the object centroids). You should also share the process you used to create the masks. The rules of the competition were written to encourage research around model construction and training practices.
\n\nThe rules also try to make the competition equitable between biologists and non-biologists. This is important since some biologists (e.g. biologists in our lab) have access to terabytes of data (😳) while others only have access to what’s available online.
", "rawMarkdown": "You should share both the images and the annotations you’re using (i.e. both the masks and the object centroids). You should also share the process you used to create the masks. The rules of the competition were written to encourage research around model construction and training practices. \n\nThe rules also try to make the competition equitable between biologists and non-biologists. This is important since some biologists (e.g. biologists in our lab) have access to terabytes of data (😳) while others only have access to what’s available online.", "votes": 3 }, { "id": 283066, "postDate": "2018-02-14T22:18:54.523Z", "content": "Hi Allen, I want to clarify if we \"should\" or \"must\" share the annotations?
\n\nI understand that there is a research motivation behind this project however this is still a competition with winners. Annotation is extremely labor intensive and costly. I think if we must share our annotation no one will do any extra annotation as it would give you no competitive advantage.
\n\nI think it is very reasonable to share the annotation to the freely available images \u001dAFTER the competition, as long as the images themselves were shared here on this thread. Given the small set of data provided for training if the aim is to build something that is ultimately useful annotation is ultimately needed and there need to be an incentive to do that.
", "rawMarkdown": "Hi Allen, I want to clarify if we \"should\" or \"must\" share the annotations?\n\nI understand that there is a research motivation behind this project however this is still a competition with winners. Annotation is extremely labor intensive and costly. I think if we must share our annotation no one will do any extra annotation as it would give you no competitive advantage. \n\nI think it is very reasonable to share the annotation to the freely available images \u001dAFTER the competition, as long as the images themselves were shared here on this thread. Given the small set of data provided for training if the aim is to build something that is ultimately useful annotation is ultimately needed and there need to be an incentive to do that.", "votes": 5 }, { "id": 283184, "postDate": "2018-02-15T02:19:36.817Z", "content": "I think both views has some valid points.
\n\nAlso there are some data sets that are partially annotated and if we use some programmatic method to extract only the areas annotated (for example this data set http://www.andrewjanowczyk.com/use-case-1-nuclei-segmentation/ has some areas annotated in the image) should we share that details in this thread?
", "rawMarkdown": "I think both views has some valid points. \n\nAlso there are some data sets that are partially annotated and if we use some programmatic method to extract only the areas annotated (for example this data set http://www.andrewjanowczyk.com/use-case-1-nuclei-segmentation/ has some areas annotated in the image) should we share that details in this thread?" }, { "id": 284665, "postDate": "2018-02-17T21:44:47.043Z", "content": "My rough guess on data you label yourself (speculating here but this is what makes sense to me): if you do label the data manually, then your labels are an external source of data, and you need to share them. If you label automatically, then your labeling is part of your algorithm and you don't need to share it.
\n\nNote that the rules say \"If you are using a source of external data, you must post the source to the official external data forum thread no later than one week prior to the deadline of the first stage.\" (emphasis mine), so if you really want to keep your competitive advantage you could withhold your data up until exactly one week before the end of the competition.
", "rawMarkdown": "My rough guess on data you label yourself (speculating here but this is what makes sense to me): if you do label the data manually, then your labels are an external source of data, and you need to share them. If you label automatically, then your labeling is part of your algorithm and you don't need to share it.\n\nNote that the rules say \"If you are using a source of external data, you must post the source to the official external data forum thread **no later than one week prior to the deadline of the first stage**.\" (emphasis mine), so if you really want to keep your competitive advantage you could withhold your data up until exactly one week before the end of the competition.", "votes": 1 }, { "id": 285489, "postDate": "2018-02-20T12:43:10.973Z", "content": "Automatic, semi-automatic or manual annotation requires time-effort. Sharing afterward the competition makes sense. Don’t agree with external data source portion though.
", "rawMarkdown": "Automatic, semi-automatic or manual annotation requires time-effort. Sharing afterward the competition makes sense. Don’t agree with external data source portion though. " }, { "id": 285740, "postDate": "2018-02-20T23:17:42.190Z", "content": "\n\n\nAllen Goodman wrote
\n \nThis is important since some biologists (e.g. biologists in our lab) have access to terabytes of data (😳)
\n
That's how you've reached 0.64 LB? :)\nSeriously, I'm asking you this to know if we can beat your result or it should be considered a proxy for the bayesian error.
", "rawMarkdown": ">**Allen Goodman wrote**\n>\n>This is important since some biologists (e.g. biologists in our lab) have access to terabytes of data (😳)\n\nThat's how you've reached 0.64 LB? :)\nSeriously, I'm asking you this to know if we can beat your result or it should be considered a proxy for the bayesian error." }, { "id": 285996, "postDate": "2018-02-21T12:57:49.297Z", "content": "\n\n\nAutomatic, semi-automatic or manual annotation requires time-effort. Sharing afterward the competition makes sense. Don’t agree with external data source portion though.
\n
Then maybe you can organize your own competition someday and make the rules how you like them ;)
\n\nBut seriously, here is the rationale: though manual annotation will definitely result in better models, we actually wanted to discourage competitors from winning that way, because it would disadvantage those who don't have the time or money to do annotations. Hopefully, people can focus on their methodology for the competition itself, and in the future more annotations can be made and then the model will just get better and better.
", "rawMarkdown": "> Automatic, semi-automatic or manual annotation requires time-effort. Sharing afterward the competition makes sense. Don’t agree with external data source portion though.\n\nThen maybe you can organize your own competition someday and make the rules how you like them ;) \n\nBut seriously, here is the rationale: though manual annotation will definitely result in better models, we actually wanted to discourage competitors from winning that way, because it would disadvantage those who don't have the time or money to do annotations. Hopefully, people can focus on their methodology for the competition itself, and in the future more annotations can be made and then the model will just get better and better.", "votes": 1 }, { "id": 292488, "postDate": "2018-03-08T02:44:00.193Z", "content": "Well, I asked about the need for sharing masks when using external data, and was answered that no, people don't have to. It means that people with biology knowledge have an unfair advantage as they can annotate more data for training.
", "rawMarkdown": "Well, I asked about the need for sharing masks when using external data, and was answered that no, people don't have to. It means that people with biology knowledge have an unfair advantage as they can annotate more data for training." }, { "id": 293412, "postDate": "2018-03-09T19:37:22.330Z", "content": "Biologists would certainly have an advantage... then again, being forced to circle cells for hours is not my idea of fun! On the other hand, people who have deep learning knowledge also have an advantage in this challenge so I suppose it is all \"fair\".
", "rawMarkdown": "Biologists would certainly have an advantage... then again, being forced to circle cells for hours is not my idea of fun! On the other hand, people who have deep learning knowledge also have an advantage in this challenge so I suppose it is all \"fair\".", "votes": 1 }, { "id": 293423, "postDate": "2018-03-09T20:04:49.703Z", "content": "Perhaps it is an encouragement for the formation of strong teams with a wide scope of expertise.
", "rawMarkdown": "Perhaps it is an encouragement for the formation of strong teams with a wide scope of expertise." } ] }, { "id": 309363, "postDate": "2018-04-05T06:17:45.900Z", "content": "Pretrained model using the Coco dataset (https://github.com/matterport/Mask_RCNN/releases)\nFixed training dataset (https://github.com/lopuhin/kaggle-dsbowl-2018-dataset-fixes)\nTNBC_NucleiSegmentation (https://cs.adelaide.edu.au/~carneiro/isbi14_challenge/dataset.html)\nPSB Crowdsourced Nuclei Annotation (https://becklab.hms.harvard.edu/software/psb-crowdsourced-nuclei-annotation-data-1)\nhttps://www.kaggle.com/voglinio/external-h-e-data-with-mask-annotations (made by Watt, converted from https://nucleisegmentationbenchmark.weebly.com/)
", "rawMarkdown": "Pretrained model using the Coco dataset (https://github.com/matterport/Mask_RCNN/releases)\nFixed training dataset (https://github.com/lopuhin/kaggle-dsbowl-2018-dataset-fixes)\nTNBC_NucleiSegmentation (https://cs.adelaide.edu.au/~carneiro/isbi14_challenge/dataset.html)\nPSB Crowdsourced Nuclei Annotation (https://becklab.hms.harvard.edu/software/psb-crowdsourced-nuclei-annotation-data-1)\nhttps://www.kaggle.com/voglinio/external-h-e-data-with-mask-annotations (made by Watt, converted from https://nucleisegmentationbenchmark.weebly.com/)\n\n", "votes": 2 }, { "id": 298355, "postDate": "2018-03-19T12:42:28.387Z", "content": "https://github.com/lopuhin/kaggle-dsbowl-2018-dataset-fixes
", "rawMarkdown": " https://github.com/lopuhin/kaggle-dsbowl-2018-dataset-fixes", "replies": [ { "id": 301849, "postDate": "2018-03-23T11:17:28.480Z", "content": "https://www.kaggle.com/voglinio/external-h-e-data-with-mask-annotations
", "rawMarkdown": " https://www.kaggle.com/voglinio/external-h-e-data-with-mask-annotations", "votes": 3 }, { "id": 308926, "postDate": "2018-04-04T11:33:22.427Z", "content": "https://becklab.hms.harvard.edu/software/psb-crowdsourced-nuclei-annotation-data-1
", "rawMarkdown": "https://becklab.hms.harvard.edu/software/psb-crowdsourced-nuclei-annotation-data-1" }, { "id": 308929, "postDate": "2018-04-04T11:35:02.633Z", "content": ".https://nucleisegmentationbenchmark.weebly.com/
", "rawMarkdown": ".https://nucleisegmentationbenchmark.weebly.com/" } ] }, { "id": 276790, "postDate": "2018-02-01T14:07:30.350Z", "content": "If we change the images (cropping, rotating, flipping etc) from the training set to train our model, do we have to post that data here as well?
", "rawMarkdown": "If we change the images (cropping, rotating, flipping etc) from the training set to train our model, do we have to post that data here as well?", "votes": 2, "replies": [ { "id": 276814, "postDate": "2018-02-01T15:12:01.293Z", "content": "No. We consider that to be part of your methodology, not external data.
", "rawMarkdown": "No. We consider that to be part of your methodology, not external data.", "votes": 3 } ] }, { "id": 273344, "postDate": "2018-01-24T13:33:45.990Z", "content": "https://ome.grc.nia.nih.gov/iicbu2008/
", "rawMarkdown": "https://ome.grc.nia.nih.gov/iicbu2008/", "votes": 2 }, { "id": 269625, "postDate": "2018-01-17T03:17:44.487Z", "content": "In the AMIDA13 MICCAI grand competition, they achieved the 3rd highest F1-score.
\n\nCell Detection with Deep Convolutional Neural Network and Compressed Sensing.
\n\n", "rawMarkdown": " In the AMIDA13 MICCAI grand competition, they achieved the **3rd highest** F1-score. \n\n**Cell Detection with Deep Convolutional Neural Network and Compressed Sensing**.\n\n[link to the arXiv paper][1]\n\n\n [1]: https://arxiv.org/pdf/1708.03307.pdf", "votes": 2, "replies": [ { "id": 271150, "postDate": "2018-01-19T19:29:11.130Z", "content": "Is the dataset downloadable?
", "rawMarkdown": "Is the dataset downloadable? " }, { "id": 276226, "postDate": "2018-01-31T02:09:39.243Z", "content": "The competition has ended, it seems you'll have to register on their site to get the data \nhttp://amida13.isi.uu.nl/?q=node/90
\n\nhowever, it seems its continuing in a separate data challenge: http://tupac.tue-image.nl. Tried to sign up but still waiting for the admins to approve.
\n\nmeanwhile i saw (http://tupac.tue-image.nl/node/62 ) some of the winning teams used some external datasets:
\n\nHope this helps everyone!
", "rawMarkdown": "The competition has ended, it seems you'll have to register on their site to get the data \nhttp://amida13.isi.uu.nl/?q=node/90\n\nhowever, it seems its continuing in a separate data challenge: http://tupac.tue-image.nl. Tried to sign up but still waiting for the admins to approve. \n\nmeanwhile i saw (http://tupac.tue-image.nl/node/62 ) some of the winning teams used some external datasets:\n\n1. http://ludo17.free.fr/mitos_2012/download.html\n2. https://mitos-atypia-14.grand-challenge.org/donwload/\n\nHope this helps everyone!", "votes": 4 } ] }, { "id": 309275, "postDate": "2018-04-04T23:47:17.123Z", "content": "http://pretrained.ml\nhttps://gist.github.com/flyyufelix/7e2eafb149f72f4d38dd661882c554a6\nhttps://github.com/Cadene/pretrained-models.pytorch\nhttps://github.com/Zehaos/MobileNet\nhttps://github.com/aaron-xichen/pytorch-playground\nhttps://github.com/endymecy/awesome-deeplearning-resources\nhttps://github.com/fchollet/deep-learning-models\nhttps://github.com/fchollet/keras/blob/master/keras/applications/xception.py\nhttps://github.com/iro-cp/FCRN-DepthPrediction\nhttps://github.com/kentsommer/keras-inceptionV4\nhttps://github.com/mrharicot/monodepth\nhttps://github.com/myutwo150/keras-inception-resnet-v2\nhttps://github.com/nutszebra/resnext\nhttps://github.com/openimages/dataset\nhttps://github.com/raulmur/ORB_SLAM2\nhttps://github.com/shicai/SENet-Caffe\nhttps://github.com/taehoonlee/tensornets\nhttps://github.com/tensorflow/models/blob/master/slim/README.md\nhttps://github.com/titu1994/MobileNetworks\nhttps://keras.io/applications/
\n\nhttps://github.com/naoto0804/cross-domain-detection\nhttps://github.com/k3nt0w/FCN_via_keras\nhttps://github.com/imatge-upc/rsis
\n\nhttps://github.com/pjreddie/darknet\nhttps://github.com/explosion/lightnet\nhttps://gitlab.com/EAVISE/lightnet\nhttps://github.com/yjmade/ios_camera_object_detection\nhttps://github.com/mrzl/ofxDarknet\nhttps://github.com/moovel/node-yolo
\n\nhttps://www.tensorflow.org/hub/\nhttps://github.com/youansheng/pytorch-cv\nhttps://github.com/chainer/chainercv\nhttps://tryolabs.com/blog/2018/01/18/faster-r-cnn-down-the-rabbit-hole-of-modern-object-detection/\nhttps://github.com/facebookresearch/Detectron\nhttps://github.com/DeNA/Chainer_Mask_R-CNN\nhttps://adamspannbauer.github.io/2017/12/28/super-smash-cv/\nhttps://github.com/sfzhang15/RefineDet\nhttps://github.com/kuangliu/pytorch-retinanet\nhttps://github.com/fizyr/keras-retinanet\nhttps://github.com/szq0214/DSOD\nhttps://github.com/mfigurnov/sact\nhttps://github.com/allanzelener/YAD2K\nhttps://github.com/daijifeng001/R-FCN\nhttps://github.com/weichengkuo/DeepBox\nhttps://github.com/endernewton/tf-faster-rcnn\nhttp://vision.cs.utexas.edu/projects/pixelobjectness/
\n\nhttps://github.com/ternaus/robot-surgery-segmentation\nhttps://github.com/knorth55/chainer-fcis\nhttps://github.com/matterport/Mask_RCNN\nhttps://github.com/fyu/drn\nhttps://github.com/mitmul/chainer-pspnet\nhttps://github.com/TobyPDE/FRRN\nhttps://github.com/msracver/FCIS\nhttps://blog.deepsense.ai/deep-learning-for-satellite-imagery-via-image-segmentation/\nhttps://github.com/isht7/pytorch-deeplab-resnet\nhttps://github.com/kmaninis/OSVOS-caffe\nhttps://github.com/hszhao/PSPNet\nhttps://github.com/HyeonwooNoh/DeconvNet\nhttps://github.com/fregu856/segmentation\nhttps://github.com/e-lab/ENet-training\nhttp://www.mathgram.xyz/entry/keras/fcn\nhttp://stanford.edu/~rqi/pointnet/\nhttps://github.com/Engineering-Course/LIP_SSL
\n\nand all models listed on this page
", "rawMarkdown": "http://pretrained.ml\nhttps://gist.github.com/flyyufelix/7e2eafb149f72f4d38dd661882c554a6\nhttps://github.com/Cadene/pretrained-models.pytorch\nhttps://github.com/Zehaos/MobileNet\nhttps://github.com/aaron-xichen/pytorch-playground\nhttps://github.com/endymecy/awesome-deeplearning-resources\nhttps://github.com/fchollet/deep-learning-models\nhttps://github.com/fchollet/keras/blob/master/keras/applications/xception.py\nhttps://github.com/iro-cp/FCRN-DepthPrediction\nhttps://github.com/kentsommer/keras-inceptionV4\nhttps://github.com/mrharicot/monodepth\nhttps://github.com/myutwo150/keras-inception-resnet-v2\nhttps://github.com/nutszebra/resnext\nhttps://github.com/openimages/dataset\nhttps://github.com/raulmur/ORB_SLAM2\nhttps://github.com/shicai/SENet-Caffe\nhttps://github.com/taehoonlee/tensornets\nhttps://github.com/tensorflow/models/blob/master/slim/README.md\nhttps://github.com/titu1994/MobileNetworks\nhttps://keras.io/applications/\nhttps://softwaremill.com/counting-objects-with-faster-rcnn/\n---\nhttps://github.com/naoto0804/cross-domain-detection\nhttps://github.com/k3nt0w/FCN_via_keras\nhttps://github.com/imatge-upc/rsis\nhttps://skrish13.github.io/articles/2018-03/fair-cv-saga\n---\nhttps://github.com/pjreddie/darknet\nhttps://github.com/explosion/lightnet\nhttps://gitlab.com/EAVISE/lightnet\nhttps://github.com/yjmade/ios_camera_object_detection\nhttps://github.com/mrzl/ofxDarknet\nhttps://github.com/moovel/node-yolo\nhttps://github.com/leetenki/YOLOv2\n---\nhttps://www.tensorflow.org/hub/\nhttps://github.com/youansheng/pytorch-cv\nhttps://github.com/chainer/chainercv\nhttps://tryolabs.com/blog/2018/01/18/faster-r-cnn-down-the-rabbit-hole-of-modern-object-detection/\nhttps://github.com/facebookresearch/Detectron\nhttps://github.com/DeNA/Chainer_Mask_R-CNN\nhttps://adamspannbauer.github.io/2017/12/28/super-smash-cv/\nhttps://github.com/sfzhang15/RefineDet\nhttps://github.com/kuangliu/pytorch-retinanet\nhttps://github.com/fizyr/keras-retinanet\nhttps://github.com/szq0214/DSOD\nhttps://github.com/mfigurnov/sact\nhttps://github.com/allanzelener/YAD2K\nhttps://github.com/daijifeng001/R-FCN\nhttps://github.com/weichengkuo/DeepBox\nhttps://github.com/endernewton/tf-faster-rcnn\nhttp://vision.cs.utexas.edu/projects/pixelobjectness/\nhttps://imatge-upc.github.io/detection-2016-nipsws/\n---\nhttps://github.com/ternaus/robot-surgery-segmentation\nhttps://github.com/knorth55/chainer-fcis\nhttps://github.com/matterport/Mask_RCNN\nhttps://github.com/fyu/drn\nhttps://github.com/mitmul/chainer-pspnet\nhttps://github.com/TobyPDE/FRRN\nhttps://github.com/msracver/FCIS\nhttps://blog.deepsense.ai/deep-learning-for-satellite-imagery-via-image-segmentation/\nhttps://github.com/isht7/pytorch-deeplab-resnet\nhttps://github.com/kmaninis/OSVOS-caffe\nhttps://github.com/hszhao/PSPNet\nhttps://github.com/HyeonwooNoh/DeconvNet\nhttps://github.com/fregu856/segmentation\nhttps://github.com/e-lab/ENet-training\nhttp://www.mathgram.xyz/entry/keras/fcn\nhttp://stanford.edu/~rqi/pointnet/\nhttps://github.com/Engineering-Course/LIP_SSL\n\nand all models listed on this page\n", "votes": -2, "replies": [ { "id": 309312, "postDate": "2018-04-05T02:47:22.240Z", "content": "These are not related to 'external data'.
", "rawMarkdown": "These are not related to 'external data'. " }, { "id": 309323, "postDate": "2018-04-05T03:29:44.633Z", "rawMarkdown": "", "isDeleted": true } ] }, { "id": 278534, "postDate": "2018-02-06T12:09:48.537Z", "rawMarkdown": "", "votes": -2 }, { "id": 270422, "postDate": "2018-01-18T07:41:08.907Z", "content": "23423
", "rawMarkdown": "23423", "votes": -7 }, { "id": 411468, "postDate": "2018-10-28T08:30:13.760Z", "content": "amazing！
", "rawMarkdown": "amazing！" }, { "id": 313436, "postDate": "2018-04-13T08:55:55.750Z", "content": "We use the following package for data augmentation: \nhttps://github.com/NVIDIA/FastPhotoStyle/blob/master/USAGE.md\nJust realized that this package contains pre-trained model weights (https://github.com/suquark/PhotoWCTModels/blob/d8a4e35f2a994eb7c6f261b5e8172ae6397040f2/photo_wct.pth) that needs to be posted here.
", "rawMarkdown": "We use the following package for data augmentation: \nhttps://github.com/NVIDIA/FastPhotoStyle/blob/master/USAGE.md\nJust realized that this package contains pre-trained model weights (https://github.com/suquark/PhotoWCTModels/blob/d8a4e35f2a994eb7c6f261b5e8172ae6397040f2/photo_wct.pth) that needs to be posted here. " }, { "id": 313203, "postDate": "2018-04-12T23:05:26.027Z", "content": "I make my version of BBBC6 dataset turned into DSB style is available at https://drive.google.com/file/d/16KVLYeKBkk_Y0W4Fra_2AhjmW-p6JjZu/view?usp=sharing . Please, view images before use to understand how the data is combined in channels.
", "rawMarkdown": "I make my version of BBBC6 dataset turned into DSB style is available at https://drive.google.com/file/d/16KVLYeKBkk_Y0W4Fra_2AhjmW-p6JjZu/view?usp=sharing . Please, view images before use to understand how the data is combined in channels." }, { "id": 312773, "postDate": "2018-04-12T10:02:00.023Z", "content": "Now, knowing what stage 2 data is like, I'm guessing, pretty much all datasets included here except BBBC ones are only gonna hurt performance on stage 2 data. Trying to fit them has only taken away effort from trying to fit the actual data for me. Stage 2 data doesn't even contain any gray black-on-white images. The gray-only solution I made in the first few days is actually 10x better for this data than the one I couldn't finish on time that was supposed to work on any colors and lighting conditions.
", "rawMarkdown": "Now, knowing what stage 2 data is like, I'm guessing, pretty much all datasets included here except BBBC ones are only gonna hurt performance on stage 2 data. Trying to fit them has only taken away effort from trying to fit the actual data for me. Stage 2 data doesn't even contain any gray black-on-white images. The gray-only solution I made in the first few days is actually 10x better for this data than the one I couldn't finish on time that was supposed to work on any colors and lighting conditions." }, { "id": 312502, "postDate": "2018-04-11T21:39:58.683Z", "content": "https://nucleisegmentationbenchmark.weebly.com/
", "rawMarkdown": "https://nucleisegmentationbenchmark.weebly.com/" }, { "id": 312241, "postDate": "2018-04-11T12:53:59.447Z", "content": "Based on Mask RCNN with pre-trained model on COCO dataset from https://github.com/matterport/Mask_RCNN/releases
", "rawMarkdown": "Based on Mask RCNN with pre-trained model on COCO dataset from https://github.com/matterport/Mask_RCNN/releases" }, { "id": 312238, "postDate": "2018-04-11T12:47:26.730Z", "content": "I konw it's a bit late, we will use the external Data as follows:\n(1) TNBC_NucleiSegmentation: https://zenodo.org/record/1175282#.Ws2n_vkdhfA\n(2) https://github.com/lopuhin/kaggle-dsbowl-2018-dataset-fixes\n(3) pretrained on https://github.com/matterport/Mask_RCNN/releases
", "rawMarkdown": "I konw it's a bit late, we will use the external Data as follows:\n(1) TNBC_NucleiSegmentation: https://zenodo.org/record/1175282#.Ws2n_vkdhfA\n(2) https://github.com/lopuhin/kaggle-dsbowl-2018-dataset-fixes\n(3) pretrained on https://github.com/matterport/Mask_RCNN/releases" }, { "id": 312100, "postDate": "2018-04-11T07:28:10.783Z", "content": "Using pre-trained model from the MS Coco dataset, as part of the Matterport Mask RCNN implementation: reference https://github.com/matterport/Mask_RCNN/releases
", "rawMarkdown": "Using pre-trained model from the MS Coco dataset, as part of the Matterport Mask RCNN implementation: reference https://github.com/matterport/Mask_RCNN/releases" }, { "id": 312066, "postDate": "2018-04-11T05:59:22.360Z", "content": "https://github.com/matterport/Mask_RCNN i use pretrained model on coco and imagenet
", "rawMarkdown": "https://github.com/matterport/Mask_RCNN i use pretrained model on coco and imagenet" }, { "id": 311858, "postDate": "2018-04-10T19:46:36.530Z", "content": "https://github.com/lopuhin/kaggle-dsbowl-2018-dataset-fixes\nhttps://www.kaggle.com/voglinio/external-h-e-data-with-mask-annotations
", "rawMarkdown": "https://github.com/lopuhin/kaggle-dsbowl-2018-dataset-fixes\nhttps://www.kaggle.com/voglinio/external-h-e-data-with-mask-annotations" }, { "id": 311412, "postDate": "2018-04-10T02:19:16.060Z", "content": "We used these images: https://www.kaggle.com/voglinio/external-h-e-data-with-mask-annotations
", "rawMarkdown": "We used these images: https://www.kaggle.com/voglinio/external-h-e-data-with-mask-annotations" }, { "id": 311194, "postDate": "2018-04-09T15:27:20.443Z", "content": "Hasn't the external data deadline passed already?
", "rawMarkdown": "Hasn't the external data deadline passed already?", "replies": [ { "id": 311198, "postDate": "2018-04-09T15:50:20.647Z", "content": "You are probably right. We misunderstood that it should be before 1st and not 2nd deadline (team merging point). Still, we will make two submissions, with augmented or only kaggle dataset. I guess we are not eligible for a prize, but given, our current position it is probably not a big deal ;)
", "rawMarkdown": "You are probably right. We misunderstood that it should be before 1st and not 2nd deadline (team merging point). Still, we will make two submissions, with augmented or only kaggle dataset. I guess we are not eligible for a prize, but given, our current position it is probably not a big deal ;)" } ] }, { "id": 311193, "postDate": "2018-04-09T15:25:55.620Z", "content": "Using an additional data from our own microscope. It is anonimized and comes from few projects. Unfortunately only fluorescent images.
\n\nhttps://www.dropbox.com/sh/i753b966e6zp7t4/AAB6KtMCgcmKzyfdxEzXVIyHa?dl=0
", "rawMarkdown": "Using an additional data from our own microscope. It is anonimized and comes from few projects. Unfortunately only fluorescent images.\n\nhttps://www.dropbox.com/sh/i753b966e6zp7t4/AAB6KtMCgcmKzyfdxEzXVIyHa?dl=0", "replies": [ { "id": 311199, "postDate": "2018-04-09T15:50:57.233Z", "content": "Any labels(masks) ? Just RGB Images?
", "rawMarkdown": "Any labels(masks) ? Just RGB Images?" } ] }, { "id": 310781, "postDate": "2018-04-08T15:15:04.700Z", "content": "Using pre-trained model from the MS Coco dataset, as part of the Matterport Mask RCNN implementation: reference https://github.com/matterport/Mask_RCNN/releases
", "rawMarkdown": "Using pre-trained model from the MS Coco dataset, as part of the Matterport Mask RCNN implementation: reference https://github.com/matterport/Mask_RCNN/releases" }, { "id": 310744, "postDate": "2018-04-08T13:18:56.753Z", "content": "Deeplab v3, v3+ pretrained weights : https://github.com/tensorflow/models/tree/master/research/deeplab
", "rawMarkdown": "Deeplab v3, v3+ pretrained weights : https://github.com/tensorflow/models/tree/master/research/deeplab" }, { "id": 310701, "postDate": "2018-04-08T10:46:46.070Z", "content": "We used https://www2.eecs.berkeley.edu/Research/Projects/CS/vision/grouping/resources.html for synthetic images generation.\nhttps://zenodo.org/record/1174353#.WprukR9fg8o\nhttps://github.com/lopuhin/kaggle-dsbowl-2018-dataset-fixes
", "rawMarkdown": "We used https://www2.eecs.berkeley.edu/Research/Projects/CS/vision/grouping/resources.html for synthetic images generation.\nhttps://zenodo.org/record/1174353#.WprukR9fg8o\nhttps://github.com/lopuhin/kaggle-dsbowl-2018-dataset-fixes" }, { "id": 310549, "postDate": "2018-04-07T23:19:57.220Z", "content": "https://data.broadinstitute.org/bbbc/BBBC006/, and models posted here
", "rawMarkdown": "https://data.broadinstitute.org/bbbc/BBBC006/, and models posted here " }, { "id": 310501, "postDate": "2018-04-07T18:25:47.367Z", "content": "I have not seen these models linked yet, and I may be using them.
\n\n\n\nhttps://github.com/tensorflow/models/blob/master/research/deeplab/g3doc/model_zoo.md
\n\nhttps://github.com/tensorflow/models/tree/master/research/slim
", "rawMarkdown": "I have not seen these models linked yet, and I may be using them. \n\nhttps://github.com/tensorflow/models/blob/master/research/object_detection/g3doc/detection_model_zoo.md\n\nhttps://github.com/tensorflow/models/blob/master/research/deeplab/g3doc/model_zoo.md\n\nhttps://github.com/tensorflow/models/tree/master/research/slim" }, { "id": 309723, "postDate": "2018-04-05T21:24:20.810Z", "content": "Mask-RCNN model pretrained on imagenet: https://github.com/matterport/Mask_RCNN/releases
", "rawMarkdown": "Mask-RCNN model pretrained on imagenet: https://github.com/matterport/Mask_RCNN/releases" }, { "id": 309653, "postDate": "2018-04-05T18:43:13.517Z", "content": "https://cs.adelaide.edu.au/~carneiro/isbi14_challenge/dataset.html\nhttps://nucleisegmentationbenchmark.weebly.com/dataset.html\nhttps://ome.grc.nia.nih.gov/iicbu2008/\nhttp://rnabiology.ircm.qc.ca/RBPImage/index.php\nhttps://zenodo.org/record/1174353#.WsGiztP4-3V\nhttps://warwick.ac.uk/fac/sci/dcs/research/tia/data/crchistolabelednucleihe\nhttps://becklab.hms.harvard.edu/software/psb-crowdsourced-nuclei-annotation-data-1
", "rawMarkdown": "https://cs.adelaide.edu.au/~carneiro/isbi14_challenge/dataset.html\nhttps://nucleisegmentationbenchmark.weebly.com/dataset.html\nhttps://ome.grc.nia.nih.gov/iicbu2008/\nhttp://rnabiology.ircm.qc.ca/RBPImage/index.php\nhttps://zenodo.org/record/1174353#.WsGiztP4-3V\nhttps://warwick.ac.uk/fac/sci/dcs/research/tia/data/crchistolabelednucleihe\nhttps://becklab.hms.harvard.edu/software/psb-crowdsourced-nuclei-annotation-data-1" }, { "id": 308983, "postDate": "2018-04-04T13:11:22.047Z", "content": "https://github.com/lopuhin/kaggle-dsbowl-2018-dataset-fixes
", "rawMarkdown": " https://github.com/lopuhin/kaggle-dsbowl-2018-dataset-fixes", "replies": [ { "id": 313913, "postDate": "2018-04-14T04:12:05.083Z", "rawMarkdown": "", "isDeleted": true } ] }, { "id": 308964, "postDate": "2018-04-04T12:44:38.007Z", "content": "Small dataset of images with no nuclei is available here https://www.kaggle.com/jakubczakon/improve-the-dataset-with-non-nuclei-images
", "rawMarkdown": "Small dataset of images with no nuclei is available here https://www.kaggle.com/jakubczakon/improve-the-dataset-with-non-nuclei-images" }, { "id": 308889, "postDate": "2018-04-04T09:02:19.087Z", "content": "A Dataset and a Technique for Generalized Nuclear Segmentation in Histological Images\nhttps://nucleisegmentationbenchmark.weebly.com
\n\nHand-segmented 2D Nuclear Images\nhttp://murphylab.web.cmu.edu/data/2009_ISBI_Nuclei.html
\n\nA dataset for nuclei segmentation based on Breast Cancer patients\nhttps://zenodo.org/record/1175282
\n\nMask-RCNN model pretrained on COCO\nhttps://github.com/matterport/Mask_RCNN/releases
", "rawMarkdown": "A Dataset and a Technique for Generalized Nuclear Segmentation in Histological Images\nhttps://nucleisegmentationbenchmark.weebly.com\n\nHand-segmented 2D Nuclear Images\nhttp://murphylab.web.cmu.edu/data/2009_ISBI_Nuclei.html\n\nA dataset for nuclei segmentation based on Breast Cancer patients\nhttps://zenodo.org/record/1175282\n\nMask-RCNN model pretrained on COCO\nhttps://github.com/matterport/Mask_RCNN/releases" }, { "id": 308776, "postDate": "2018-04-04T04:27:35.090Z", "content": "https://github.com/matterport/Mask_RCNN/releases \n1.we use the link to fix the wrong labels(https://github.com/lopuhin/kaggle-dsbowl-2018-dataset-fixes).
\n\n2.https://www.kaggle.com/voglinio/external-h-e-data-with-mask-annotations
\n\n3.https://nucleisegmentationbenchmark.weebly.com/
\n\n4.https://becklab.hms.harvard.edu/software/psb-crowdsourced-nuclei-annotation-data-1
\n\n5.https://cs.adelaide.edu.au/~carneiro/isbi14_challenge/\nBroad Institute Image Sets: https://data.broadinstitute.org/bbbc/image_sets.html
\n\nISBI '14 Pap Smears: https://cs.adelaide.edu.au/~carneiro/isbi14_challenge/dataset.html
\n\nBioimage Datasets: https://bioimage.ucsb.edu/research/bio-segmentation
\n\nNuclear Segmentation for Deep Learning: http://www.andrewjanowczyk.com/use-case-1-nuclei-segmentation/
\n\nMurphy Dataset: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC2901896/\n https://github.com/lopuhin/kaggle-dsbowl-2018-dataset-fixes \nhttps://www.kaggle.com/voglinio/external-h-e-data-with-mask-annotations \nhttps://nucleisegmentationbenchmark.weebly.com/:\n https://zenodo.org/record/1174353#.WprukR9fg8o
", "rawMarkdown": "https://github.com/matterport/Mask_RCNN/releases \n1.we use the link to fix the wrong labels(https://github.com/lopuhin/kaggle-dsbowl-2018-dataset-fixes).\n\n2.https://www.kaggle.com/voglinio/external-h-e-data-with-mask-annotations\n\n3.https://nucleisegmentationbenchmark.weebly.com/\n\n4.https://becklab.hms.harvard.edu/software/psb-crowdsourced-nuclei-annotation-data-1\n\n5.https://cs.adelaide.edu.au/~carneiro/isbi14_challenge/\nBroad Institute Image Sets: https://data.broadinstitute.org/bbbc/image_sets.html\n\nISBI '14 Pap Smears: https://cs.adelaide.edu.au/~carneiro/isbi14_challenge/dataset.html\n\nBioimage Datasets: https://bioimage.ucsb.edu/research/bio-segmentation\n\nNuclear Segmentation for Deep Learning: http://www.andrewjanowczyk.com/use-case-1-nuclei-segmentation/\n\nMurphy Dataset: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC2901896/\n https://github.com/lopuhin/kaggle-dsbowl-2018-dataset-fixes \nhttps://www.kaggle.com/voglinio/external-h-e-data-with-mask-annotations \nhttps://nucleisegmentationbenchmark.weebly.com/:\n https://zenodo.org/record/1174353#.WprukR9fg8o\n" }, { "id": 308774, "postDate": "2018-04-04T04:20:18.373Z", "content": "https://downloads.openmicroscopy.org/images/
\n\nhttps://www.openmicroscopy.org/
\n\nhttps://warwick.ac.uk/fac/sci/dcs/research/tia/data/
", "rawMarkdown": "https://downloads.openmicroscopy.org/images/\n\nhttps://www.openmicroscopy.org/\n\nhttps://warwick.ac.uk/fac/sci/dcs/research/tia/data/\n" }, { "id": 308773, "postDate": "2018-04-04T04:19:46.607Z", "content": "https://github.com/sanuj/nuclei-net
\n\nhttps://sanuj.github.io/nuclei-net-btp.html
\n\nhttps://nucleisegmentationbenchmark.weebly.com/dataset.html
\n\nhttps://sites.google.com/site/aacruzr/image-datasets, see (Histology and Histopathology, etc)
\n\nhttp://histo.anat.ubc.ca/HISTOLOGY/Basic%20Tissues/view.apml?
\n\nhuman altas
\n\nhttps://www.proteinatlas.org/humanproteome
", "rawMarkdown": "https://github.com/sanuj/nuclei-net\n\nhttps://sanuj.github.io/nuclei-net-btp.html\n\nhttps://nucleisegmentationbenchmark.weebly.com/dataset.html\n\nhttps://sites.google.com/site/aacruzr/image-datasets, see (Histology and Histopathology, etc)\n\nhttp://histo.anat.ubc.ca/HISTOLOGY/Basic%20Tissues/view.apml?\n\nhuman altas\n\nhttps://www.proteinatlas.org/humanproteome" }, { "id": 308771, "postDate": "2018-04-04T04:18:59.793Z", "content": "http://histo.anat.ubc.ca/HISTOLOGY/Basic%20Tissues/view.apml\nhttps://ome.grc.nia.nih.gov/iicbu2008/\nhttps://data.broadinstitute.org/bbbc/BBBC006/
\n\nhttps://bioimage.ucsb.edu/research/bio-segmentation
", "rawMarkdown": "http://histo.anat.ubc.ca/HISTOLOGY/Basic%20Tissues/view.apml\nhttps://ome.grc.nia.nih.gov/iicbu2008/\nhttps://data.broadinstitute.org/bbbc/BBBC006/\n\nhttps://bioimage.ucsb.edu/research/bio-segmentation" }, { "id": 308738, "postDate": "2018-04-04T02:04:49.227Z", "content": "but I may use some of the following datasets on the final solution. They all have already been shared.
\n\n| dataset name | link | |--------------|----------------| | crchistophenotypes_2016_04_28 | https://warwick.ac.uk/fac/sci/dcs/research/tia/data/crchistolabelednucleihe/ | | nuclei-net-master | https://github.com/sanuj/nuclei-net | | nucleisegmentationbenchmark | https://nucleisegmentationbenchmark.weebly.com/dataset.html | | isbi14_challenge | https://cs.adelaide.edu.au/~carneiro/isbi14_challenge/dataset.html | | TNBC_NucleiSegmentation | https://zenodo.org/record/1175282#.WsOx73XFJFs | | nuclei | http://andrewjanowczyk.com/wp-static/nuclei.tgz | | PSB_2015_ImageSize_400 | https://becklab.hms.harvard.edu/software/psb-crowdsourced-nuclei-annotation-data-1 | | cluster nuclei | https://imagej.nih.gov/ij/plugins/ihc-toolbox/index.html |
\n\nDataset at:
\n\nhttps://github.com/sanuj/nuclei-net
\n\nhttps://sanuj.github.io/nuclei-net-btp.html
\n\nhttps://nucleisegmentationbenchmark.weebly.com/dataset.html
\n\nhttps://sites.google.com/site/aacruzr/image-datasets, see (Histology and Histopathology, etc)
\n\nhttp://histo.anat.ubc.ca/HISTOLOGY/Basic%20Tissues/view.apml?
\n\nhuman altas
\n\nhttps://www.proteinatlas.org/humanproteome
\n\nAlso COCO weights from https://github.com/matterport/Mask_RCNN/releases may be used.
\n\nhttps://downloads.openmicroscopy.org/images/
\n\nhttps://www.openmicroscopy.org/
\n\nhttps://warwick.ac.uk/fac/sci/dcs/research/tia/data/
\n\nSome potential resources with related datasets :
\n\nA comprehensive tutorial with selected use cases http://www.andrewjanowczyk.com/use-case-1-nuclei-segmentation/ (dataset : http://andrewjanowczyk.com/wp-static/nuclei.tgz)
\n\nA Dataset and a Technique for Generalized Nuclear Segmentation in Histological Images https://nucleisegmentationbenchmark.weebly.com/
\n\n2D images of 3T3 and U20S cells hand-segmented to show nuclear boundaries http://murphylab.web.cmu.edu/data/#2DNuclei (dataset : http://murphylab.web.cmu.edu/data/2009_ISBI_Nuclei.html)
\n\nDigital Pathology : Nuclei Segmentation MICCAI 2017 Challenge http://miccai.cloudapp.net/competitions/57 (currently not responding)
\n\nMitochondria segmentation dataset https://cvlab.epfl.ch/data/em
", "rawMarkdown": "but I may use some of the following datasets on the final solution. They all have already been shared.\n\n| dataset name | link | |--------------|----------------| | crchistophenotypes_2016_04_28 | https://warwick.ac.uk/fac/sci/dcs/research/tia/data/crchistolabelednucleihe/ | | nuclei-net-master | https://github.com/sanuj/nuclei-net | | nucleisegmentationbenchmark | https://nucleisegmentationbenchmark.weebly.com/dataset.html | | isbi14_challenge | https://cs.adelaide.edu.au/~carneiro/isbi14_challenge/dataset.html | | TNBC_NucleiSegmentation | https://zenodo.org/record/1175282#.WsOx73XFJFs | | nuclei | http://andrewjanowczyk.com/wp-static/nuclei.tgz | | PSB_2015_ImageSize_400 | https://becklab.hms.harvard.edu/software/psb-crowdsourced-nuclei-annotation-data-1 | | cluster nuclei | https://imagej.nih.gov/ij/plugins/ihc-toolbox/index.html |\n\nDataset at:\n\nhttps://github.com/sanuj/nuclei-net\n\nhttps://sanuj.github.io/nuclei-net-btp.html\n\nhttps://nucleisegmentationbenchmark.weebly.com/dataset.html\n\nhttps://sites.google.com/site/aacruzr/image-datasets, see (Histology and Histopathology, etc)\n\nhttp://histo.anat.ubc.ca/HISTOLOGY/Basic%20Tissues/view.apml?\n\nhuman altas\n\nhttps://www.proteinatlas.org/humanproteome\n\nAlso COCO weights from https://github.com/matterport/Mask_RCNN/releases may be used.\n\nhttps://downloads.openmicroscopy.org/images/\n\nhttps://www.openmicroscopy.org/\n\nhttps://warwick.ac.uk/fac/sci/dcs/research/tia/data/\n\n\nSome potential resources with related datasets :\n\nA comprehensive tutorial with selected use cases http://www.andrewjanowczyk.com/use-case-1-nuclei-segmentation/ (dataset : http://andrewjanowczyk.com/wp-static/nuclei.tgz)\n\nA Dataset and a Technique for Generalized Nuclear Segmentation in Histological Images https://nucleisegmentationbenchmark.weebly.com/\n\n2D images of 3T3 and U20S cells hand-segmented to show nuclear boundaries http://murphylab.web.cmu.edu/data/#2DNuclei (dataset : http://murphylab.web.cmu.edu/data/2009_ISBI_Nuclei.html)\n\nDigital Pathology : Nuclei Segmentation MICCAI 2017 Challenge http://miccai.cloudapp.net/competitions/57 (currently not responding)\n\nMitochondria segmentation dataset https://cvlab.epfl.ch/data/em" }, { "id": 308707, "postDate": "2018-04-04T00:45:59.357Z", "content": "Using tensorflow slim pre-trained weights:
\n\nhttps://github.com/tensorflow/models/tree/master/research/slim
", "rawMarkdown": "Using tensorflow slim pre-trained weights:\n\nhttps://github.com/tensorflow/models/tree/master/research/slim" }, { "id": 308673, "postDate": "2018-04-03T22:32:55.070Z", "content": "Anyone happen to know if our training data isn't partially taken from https://data.broadinstitute.org/bbbc/BBBC006/ ?
", "rawMarkdown": "Anyone happen to know if our training data isn't partially taken from https://data.broadinstitute.org/bbbc/BBBC006/ ?" }, { "id": 308634, "postDate": "2018-04-03T20:30:52.193Z", "content": "https://tma.im/cgi-bin/home.pl
\n\nand also
\n\n", "rawMarkdown": "https://tma.im/cgi-bin/home.pl\n\nand also\n\nhttps://openi.nlm.nih.gov/" }, { "id": 308572, "postDate": "2018-04-03T17:35:50.443Z", "content": "Nothing new in this post, but I may use some of the following datasets on the final solution. They all have already been shared.
\n\n| dataset name | link | \n|--------------|----------------|\n| crchistophenotypes_2016_04_28 | https://warwick.ac.uk/fac/sci/dcs/research/tia/data/crchistolabelednucleihe/ |\n| nuclei-net-master | https://github.com/sanuj/nuclei-net |\n| nucleisegmentationbenchmark | https://nucleisegmentationbenchmark.weebly.com/dataset.html |\n| isbi14_challenge | https://cs.adelaide.edu.au/~carneiro/isbi14_challenge/dataset.html |\n| TNBC_NucleiSegmentation | https://zenodo.org/record/1175282#.WsOx73XFJFs |\n| nuclei | http://andrewjanowczyk.com/wp-static/nuclei.tgz |\n| PSB_2015_ImageSize_400 | https://becklab.hms.harvard.edu/software/psb-crowdsourced-nuclei-annotation-data-1 |\n| cluster nuclei | https://imagej.nih.gov/ij/plugins/ihc-toolbox/index.html |
\n\nAlso COCO weights from https://github.com/matterport/Mask_RCNN/releases may be used.
\n\nSupport for markdown tables will be appreciated.
", "rawMarkdown": "Nothing new in this post, but I may use some of the following datasets on the final solution. They all have already been shared.\n\n| dataset name | link | \n|--------------|----------------|\n| crchistophenotypes_2016_04_28 | https://warwick.ac.uk/fac/sci/dcs/research/tia/data/crchistolabelednucleihe/ |\n| nuclei-net-master | https://github.com/sanuj/nuclei-net |\n| nucleisegmentationbenchmark | https://nucleisegmentationbenchmark.weebly.com/dataset.html |\n| isbi14_challenge | https://cs.adelaide.edu.au/~carneiro/isbi14_challenge/dataset.html |\n| TNBC_NucleiSegmentation | https://zenodo.org/record/1175282#.WsOx73XFJFs |\n| nuclei | http://andrewjanowczyk.com/wp-static/nuclei.tgz |\n| PSB_2015_ImageSize_400 | https://becklab.hms.harvard.edu/software/psb-crowdsourced-nuclei-annotation-data-1 |\n| cluster nuclei | https://imagej.nih.gov/ij/plugins/ihc-toolbox/index.html |\n\nAlso COCO weights from https://github.com/matterport/Mask_RCNN/releases may be used.\n\nSupport for markdown tables will be appreciated." }, { "id": 308464, "postDate": "2018-04-03T14:48:56.480Z", "content": "Mask-RCNN model pretrained on COCO: https://github.com/matterport/Mask_RCNN/releases \nDetectron (https://github.com/facebookresearch/Detectron) along with several weight files from the model zoo (https://github.com/facebookresearch/Detectron/blob/master/MODEL_ZOO.md).\nOf course Allen's classes.csv from https://www.kaggle.com/c/data-science-bowl-2018/discussion/48130.
", "rawMarkdown": "Mask-RCNN model pretrained on COCO: https://github.com/matterport/Mask_RCNN/releases \nDetectron (https://github.com/facebookresearch/Detectron) along with several weight files from the model zoo (https://github.com/facebookresearch/Detectron/blob/master/MODEL_ZOO.md).\nOf course Allen's classes.csv from https://www.kaggle.com/c/data-science-bowl-2018/discussion/48130.\n" }, { "id": 308426, "postDate": "2018-04-03T13:49:52.427Z", "content": "http://www.robots.ox.ac.uk/~vgg/research/mitosis/\nAutomated Labelling of Cell Cycle Phases
", "rawMarkdown": "http://www.robots.ox.ac.uk/~vgg/research/mitosis/\nAutomated Labelling of Cell Cycle Phases" }, { "id": 308286, "postDate": "2018-04-03T09:19:57.170Z", "content": "We use the External Data, so we claim it here.
\n\n1.we use the link to fix the wrong labels(https://github.com/lopuhin/kaggle-dsbowl-2018-dataset-fixes).
\n\n2.https://www.kaggle.com/voglinio/external-h-e-data-with-mask-annotations
\n\n3.https://nucleisegmentationbenchmark.weebly.com/
\n\n4.https://becklab.hms.harvard.edu/software/psb-crowdsourced-nuclei-annotation-data-1
\n\n5.https://cs.adelaide.edu.au/~carneiro/isbi14_challenge/
", "rawMarkdown": "We use the External Data, so we claim it here.\n\n1.we use the link to fix the wrong labels(https://github.com/lopuhin/kaggle-dsbowl-2018-dataset-fixes).\n\n2.https://www.kaggle.com/voglinio/external-h-e-data-with-mask-annotations\n\n3.https://nucleisegmentationbenchmark.weebly.com/\n\n4.https://becklab.hms.harvard.edu/software/psb-crowdsourced-nuclei-annotation-data-1\n\n5.https://cs.adelaide.edu.au/~carneiro/isbi14_challenge/\n\n" }, { "id": 308204, "postDate": "2018-04-03T05:51:52.120Z", "content": "Some (mostly unannotated) phase contrast images of cells & nuclei from Google Images, filtered to \"Labeled for reuse with modification. Phase contrast is probably the most difficult imaging modality to generalize to.
", "rawMarkdown": "[Some (mostly unannotated) phase contrast images of cells & nuclei from Google Images, filtered to \"Labeled for reuse with modification](https://www.google.com/search?q=phase+contrast+(cells%7Cnuclei%7Ccytology)&tbm=isch&tbs=sur:fmc). Phase contrast is probably the most difficult imaging modality to generalize to." }, { "id": 308159, "postDate": "2018-04-03T03:47:34.223Z", "content": "How about pretrained models included in common ML libraries? For example, I'm using a pretrained VGG-19 from PyTorch/torchvision (models.vgg19(pretrained=True)) at the moment, though that may change. I thought I'd read somewhere that if it was included in a popular library it didn't need to be declared here (because it would already be available to everyone), but I'm not sure.
Just in case, the torchvision models are here.
", "rawMarkdown": "How about pretrained models included in common ML libraries? For example, I'm using a pretrained VGG-19 from PyTorch/torchvision (`models.vgg19(pretrained=True)`) at the moment, though that may change. I thought I'd read somewhere that if it was included in a popular library it didn't need to be declared here (because it would already be available to everyone), but I'm not sure.\n\nJust in case, the torchvision models are [here](http://pytorch.org/docs/master/torchvision/models.html)." }, { "id": 308046, "postDate": "2018-04-02T21:31:21.777Z", "content": "Broad Institute Image Sets:\nhttps://data.broadinstitute.org/bbbc/image_sets.html
\n\nISBI '14 Pap Smears:\nhttps://cs.adelaide.edu.au/~carneiro/isbi14_challenge/dataset.html
\n\nBioimage Datasets:\nhttps://bioimage.ucsb.edu/research/bio-segmentation
\n\nNuclear Segmentation for Deep Learning:\nhttp://www.andrewjanowczyk.com/use-case-1-nuclei-segmentation/
\n\nMurphy Dataset:\nhttps://www.ncbi.nlm.nih.gov/pmc/articles/PMC2901896/
", "rawMarkdown": "Broad Institute Image Sets:\nhttps://data.broadinstitute.org/bbbc/image_sets.html\n\nISBI '14 Pap Smears:\nhttps://cs.adelaide.edu.au/~carneiro/isbi14_challenge/dataset.html\n\nBioimage Datasets:\nhttps://bioimage.ucsb.edu/research/bio-segmentation\n\nNuclear Segmentation for Deep Learning:\nhttp://www.andrewjanowczyk.com/use-case-1-nuclei-segmentation/\n\nMurphy Dataset:\nhttps://www.ncbi.nlm.nih.gov/pmc/articles/PMC2901896/" }, { "id": 307999, "postDate": "2018-04-02T19:27:59.227Z", "content": "Fixed Stage 1 Train Masks by lopuhin: \nhttps://github.com/lopuhin/kaggle-dsbowl-2018-dataset-fixes
\n\nH&E Dataset by voglinio: \nhttps://www.kaggle.com/voglinio/external-h-e-data-with-mask-annotations\nBased on:\nhttps://nucleisegmentationbenchmark.weebly.com/
\n\nA dataset for nuclei segmentation based on Breast Cancer patients:\nhttps://zenodo.org/record/1174353#.WprukR9fg8o
", "rawMarkdown": "Fixed Stage 1 Train Masks by lopuhin: \nhttps://github.com/lopuhin/kaggle-dsbowl-2018-dataset-fixes\n\nH&E Dataset by voglinio: \nhttps://www.kaggle.com/voglinio/external-h-e-data-with-mask-annotations\nBased on:\nhttps://nucleisegmentationbenchmark.weebly.com/\n\nA dataset for nuclei segmentation based on Breast Cancer patients:\nhttps://zenodo.org/record/1174353#.WprukR9fg8o", "replies": [ { "id": 308660, "postDate": "2018-04-03T21:48:29.643Z", "content": "https://cs.adelaide.edu.au/~carneiro/isbi14_challenge/dataset.html
", "rawMarkdown": "https://cs.adelaide.edu.au/~carneiro/isbi14_challenge/dataset.html" } ] }, { "id": 307499, "postDate": "2018-04-01T20:35:44.947Z", "content": "https://warwick.ac.uk/fac/sci/dcs/research/tia/data/crchistolabelednucleihe\nhttp://www.andrewjanowczyk.com/use-case-1-nuclei-segmentation
", "rawMarkdown": "https://warwick.ac.uk/fac/sci/dcs/research/tia/data/crchistolabelednucleihe\nhttp://www.andrewjanowczyk.com/use-case-1-nuclei-segmentation" }, { "id": 307488, "postDate": "2018-04-01T19:26:01.387Z", "content": "https://nucleisegmentationbenchmark.weebly.com/
", "rawMarkdown": "https://nucleisegmentationbenchmark.weebly.com/" }, { "id": 307417, "postDate": "2018-04-01T15:13:51.013Z", "content": "https://github.com/gigascience/paper-bray2017
\n\nSupporting data for \"A dataset of images and morphological profiles of 30,000 small-molecule treatments using the Cell Painting assay\"
\n\nhttp://www.cellimagelibrary.org/browse/celltype
", "rawMarkdown": "https://github.com/gigascience/paper-bray2017\n\n\nSupporting data for \"A dataset of images and morphological profiles of 30,000 small-molecule treatments using the Cell Painting assay\"\n\nhttp://www.cellimagelibrary.org/browse/celltype" }, { "id": 307275, "postDate": "2018-04-01T09:09:31.907Z", "content": "https://zenodo.org/record/1174353#.WprukR9fg8o
", "rawMarkdown": "https://zenodo.org/record/1174353#.WprukR9fg8o" }, { "id": 306863, "postDate": "2018-03-31T07:29:55.263Z", "content": "hence I assume that any programming language other than Python like java etc will also work right ?
", "rawMarkdown": "hence I assume that any programming language other than Python like java etc will also work right ? " }, { "id": 306847, "postDate": "2018-03-31T06:29:43.633Z", "content": "Do we need to use specific programming languages like Python , R ? or We can use any other programming languages like java ? In the competition rules its mentioned we have to develop the \"algorithm\" they have not specified any language restrictions hence I assume that any programming language other than java will also work right ?
", "rawMarkdown": "Do we need to use specific programming languages like Python , R ? or We can use any other programming languages like java ? In the competition rules its mentioned we have to develop the \"algorithm\" they have not specified any language restrictions hence I assume that any programming language other than java will also work right ? ", "replies": [ { "id": 308157, "postDate": "2018-04-03T03:40:49.993Z", "content": "Any language you want. Submissions are just the output of your code. If you want to be eligible to win a prize, though, you do have to submit your code (in any language) at the stage 1 deadline.
", "rawMarkdown": "Any language you want. Submissions are just the output of your code. If you want to be eligible to win a prize, though, you do have to submit your code (in any language) at the stage 1 deadline." } ] }, { "id": 306830, "postDate": "2018-03-31T05:15:10.087Z", "content": "https://www.kaggle.com/voglinio/external-h-e-data-with-mask-annotations\nhttps://www.kaggle.com/bonlime/train-test-image-mosaic\nhttps://github.com/lopuhin/kaggle-dsbowl-2018-dataset-fixes
", "rawMarkdown": "https://www.kaggle.com/voglinio/external-h-e-data-with-mask-annotations\nhttps://www.kaggle.com/bonlime/train-test-image-mosaic\nhttps://github.com/lopuhin/kaggle-dsbowl-2018-dataset-fixes" }, { "id": 306192, "postDate": "2018-03-30T02:37:59.770Z", "content": "https://imagej.nih.gov/ij/plugins/ihc-toolbox/index.html
", "rawMarkdown": "https://imagej.nih.gov/ij/plugins/ihc-toolbox/index.html" }, { "id": 304897, "postDate": "2018-03-28T07:20:36.867Z", "content": "https://commons.wikimedia.org/w/index.php?title=Category:Histopathology
", "rawMarkdown": "https://commons.wikimedia.org/w/index.php?title=Category:Histopathology" }, { "id": 304302, "postDate": "2018-03-27T10:28:07.730Z", "content": "https://storage.googleapis.com/kaggle-forum-message-attachments/273531/8366/classes.csv
", "rawMarkdown": "https://kaggle2.blob.core.windows.net/forum-message-attachments/273531/8366/classes.csv" }, { "id": 303714, "postDate": "2018-03-26T15:22:15.730Z", "content": "https://data.broadinstitute.org/bbbc/image_sets.html
", "rawMarkdown": "https://data.broadinstitute.org/bbbc/image_sets.html" }, { "id": 303452, "postDate": "2018-03-26T07:59:47.043Z", "content": "I'll use fixed stage1_train masks https://github.com/lopuhin/kaggle-dsbowl-2018-dataset-fixes
", "rawMarkdown": "I'll use fixed stage1_train masks https://github.com/lopuhin/kaggle-dsbowl-2018-dataset-fixes" }, { "id": 303227, "postDate": "2018-03-25T20:01:49.400Z", "content": "Pretrained models from torchvision as well as pretrained models linked in faster-rcnn-pytorch
", "rawMarkdown": "Pretrained models from torchvision as well as pretrained models linked in [faster-rcnn-pytorch][1]\n\n\n [1]: https://github.com/jwyang/faster-rcnn.pytorch" }, { "id": 301461, "postDate": "2018-03-22T19:34:19.173Z", "content": "ImageNet weights for Keras: https://keras.io/applications/
", "rawMarkdown": "ImageNet weights for Keras: https://keras.io/applications/" }, { "id": 293522, "postDate": "2018-03-10T03:31:38.247Z", "content": "http://rnabiology.ircm.qc.ca/RBPImage/index.php
", "rawMarkdown": "http://rnabiology.ircm.qc.ca/RBPImage/index.php" }, { "id": 292546, "postDate": "2018-03-08T06:12:40.377Z", "content": "Hello, I have a question. If I use the data other people post here should I mention it? Won't this create an echo effect every time somebody publishes a new DB? Is it enough to say I will try to use all the data available here? If yes, it is said. Thanks
", "rawMarkdown": "Hello, I have a question. If I use the data other people post here should I mention it? Won't this create an echo effect every time somebody publishes a new DB? Is it enough to say I will try to use all the data available here? If yes, it is said. Thanks", "replies": [ { "id": 293409, "postDate": "2018-03-09T19:34:17.383Z", "content": "You do not need to mention, in this thread, your use of data in this thread. Once people post external data in this thread, it means it's fair game for anyone else to use (subject to license restrictions on the data of course).
\n\nYou only need to mention use of external data when submitting your solution (privately).
", "rawMarkdown": "You do not need to mention, in this thread, your use of data in this thread. Once people post external data in this thread, it means it's fair game for anyone else to use (subject to license restrictions on the data of course).\n\nYou only need to mention use of external data when submitting your solution (privately).", "votes": 1 } ] }, { "id": 292005, "postDate": "2018-03-07T07:57:54.270Z", "content": "I am new to Kaggle. Is there a way to upload fixed-up or external data onto a notebook, so that we can run our kernel against that new data? I can't seem to find the upload button. Many of the public kernels are accessing the provided data under \"../input\", so it'd be nice to be able to add/replace the data there. Sorry if this is something everyone already knows how to do.
", "rawMarkdown": "I am new to Kaggle. Is there a way to upload fixed-up or external data onto a notebook, so that we can run our kernel against that new data? I can't seem to find the upload button. Many of the public kernels are accessing the provided data under \"../input\", so it'd be nice to be able to add/replace the data there. Sorry if this is something everyone already knows how to do.", "replies": [ { "id": 294058, "postDate": "2018-03-11T07:19:49.030Z", "content": "Does anyone know the answer to this? Just wondering how people make use of the fixed-up (or external) data.
", "rawMarkdown": "Does anyone know the answer to this? Just wondering how people make use of the fixed-up (or external) data." }, { "id": 294676, "postDate": "2018-03-12T12:03:06.950Z", "content": "I suggest starting a new thread with the question so it gets more attention; I don't know the answer myself.
", "rawMarkdown": "I suggest starting a new thread with the question so it gets more attention; I don't know the answer myself." }, { "id": 295129, "postDate": "2018-03-13T05:12:18.343Z", "content": "Thanks Anne! I'll do that.
", "rawMarkdown": "Thanks Anne! I'll do that." } ] }, { "id": 290099, "postDate": "2018-03-02T23:29:51.303Z", "content": "https://github.com/facebookresearch/Detectron
", "rawMarkdown": "https://github.com/facebookresearch/Detectron" }, { "id": 289215, "postDate": "2018-03-01T06:24:50.940Z", "content": "Images from wikimedia \nhttps://commons.wikimedia.org/wiki/Category:Cytopathology_of_cancers
", "rawMarkdown": "Images from wikimedia \nhttps://commons.wikimedia.org/wiki/Category:Cytopathology_of_cancers" }, { "id": 281586, "postDate": "2018-02-12T21:12:53.807Z", "content": "Can I use the pre-trained model on COCO datasets for competition for prize?
", "rawMarkdown": "Can I use the pre-trained model on COCO datasets for competition for prize?", "replies": [ { "id": 282731, "postDate": "2018-02-14T13:32:03.137Z", "content": "Sure
", "rawMarkdown": "Sure", "votes": 1 }, { "id": 285729, "postDate": "2018-02-20T22:41:10.147Z", "content": "You need to post the link(s) to them here and then you're good to go.
", "rawMarkdown": "You need to post the link(s) to them here and then you're good to go. " }, { "id": 291534, "postDate": "2018-03-06T10:02:56.763Z", "content": "Mask-RCNN model pretrained on COCO:\nhttps://github.com/matterport/Mask_RCNN/releases
", "rawMarkdown": "Mask-RCNN model pretrained on COCO:\nhttps://github.com/matterport/Mask_RCNN/releases" } ] }, { "id": 280402, "postDate": "2018-02-09T22:27:03.320Z", "content": "Found one more dataset from CellTrackingChallenge: http://www.celltrackingchallenge.net/datasets.html
\n\nThere are plenty of time-lapse data:
", "rawMarkdown": "Found one more dataset from CellTrackingChallenge: http://www.celltrackingchallenge.net/datasets.html\n\nThere are plenty of time-lapse data:\n\n" }, { "id": 280159, "postDate": "2018-02-09T13:44:51.420Z", "content": "Tumor Proliferation Assessment Challenge 2016\nhttp://tupac.tue-image.nl
", "rawMarkdown": "Tumor Proliferation Assessment Challenge 2016\nhttp://tupac.tue-image.nl" }, { "id": 279666, "postDate": "2018-02-08T13:56:50.620Z", "content": "50 H&E stained histopathology images, breast cancer. This are images with class (purple) that has low samples count in train set and included in test set. http://bioimage.ucsb.edu/research/bio-segmentation
\n\n@inproceedings{\nDrelie08-298,\nauthor = {Elisa Drelie Gelasca and Jiyun Byun and Boguslaw Obara and B.S. Manjunath},\ntitle = {Evaluation and Benchmark for Biological Image Segmentation},\nbooktitle = {IEEE International Conference on Image Processing},\nlocation = {San Diego, CA},\nmonth = {Oct},\nyear = {2008},\nurl = {http://vision.ece.ucsb.edu/publications/elisa_ICIP08.pdf}\n}
", "rawMarkdown": "50 H&E stained histopathology images, breast cancer. This are images with class (purple) that has low samples count in train set and included in test set. [http://bioimage.ucsb.edu/research/bio-segmentation][1]\n\n@inproceedings{\nDrelie08-298,\nauthor = {Elisa Drelie Gelasca and Jiyun Byun and Boguslaw Obara and B.S. Manjunath},\ntitle = {Evaluation and Benchmark for Biological Image Segmentation},\nbooktitle = {IEEE International Conference on Image Processing},\nlocation = {San Diego, CA},\nmonth = {Oct},\nyear = {2008},\nurl = {http://vision.ece.ucsb.edu/publications/elisa_ICIP08.pdf}\n} \n\n\n [1]: http://bioimage.ucsb.edu/research/bio-segmentation" }, { "id": 277409, "postDate": "2018-02-03T02:12:59.880Z", "content": "Hi Anne, I am planning to get nuclei images from google to supplement the training set. I have applied the search filter of \"Labeled or Reuse\" on Google. Can I use these images for training?
", "rawMarkdown": "Hi Anne, I am planning to get nuclei images from google to supplement the training set. I have applied the search filter of \"Labeled or Reuse\" on Google. Can I use these images for training?" }, { "id": 309640, "postDate": "2018-04-05T17:59:44.577Z", "content": "PyTorch Mask RCNN and Coco pre-trained weights:\nhttps://github.com/multimodallearning/pytorch-mask-rcnn
", "rawMarkdown": "PyTorch Mask RCNN and Coco pre-trained weights:\nhttps://github.com/multimodallearning/pytorch-mask-rcnn", "isDeleted": true }, { "id": 308512, "postDate": "2018-04-03T16:03:11.370Z", "rawMarkdown": "", "isDeleted": true } ], "comments": [ { "id": 269406, "author_name": "Alexandre Cadrin-Chênevert", "author_url": "", "post_date": "2018-01-16T19:13:33.520000", "content": "Some potential resources with related datasets :
\n\nA comprehensive tutorial with selected use cases\nhttp://www.andrewjanowczyk.com/use-case-1-nuclei-segmentation/ (dataset : http://andrewjanowczyk.com/wp-static/nuclei.tgz)
\n\nA Dataset and a Technique for Generalized Nuclear Segmentation in Histological Images\nhttps://nucleisegmentationbenchmark.weebly.com/
\n\n2D images of 3T3 and U20S cells hand-segmented to show nuclear boundaries\nhttp://murphylab.web.cmu.edu/data/#2DNuclei (dataset : http://murphylab.web.cmu.edu/data/2009_ISBI_Nuclei.html)
\n\nDigital Pathology : Nuclei Segmentation MICCAI 2017 Challenge\nhttp://miccai.cloudapp.net/competitions/57 (currently not responding)
\n\nMitochondria segmentation dataset\nhttps://cvlab.epfl.ch/data/em
", "votes": 30, "replies": [ { "id": 281896, "author_name": "Riccardo Gallina", "author_url": "", "post_date": "2018-02-13T08:40:04.667000", "content": "Thanks for sharing!
", "votes": 2, "replies": [] }, { "id": 284570, "author_name": "Soonhwan Kwon", "author_url": "", "post_date": "2018-02-17T15:50:27.130000", "content": "thank you for great references!
", "votes": 0, "replies": [] }, { "id": 305078, "author_name": "DStjhb", "author_url": "", "post_date": "2018-03-28T13:14:12.490000", "content": "Thanks for these links. Very helpful. The Github of the second set (Histological Images) is also useful.
", "votes": 1, "replies": [] } ] }, { "id": 288797, "author_name": "Konstantin Lopukhin", "author_url": "", "post_date": "2018-02-28T07:45:56.750000", "content": "I'll use fixed stage1_train masks https://github.com/lopuhin/kaggle-dsbowl-2018-dataset-fixes
", "votes": 17, "replies": [ { "id": 288844, "author_name": "CPMP", "author_url": "", "post_date": "2018-02-28T10:16:46.833000", "content": "Thanks for sharing this, saving others lots of work!
", "votes": 1, "replies": [] }, { "id": 288914, "author_name": "Grzegorz Sionkowski", "author_url": "", "post_date": "2018-02-28T14:15:56.263000", "content": "@Konstantin, If you will really use your fixes instead of raw masks, you have no chance to win this competition. First of all you should ask yourself and answer the questions, why some nuclei are not separated and why some others are omitted. Did you do it? I am not a biologist and I do not know the answer for the first question (yet), but I think I know the answer for the second one. E.g. I am 99% sure, your fix regarding the image named a7f7... (adding one more nucleus at the left bottom) is wrong.
", "votes": -4, "replies": [] }, { "id": 288919, "author_name": "Konstantin Lopukhin", "author_url": "", "post_date": "2018-02-28T14:33:26.383000", "content": "Sorry, I didn't understand the point about winning. I tried to make only fixes confirmed in the data quality issues thread. Edit: and yes, I'm really using them, although I didn't notice a significant change on the public LB.
\n\n\n\n\nE.g. I am 99% sure, your fix regarding the image named a7f7... (adding one more nucleus at the left bottom) is wrong.
\n
You may be right - I was basing my fix on this comment by Anne Carpenter: https://www.kaggle.com/c/data-science-bowl-2018/discussion/47770#272208 but it's indeed not 100% clear if that comment referred to the nuclei annotation on the bottom left of a7f7 that I added - I presume you meant this one?\nI made each fix in a separate commit, so if you are happy with other fixes but not this one, then you can revert this commit in your fork.
", "votes": 2, "replies": [] }, { "id": 289761, "author_name": "CPMP", "author_url": "", "post_date": "2018-03-02T07:54:12.097000", "content": "\n\n\nI didn't notice a significant change on the public LB.
\n
Do you see a change in your local CV score?
\n\nI haven't used your fixed dataset but will probably start today. Thanks again for setting this, and thanks to Kaggle and sponsor for allowing it!
", "votes": 0, "replies": [] }, { "id": 289796, "author_name": "Grzegorz Sionkowski", "author_url": "", "post_date": "2018-03-02T09:31:33.733000", "content": "@Konstantin, Just to clear my previous comment. Train dataset contains evident errors, e.g. a mask in the form of long bar or curious set of masks containing one mask only, where presence of many nuclei is evident. Removing those errors should improve a little bit the quality of the results at any level: local CV, public stage 1 and private stage 2. I can't imagine someone who would prefer a dataset with evident errors over a dataset without them.
\nTrain dataset contains also objects which the annotators did not annotate as nuclei or annotated them by using a mask of strange shape. IMHO, correcting the annotators by adding more masks or separating one mask into many may only accidentally improve the score. Generally, the effect will be the opposite.
Reasumming, I would like to use a branch of train dataset without evident technical errors, but I do not want to lose time for cleaning it from the \"corrections\".
", "votes": 1, "replies": [] }, { "id": 289879, "author_name": "Anne Carpenter", "author_url": "", "post_date": "2018-03-02T14:26:27.160000", "content": "Thanks a lot, Konstantin. This is a GREAT initiative.
\n\nI would suggest that in the official Kaggle discussion thread pointing out errors, that when people make a fix, they reply to the comment that reported the problem and say it's been addressed in your github repo, pointing to the PR or merge, so it's easy to see what's been fixed already vs not. My team can try (no promises! We are volunteers here!) to keep an eye on things where there are open questions, to clarify what would be correct.
", "votes": 2, "replies": [] }, { "id": 437375, "author_name": "Daniel R. Armstrong", "author_url": "", "post_date": "2018-12-11T19:37:34.533000", "content": "Did you see that the 5th place winner used your fixes, https://www.kaggle.com/c/data-science-bowl-2018/discussion/56326. Great work!
", "votes": 1, "replies": [] } ] }, { "id": 275397, "author_name": "Wesley Goi", "author_url": "", "post_date": "2018-01-29T02:50:35.660000", "content": "Think this will be really useful. \nLink to dataset
\n\nBasically, the cells (brightfield) were first unstained and later stained (nuceli and cytoplasmic) using fluorescence microscopy.
\n\nIt'll be both a training set for brightfield and flourenscent images. Especially useful for helping to locate the nuclei for brightfield samples in the dataset.
\n\n
Abstract
\n\nDeep Convolutional Neural Networks (DCNN) have recently emerged as superior for many image segmentation tasks. The DCNN performance is however heavily dependent on the availability of large amounts of problem-specific training samples. Here we show that DCNNs trained on ground truth created automatically using fluorescently labeled cells, perform similar to manual annotations.
\n\nReference
\n\nSadanandan, SK, Ranefall, P, Le Guyader, S, Wählby, C (2017). Automated Training of Deep Convolutional Neural Networks for Cell Segmentation. Sci Rep, 7, 1:7860.
", "votes": 11, "replies": [] }, { "id": 269253, "author_name": "Joel Lee", "author_url": "", "post_date": "2018-01-16T14:27:07.937000", "content": "Adenocarcinomas with labelled nuclei dataset\nhttps://warwick.ac.uk/fac/sci/dcs/research/tia/data/crchistolabelednucleihe/crchistophenotypes_2016_04_28.zip
\n\nSource: K. Sirinukunwattana, S.E.A. Raza, Y.W Tsang, I.A. Cree, D.R.J. Snead, N.M. Rajpoot, ‘Locality Sensitive Deep Learning for Detection and Classification of Nuclei in Routine Colon Cancer Histology Images,’ IEEE Transactions on Medical Imaging, 2016
\n\nOther Relevant Papers:
\n\nunfortunetly only centers, not masks(((
", "votes": 2, "replies": [] } ] }, { "id": 272203, "author_name": "hengck23", "author_url": "", "post_date": "2018-01-22T13:38:57.733000", "content": "virtual fluorescence microscopy:
\n\nhttp://cbia.fi.muni.cz/projects/simulator.html
\n\nhttp://cbia.fi.muni.cz/projects/cytopacq-a-simulation-toolbox_5.html
\n\nhttp://cbia.fi.muni.cz/simulator/index.php
", "votes": 9, "replies": [] }, { "id": 271360, "author_name": "hengck23", "author_url": "", "post_date": "2018-01-20T08:59:43.077000", "content": "https://cs.adelaide.edu.au/~carneiro/isbi14_challenge/dataset.html
\n\nI find some train_stage1 images are from the cervical cytology images of isbi 14 challenge. I suspect they may be use in test_stage2. If they are not, I can all for isbi14 challenge images for training.
\n\nSo I can I use these images?
", "votes": 8, "replies": [] }, { "id": 312659, "author_name": "QingYuan", "author_url": "", "post_date": "2018-04-12T05:25:24.163000", "content": "Stage 1 test data masks (generated from the csv file): https://github.com/yuanqing811/DSB2018_stage1_test
", "votes": 6, "replies": [] }, { "id": 284672, "author_name": "Aengus Tran", "author_url": "", "post_date": "2018-02-17T22:10:38.287000", "content": "Dear Organisers,
\n\nI need an official word from Anne, Kyle or Allen about whether or not user-generated annotations to images already shared here on the thread are to be posted one week before the deadline.
\n\nMy strategy is to use my existing model to pre-label the images and then manually touch up the missing masks or re-fine poor masks. As such my annotations are semi-automatic and are part of my methodology.
\n\nI will likely abandon this strategy if I must share these masks as they are partly generated by my model.
\n\nIf you could kindly respond in a timely manner that would be much appreciated.
\n\nThanks.
", "votes": 5, "replies": [ { "id": 285727, "author_name": "Wendy Kan", "author_url": "", "post_date": "2018-02-20T22:40:26.633000", "content": "You don't need to post them as that is considered part of your methodology, not external data.
", "votes": 2, "replies": [] } ] }, { "id": 285466, "author_name": "hengck23", "author_url": "", "post_date": "2018-02-20T10:12:45.613000", "content": "https://downloads.openmicroscopy.org/images/
\n\nhttps://www.openmicroscopy.org/
\n\nhttps://warwick.ac.uk/fac/sci/dcs/research/tia/data/
", "votes": 6, "replies": [ { "id": 285943, "author_name": "Kostiantyn Maksymov", "author_url": "", "post_date": "2018-02-21T10:14:19.183000", "content": "Is it data without ground truth masks?
", "votes": 1, "replies": [] }, { "id": 302141, "author_name": "Collin Donahue-Oponski", "author_url": "", "post_date": "2018-03-23T18:23:00.397000", "content": "Great links. Thank you!
", "votes": 1, "replies": [] } ] }, { "id": 271537, "author_name": "hengck23", "author_url": "", "post_date": "2018-01-20T17:42:01.360000", "content": "Dataset at:
\n\nhttps://github.com/sanuj/nuclei-net
\n\nhttps://sanuj.github.io/nuclei-net-btp.html
\n\nhttps://nucleisegmentationbenchmark.weebly.com/dataset.html
\n\nhttps://sites.google.com/site/aacruzr/image-datasets, see (Histology and Histopathology, etc)
\n\nhttp://histo.anat.ubc.ca/HISTOLOGY/Basic%20Tissues/view.apml?
\n\nhuman altas
\n\nhttps://www.proteinatlas.org/humanproteome
", "votes": 6, "replies": [] }, { "id": 282734, "author_name": "S. Panchenko", "author_url": "", "post_date": "2018-02-14T13:45:42.803000", "content": "If we change the train set (correct mistakes in the masks), do we have to post that data here as well?
", "votes": 3, "replies": [ { "id": 285728, "author_name": "Wendy Kan", "author_url": "", "post_date": "2018-02-20T22:40:39.430000", "content": "You don't need to post them as that is considered part of your methodology, not external data.
", "votes": 1, "replies": [] } ] }, { "id": 271270, "author_name": "Serhiy Danchenko", "author_url": "", "post_date": "2018-01-20T01:54:19.310000", "content": "Broad Bioimage Benchmark Collection https://data.broadinstitute.org/bbbc/BBBC006/
", "votes": 4, "replies": [ { "id": 308671, "author_name": "theNickName", "author_url": "", "post_date": "2018-04-03T22:29:22.287000", "content": "Just so you know, the annotations on that dataset are almost as crappy as on our training data. I'm not entirely sure, though, that it's different from the training dataset.\nEdit: no, it may actually be even worse. One could look at mcf-z-stacks-03212011_m19_s1-27 case to see just how awful it is at instance segmentation.
", "votes": 0, "replies": [] } ] }, { "id": 312803, "author_name": "theNickName", "author_url": "", "post_date": "2018-04-12T11:09:09.873000", "content": "As I know I'm definitely not gonna win now, I make my revision of the training set and my annotation tool available at https://github.com/ibmua/data-science-bowl-2018-train-set . I think the training set is probably the best there is, as I went through each white-on-black sample and fixed it by hand. Annotation tool is also pretty useful, though, probably limitedly so now after the first stage is over. Maybe, for some future competitions, I donno.
", "votes": 1, "replies": [ { "id": 313218, "author_name": "theNickName", "author_url": "", "post_date": "2018-04-12T23:59:37.600000", "content": "BBBC6 https://drive.google.com/file/d/16KVLYeKBkk_Y0W4Fra_2AhjmW-p6JjZu/view?usp=sharing\nRed channel is a combo similar to \n383d8ad7069bf55cc2e40f638c10565d9967d25dd38258afb048ab8d012d72ce , 486fe606382b28907ffd6f4e0b45d6a1b1ea3ef798eb59d5a4dd917fa8ec5f90 , 572fe655c8d8091ab4abbcdc47aa9c6989ed4becee377a43cea689903a1364be , 3972f3d313a6f5152801f958f26409fc6c6003646087abe1713139d2ab2adc88
\n\n
(https://cs.adelaide.edu.au/~carneiro/isbi14_challenge/dataset.html) PSB Crowdsourced Nuclei Annotation (https://becklab.hms.harvard.edu/software/psb-crowdsourced-nuclei-annotation-data-1) https://www.kaggle.com/voglinio/external-h-e-data-with-mask-annotations (made by Watt, converted from https://nucleisegmentationbenchmark.weebly.com/)\nhttp://www.andrewjanowczyk.com/use-case-1-nuclei-segmentation/\nhttps://bioimage.ucsb.edu/research/bio-segmentation
", "votes": 1, "replies": [ { "id": 405751, "author_name": "", "author_url": "", "post_date": "2018-10-18T03:43:55.083000", "content": "", "votes": 0, "replies": [] } ] }, { "id": 308586, "author_name": "Timea Toth", "author_url": "", "post_date": "2018-04-03T17:56:35.043000", "content": "We used these images (sources in the 'sources.txt' file)\nhttps://drive.google.com/file/d/1A_x-40lfh9FaBEsRv2zTjjvNOlcItnTX/view?usp=sharing
", "votes": 1, "replies": [] }, { "id": 308476, "author_name": "Mizuki Ono", "author_url": "", "post_date": "2018-04-03T14:54:34.303000", "content": "https://zenodo.org/record/1175282\nhttp://cbia.fi.muni.cz/projects/cytopacq-a-simulation-toolbox_6.html
", "votes": 1, "replies": [ { "id": 308536, "author_name": "Guillermo Barbadillo", "author_url": "", "post_date": "2018-04-03T16:56:33.307000", "content": "Hi,\nThanks for the links, however the second one is broken.
", "votes": 0, "replies": [] } ] }, { "id": 307796, "author_name": "Selim Seferbekov", "author_url": "", "post_date": "2018-04-02T13:36:06.493000", "content": "Our team will probably use some pretrained models from the following list:
\n\nhttps://github.com/Cadene/pretrained-models.pytorch\nhttps://keras.io/applications/\nhttps://github.com/fizyr/keras-models/releases/tag/v0.0.1\nhttps://mxnet.incubator.apache.org/api/python/gluon/model_zoo.html
\n\nAdditionally we use some external datasets mentioned in this thread.
", "votes": 1, "replies": [] }, { "id": 307460, "author_name": "Akila Wajirasena", "author_url": "", "post_date": "2018-04-01T16:50:21.557000", "content": "PSB Crowdsourced Nuclei Annotation
\n\nhttps://becklab.hms.harvard.edu/software/psb-crowdsourced-nuclei-annotation-data-1
", "votes": 1, "replies": [] }, { "id": 269385, "author_name": "Thomas O'Malley", "author_url": "", "post_date": "2018-01-16T18:17:26.230000", "content": "Are pre-trained models allowed in this competition?
", "votes": 1, "replies": [ { "id": 269394, "author_name": "Allen Goodman", "author_url": "", "post_date": "2018-01-16T18:39:54.260000", "content": "It really depends on what you mean by “pre-trained.” 😎 Here’s the relevant section of the rules:
\n\n\n\n\nExternal Data Use: The following provision supersedes General Rules\n Section 7.C. below: \"You may use data, other than the Competition\n Data, as allowed on the Competition Website to develop and test your\n models and Submissions; provided, you have the right and authority to\n use such external data for the purposes of the Competition. If you are\n using a source of external data, you must post the source to the\n official external data forum thread no later than one week prior to\n the deadline of the first stage.
\n
If you mean use pre-trained ImageNet (or other publicly available) weights, I don’t see a problem. I would expect most competitors will use pre-trained weights. However, I should note that biological images are less rich than natural images so training from scratch is entirely reasonable (and we often do that to simplify our training architectures). You’re also dealing with one class rather than thousands.
", "votes": 4, "replies": [] }, { "id": 269663, "author_name": "Oleg Trott", "author_url": "", "post_date": "2018-01-17T04:30:07.130000", "content": "\n\n\nAllen Goodman wrote
\n \nIf you mean use pre-trained ImageNet (or other publicly available) weights, I don’t see a problem.
\n
Strictly speaking, ImageNet just contains the web URLs of the images. The images themselves are copyrighted and, in many cases, already gone from the web. It's possible the the copyright owners' intent was to completely prevent their sharing.
\n\nIf using ImagetNet images is not allowed, our right to use models pretrained on them also seems unclear.
", "votes": 1, "replies": [] }, { "id": 269941, "author_name": "Allen Goodman", "author_url": "", "post_date": "2018-01-17T14:52:39.780000", "content": "I was addressing weights trained with ImageNet images.
", "votes": 0, "replies": [] }, { "id": 269970, "author_name": "Oleg Trott", "author_url": "", "post_date": "2018-01-17T15:45:27.957000", "content": "\n\n\nAllen Goodman wrote
\n \n\n\nI was addressing weights trained with ImageNet images.
\n
Allen,
\n\nI understand that (see the last sentence). If competitors use those weights for initialization and then train the models on Kaggle's data before submitting them, they will be submitting models trained, in part, on ImageNet images (by a third party).
", "votes": 0, "replies": [] }, { "id": 288542, "author_name": "Oscar Takeshita", "author_url": "", "post_date": "2018-02-27T18:13:12.523000", "content": "Do all competitors using ImageNet weights need to report in this thread they might be using it for transfer learning or does it suffice to have someone having it posted?
", "votes": 0, "replies": [] } ] }, { "id": 269291, "author_name": "Robert", "author_url": "", "post_date": "2018-01-16T15:46:17.463000", "content": "Is external data allowed in this competition?
", "votes": 1, "replies": [ { "id": 269352, "author_name": "Allen Goodman", "author_url": "", "post_date": "2018-01-16T17:16:13.717000", "content": "Yes!
\n\nIf you do use external data, it’d be awesome to share your sources.
\n\nI’d also be mindful of licensing to ensure the external data can be used for future research.
", "votes": 5, "replies": [] }, { "id": 271061, "author_name": "Dufan Wu", "author_url": "", "post_date": "2018-01-19T16:04:24.270000", "content": "What about manual labeling on external data? E.g. some of the data may come without any label.
", "votes": 4, "replies": [] }, { "id": 282461, "author_name": "Akila Wajirasena", "author_url": "", "post_date": "2018-02-14T02:29:15.933000", "content": "I also have the same question. Most of the external data has the coordinates of the nuclei location but doesn't have an annotated mask. What if I create masks based on those data do I have to share the masks in here?
", "votes": 0, "replies": [] }, { "id": 282920, "author_name": "Allen Goodman", "author_url": "", "post_date": "2018-02-14T19:21:31.053000", "content": "You should share both the images and the annotations you’re using (i.e. both the masks and the object centroids). You should also share the process you used to create the masks. The rules of the competition were written to encourage research around model construction and training practices.
\n\nThe rules also try to make the competition equitable between biologists and non-biologists. This is important since some biologists (e.g. biologists in our lab) have access to terabytes of data (😳) while others only have access to what’s available online.
", "votes": 3, "replies": [] }, { "id": 283066, "author_name": "DimitryTran", "author_url": "", "post_date": "2018-02-14T22:18:54.523000", "content": "Hi Allen, I want to clarify if we \"should\" or \"must\" share the annotations?
\n\nI understand that there is a research motivation behind this project however this is still a competition with winners. Annotation is extremely labor intensive and costly. I think if we must share our annotation no one will do any extra annotation as it would give you no competitive advantage.
\n\nI think it is very reasonable to share the annotation to the freely available images \u001dAFTER the competition, as long as the images themselves were shared here on this thread. Given the small set of data provided for training if the aim is to build something that is ultimately useful annotation is ultimately needed and there need to be an incentive to do that.
", "votes": 5, "replies": [] }, { "id": 283184, "author_name": "Akila Wajirasena", "author_url": "", "post_date": "2018-02-15T02:19:36.817000", "content": "I think both views has some valid points.
\n\nAlso there are some data sets that are partially annotated and if we use some programmatic method to extract only the areas annotated (for example this data set http://www.andrewjanowczyk.com/use-case-1-nuclei-segmentation/ has some areas annotated in the image) should we share that details in this thread?
", "votes": 0, "replies": [] }, { "id": 284665, "author_name": "Adrien Lucas Ecoffet", "author_url": "", "post_date": "2018-02-17T21:44:47.043000", "content": "My rough guess on data you label yourself (speculating here but this is what makes sense to me): if you do label the data manually, then your labels are an external source of data, and you need to share them. If you label automatically, then your labeling is part of your algorithm and you don't need to share it.
\n\nNote that the rules say \"If you are using a source of external data, you must post the source to the official external data forum thread no later than one week prior to the deadline of the first stage.\" (emphasis mine), so if you really want to keep your competitive advantage you could withhold your data up until exactly one week before the end of the competition.
", "votes": 1, "replies": [] }, { "id": 285489, "author_name": "Sachin Naik", "author_url": "", "post_date": "2018-02-20T12:43:10.973000", "content": "Automatic, semi-automatic or manual annotation requires time-effort. Sharing afterward the competition makes sense. Don’t agree with external data source portion though.
", "votes": 0, "replies": [] }, { "id": 285740, "author_name": "Brasnold", "author_url": "", "post_date": "2018-02-20T23:17:42.190000", "content": "\n\n\nAllen Goodman wrote
\n \nThis is important since some biologists (e.g. biologists in our lab) have access to terabytes of data (😳)
\n
That's how you've reached 0.64 LB? :)\nSeriously, I'm asking you this to know if we can beat your result or it should be considered a proxy for the bayesian error.
", "votes": 0, "replies": [] }, { "id": 285996, "author_name": "Anne Carpenter", "author_url": "", "post_date": "2018-02-21T12:57:49.297000", "content": "\n\n\nAutomatic, semi-automatic or manual annotation requires time-effort. Sharing afterward the competition makes sense. Don’t agree with external data source portion though.
\n
Then maybe you can organize your own competition someday and make the rules how you like them ;)
\n\nBut seriously, here is the rationale: though manual annotation will definitely result in better models, we actually wanted to discourage competitors from winning that way, because it would disadvantage those who don't have the time or money to do annotations. Hopefully, people can focus on their methodology for the competition itself, and in the future more annotations can be made and then the model will just get better and better.
", "votes": 1, "replies": [] }, { "id": 292488, "author_name": "CPMP", "author_url": "", "post_date": "2018-03-08T02:44:00.193000", "content": "Well, I asked about the need for sharing masks when using external data, and was answered that no, people don't have to. It means that people with biology knowledge have an unfair advantage as they can annotate more data for training.
", "votes": 0, "replies": [] }, { "id": 293412, "author_name": "Anne Carpenter", "author_url": "", "post_date": "2018-03-09T19:37:22.330000", "content": "Biologists would certainly have an advantage... then again, being forced to circle cells for hours is not my idea of fun! On the other hand, people who have deep learning knowledge also have an advantage in this challenge so I suppose it is all \"fair\".
", "votes": 1, "replies": [] }, { "id": 293423, "author_name": "Oscar Takeshita", "author_url": "", "post_date": "2018-03-09T20:04:49.703000", "content": "Perhaps it is an encouragement for the formation of strong teams with a wide scope of expertise.
", "votes": 0, "replies": [] } ] }, { "id": 309363, "author_name": "Ruobing", "author_url": "", "post_date": "2018-04-05T06:17:45.900000", "content": "Pretrained model using the Coco dataset (https://github.com/matterport/Mask_RCNN/releases)\nFixed training dataset (https://github.com/lopuhin/kaggle-dsbowl-2018-dataset-fixes)\nTNBC_NucleiSegmentation (https://cs.adelaide.edu.au/~carneiro/isbi14_challenge/dataset.html)\nPSB Crowdsourced Nuclei Annotation (https://becklab.hms.harvard.edu/software/psb-crowdsourced-nuclei-annotation-data-1)\nhttps://www.kaggle.com/voglinio/external-h-e-data-with-mask-annotations (made by Watt, converted from https://nucleisegmentationbenchmark.weebly.com/)
", "votes": 2, "replies": [] }, { "id": 298355, "author_name": "Ashish Lal", "author_url": "", "post_date": "2018-03-19T12:42:28.387000", "content": "https://github.com/lopuhin/kaggle-dsbowl-2018-dataset-fixes
", "votes": 0, "replies": [ { "id": 301849, "author_name": "Ashish Lal", "author_url": "", "post_date": "2018-03-23T11:17:28.480000", "content": "https://www.kaggle.com/voglinio/external-h-e-data-with-mask-annotations
", "votes": 3, "replies": [] }, { "id": 308926, "author_name": "Ashish Lal", "author_url": "", "post_date": "2018-04-04T11:33:22.427000", "content": "https://becklab.hms.harvard.edu/software/psb-crowdsourced-nuclei-annotation-data-1
", "votes": 0, "replies": [] }, { "id": 308929, "author_name": "Ashish Lal", "author_url": "", "post_date": "2018-04-04T11:35:02.633000", "content": ".https://nucleisegmentationbenchmark.weebly.com/
", "votes": 0, "replies": [] } ] }, { "id": 276790, "author_name": "davidkijlstra", "author_url": "", "post_date": "2018-02-01T14:07:30.350000", "content": "If we change the images (cropping, rotating, flipping etc) from the training set to train our model, do we have to post that data here as well?
", "votes": 2, "replies": [ { "id": 276814, "author_name": "Will Cukierski", "author_url": "", "post_date": "2018-02-01T15:12:01.293000", "content": "No. We consider that to be part of your methodology, not external data.
", "votes": 3, "replies": [] } ] }, { "id": 273344, "author_name": "Morteza Milani", "author_url": "", "post_date": "2018-01-24T13:33:45.990000", "content": "https://ome.grc.nia.nih.gov/iicbu2008/
", "votes": 2, "replies": [] }, { "id": 269625, "author_name": "pooh", "author_url": "", "post_date": "2018-01-17T03:17:44.487000", "content": "In the AMIDA13 MICCAI grand competition, they achieved the 3rd highest F1-score.
\n\nCell Detection with Deep Convolutional Neural Network and Compressed Sensing.
\n\n", "votes": 2, "replies": [ { "id": 271150, "author_name": "Josan", "author_url": "", "post_date": "2018-01-19T19:29:11.130000", "content": "Is the dataset downloadable?
", "votes": 0, "replies": [] }, { "id": 276226, "author_name": "Wesley Goi", "author_url": "", "post_date": "2018-01-31T02:09:39.243000", "content": "The competition has ended, it seems you'll have to register on their site to get the data \nhttp://amida13.isi.uu.nl/?q=node/90
\n\nhowever, it seems its continuing in a separate data challenge: http://tupac.tue-image.nl. Tried to sign up but still waiting for the admins to approve.
\n\nmeanwhile i saw (http://tupac.tue-image.nl/node/62 ) some of the winning teams used some external datasets:
\n\nHope this helps everyone!
", "votes": 4, "replies": [] } ] }, { "id": 309275, "author_name": "Hiromichi NOMATA", "author_url": "", "post_date": "2018-04-04T23:47:17.123000", "content": "http://pretrained.ml\nhttps://gist.github.com/flyyufelix/7e2eafb149f72f4d38dd661882c554a6\nhttps://github.com/Cadene/pretrained-models.pytorch\nhttps://github.com/Zehaos/MobileNet\nhttps://github.com/aaron-xichen/pytorch-playground\nhttps://github.com/endymecy/awesome-deeplearning-resources\nhttps://github.com/fchollet/deep-learning-models\nhttps://github.com/fchollet/keras/blob/master/keras/applications/xception.py\nhttps://github.com/iro-cp/FCRN-DepthPrediction\nhttps://github.com/kentsommer/keras-inceptionV4\nhttps://github.com/mrharicot/monodepth\nhttps://github.com/myutwo150/keras-inception-resnet-v2\nhttps://github.com/nutszebra/resnext\nhttps://github.com/openimages/dataset\nhttps://github.com/raulmur/ORB_SLAM2\nhttps://github.com/shicai/SENet-Caffe\nhttps://github.com/taehoonlee/tensornets\nhttps://github.com/tensorflow/models/blob/master/slim/README.md\nhttps://github.com/titu1994/MobileNetworks\nhttps://keras.io/applications/
\n\nhttps://github.com/naoto0804/cross-domain-detection\nhttps://github.com/k3nt0w/FCN_via_keras\nhttps://github.com/imatge-upc/rsis
\n\nhttps://github.com/pjreddie/darknet\nhttps://github.com/explosion/lightnet\nhttps://gitlab.com/EAVISE/lightnet\nhttps://github.com/yjmade/ios_camera_object_detection\nhttps://github.com/mrzl/ofxDarknet\nhttps://github.com/moovel/node-yolo
\n\nhttps://www.tensorflow.org/hub/\nhttps://github.com/youansheng/pytorch-cv\nhttps://github.com/chainer/chainercv\nhttps://tryolabs.com/blog/2018/01/18/faster-r-cnn-down-the-rabbit-hole-of-modern-object-detection/\nhttps://github.com/facebookresearch/Detectron\nhttps://github.com/DeNA/Chainer_Mask_R-CNN\nhttps://adamspannbauer.github.io/2017/12/28/super-smash-cv/\nhttps://github.com/sfzhang15/RefineDet\nhttps://github.com/kuangliu/pytorch-retinanet\nhttps://github.com/fizyr/keras-retinanet\nhttps://github.com/szq0214/DSOD\nhttps://github.com/mfigurnov/sact\nhttps://github.com/allanzelener/YAD2K\nhttps://github.com/daijifeng001/R-FCN\nhttps://github.com/weichengkuo/DeepBox\nhttps://github.com/endernewton/tf-faster-rcnn\nhttp://vision.cs.utexas.edu/projects/pixelobjectness/
\n\nhttps://github.com/ternaus/robot-surgery-segmentation\nhttps://github.com/knorth55/chainer-fcis\nhttps://github.com/matterport/Mask_RCNN\nhttps://github.com/fyu/drn\nhttps://github.com/mitmul/chainer-pspnet\nhttps://github.com/TobyPDE/FRRN\nhttps://github.com/msracver/FCIS\nhttps://blog.deepsense.ai/deep-learning-for-satellite-imagery-via-image-segmentation/\nhttps://github.com/isht7/pytorch-deeplab-resnet\nhttps://github.com/kmaninis/OSVOS-caffe\nhttps://github.com/hszhao/PSPNet\nhttps://github.com/HyeonwooNoh/DeconvNet\nhttps://github.com/fregu856/segmentation\nhttps://github.com/e-lab/ENet-training\nhttp://www.mathgram.xyz/entry/keras/fcn\nhttp://stanford.edu/~rqi/pointnet/\nhttps://github.com/Engineering-Course/LIP_SSL
\n\nand all models listed on this page
", "votes": -2, "replies": [ { "id": 309312, "author_name": "Ildoo Kim", "author_url": "", "post_date": "2018-04-05T02:47:22.240000", "content": "These are not related to 'external data'.
", "votes": 0, "replies": [] }, { "id": 309323, "author_name": "", "author_url": "", "post_date": "2018-04-05T03:29:44.633000", "content": "", "votes": 0, "replies": [] } ] }, { "id": 278534, "author_name": "AviWolfson", "author_url": "", "post_date": "2018-02-06T12:09:48.537000", "content": "", "votes": -2, "replies": [] }, { "id": 270422, "author_name": "brad222", "author_url": "", "post_date": "2018-01-18T07:41:08.907000", "content": "23423
", "votes": -7, "replies": [] }, { "id": 411468, "author_name": "leon", "author_url": "", "post_date": "2018-10-28T08:30:13.760000", "content": "amazing！
", "votes": 0, "replies": [] }, { "id": 313436, "author_name": "", "author_url": "", "post_date": "2018-04-13T08:55:55.750000", "content": "", "votes": 0, "replies": [] }, { "id": 313203, "author_name": "", "author_url": "", "post_date": "2018-04-12T23:05:26.027000", "content": "", "votes": 0, "replies": [] }, { "id": 312773, "author_name": "", "author_url": "", "post_date": "2018-04-12T10:02:00.023000", "content": "", "votes": 0, "replies": [] }, { "id": 312502, "author_name": "", "author_url": "", "post_date": "2018-04-11T21:39:58.683000", "content": "", "votes": 0, "replies": [] }, { "id": 312241, "author_name": "", "author_url": "", "post_date": "2018-04-11T12:53:59.447000", "content": "", "votes": 0, "replies": [] }, { "id": 312238, "author_name": "", "author_url": "", "post_date": "2018-04-11T12:47:26.730000", "content": "", "votes": 0, "replies": [] }, { "id": 312100, "author_name": "", "author_url": "", "post_date": "2018-04-11T07:28:10.783000", "content": "", "votes": 0, "replies": [] }, { "id": 312066, "author_name": "", "author_url": "", "post_date": "2018-04-11T05:59:22.360000", "content": "", "votes": 0, "replies": [] }, { "id": 311858, "author_name": "", "author_url": "", "post_date": "2018-04-10T19:46:36.530000", "content": "", "votes": 0, "replies": [] }, { "id": 311412, "author_name": "", "author_url": "", "post_date": "2018-04-10T02:19:16.060000", "content": "", "votes": 0, "replies": [] }, { "id": 311194, "author_name": "", "author_url": "", "post_date": "2018-04-09T15:27:20.443000", "content": "", "votes": 0, "replies": [ { "id": 311198, "author_name": "", "author_url": "", "post_date": "2018-04-09T15:50:20.647000", "content": "", "votes": 0, "replies": [] } ] }, { "id": 311193, "author_name": "", "author_url": "", "post_date": "2018-04-09T15:25:55.620000", "content": "", "votes": 0, "replies": [ { "id": 311199, "author_name": "", "author_url": "", "post_date": "2018-04-09T15:50:57.233000", "content": "", "votes": 0, "replies": [] } ] }, { "id": 310781, "author_name": "", "author_url": "", "post_date": "2018-04-08T15:15:04.700000", "content": "", "votes": 0, "replies": [] }, { "id": 310744, "author_name": "", "author_url": "", "post_date": "2018-04-08T13:18:56.753000", "content": "", "votes": 0, "replies": [] }, { "id": 310701, "author_name": "", "author_url": "", "post_date": "2018-04-08T10:46:46.070000", "content": "", "votes": 0, "replies": [] }, { "id": 310549, "author_name": "", "author_url": "", "post_date": "2018-04-07T23:19:57.220000", "content": "", "votes": 0, "replies": [] }, { "id": 310501, "author_name": "", "author_url": "", "post_date": "2018-04-07T18:25:47.367000", "content": "", "votes": 0, "replies": [] }, { "id": 309723, "author_name": "", "author_url": "", "post_date": "2018-04-05T21:24:20.810000", "content": "", "votes": 0, "replies": [] }, { "id": 309653, "author_name": "", "author_url": "", "post_date": "2018-04-05T18:43:13.517000", "content": "", "votes": 0, "replies": [] }, { "id": 308983, "author_name": "", "author_url": "", "post_date": "2018-04-04T13:11:22.047000", "content": "", "votes": 0, "replies": [ { "id": 313913, "author_name": "", "author_url": "", "post_date": "2018-04-14T04:12:05.083000", "content": "", "votes": 0, "replies": [] } ] }, { "id": 308964, "author_name": "", "author_url": "", "post_date": "2018-04-04T12:44:38.007000", "content": "", "votes": 0, "replies": [] }, { "id": 308889, "author_name": "", "author_url": "", "post_date": "2018-04-04T09:02:19.087000", "content": "", "votes": 0, "replies": [] }, { "id": 308776, "author_name": "", "author_url": "", "post_date": "2018-04-04T04:27:35.090000", "content": "", "votes": 0, "replies": [] }, { "id": 308774, "author_name": "", "author_url": "", "post_date": "2018-04-04T04:20:18.373000", "content": "", "votes": 0, "replies": [] }, { "id": 308773, "author_name": "", "author_url": "", "post_date": "2018-04-04T04:19:46.607000", "content": "", "votes": 0, "replies": [] }, { "id": 308771, "author_name": "", "author_url": "", "post_date": "2018-04-04T04:18:59.793000", "content": "", "votes": 0, "replies": [] }, { "id": 308738, "author_name": "", "author_url": "", "post_date": "2018-04-04T02:04:49.227000", "content": "", "votes": 0, "replies": [] }, { "id": 308707, "author_name": "", "author_url": "", "post_date": "2018-04-04T00:45:59.357000", "content": "", "votes": 0, "replies": [] }, { "id": 308673, "author_name": "", "author_url": "", "post_date": "2018-04-03T22:32:55.070000", "content": "", "votes": 0, "replies": [] }, { "id": 308634, "author_name": "", "author_url": "", "post_date": "2018-04-03T20:30:52.193000", "content": "", "votes": 0, "replies": [] }, { "id": 308572, "author_name": "", "author_url": "", "post_date": "2018-04-03T17:35:50.443000", "content": "", "votes": 0, "replies": [] }, { "id": 308464, "author_name": "", "author_url": "", "post_date": "2018-04-03T14:48:56.480000", "content": "", "votes": 0, "replies": [] }, { "id": 308426, "author_name": "", "author_url": "", "post_date": "2018-04-03T13:49:52.427000", "content": "", "votes": 0, "replies": [] }, { "id": 308286, "author_name": "", "author_url": "", "post_date": "2018-04-03T09:19:57.170000", "content": "", "votes": 0, "replies": [] }, { "id": 308204, "author_name": "", "author_url": "", "post_date": "2018-04-03T05:51:52.120000", "content": "", "votes": 0, "replies": [] }, { "id": 308159, "author_name": "", "author_url": "", "post_date": "2018-04-03T03:47:34.223000", "content": "", "votes": 0, "replies": [] }, { "id": 308046, "author_name": "", "author_url": "", "post_date": "2018-04-02T21:31:21.777000", "content": "", "votes": 0, "replies": [] }, { "id": 307999, "author_name": "", "author_url": "", "post_date": "2018-04-02T19:27:59.227000", "content": "", "votes": 0, "replies": [ { "id": 308660, "author_name": "", "author_url": "", "post_date": "2018-04-03T21:48:29.643000", "content": "", "votes": 0, "replies": [] } ] }, { "id": 307499, "author_name": "", "author_url": "", "post_date": "2018-04-01T20:35:44.947000", "content": "", "votes": 0, "replies": [] }, { "id": 307488, "author_name": "", "author_url": "", "post_date": "2018-04-01T19:26:01.387000", "content": "", "votes": 0, "replies": [] }, { "id": 307417, "author_name": "", "author_url": "", "post_date": "2018-04-01T15:13:51.013000", "content": "", "votes": 0, "replies": [] }, { "id": 307275, "author_name": "", "author_url": "", "post_date": "2018-04-01T09:09:31.907000", "content": "", "votes": 0, "replies": [] }, { "id": 306931, "author_name": "", "author_url": "", "post_date": "2018-03-31T11:00:54.413000", "content": "", "votes": 0, "replies": [] }, { "id": 306863, "author_name": "", "author_url": "", "post_date": "2018-03-31T07:29:55.263000", "content": "", "votes": 0, "replies": [] }, { "id": 306847, "author_name": "", "author_url": "", "post_date": "2018-03-31T06:29:43.633000", "content": "", "votes": 0, "replies": [ { "id": 308157, "author_name": "", "author_url": "", "post_date": "2018-04-03T03:40:49.993000", "content": "", "votes": 0, "replies": [] } ] }, { "id": 306830, "author_name": "", "author_url": "", "post_date": "2018-03-31T05:15:10.087000", "content": "", "votes": 0, "replies": [] }, { "id": 306192, "author_name": "", "author_url": "", "post_date": "2018-03-30T02:37:59.770000", "content": "", "votes": 0, "replies": [] }, { "id": 304897, "author_name": "", "author_url": "", "post_date": "2018-03-28T07:20:36.867000", "content": "", "votes": 0, "replies": [] }, { "id": 304302, "author_name": "", "author_url": "", "post_date": "2018-03-27T10:28:07.730000", "content": "", "votes": 0, "replies": [] }, { "id": 303714, "author_name": "", "author_url": "", "post_date": "2018-03-26T15:22:15.730000", "content": "", "votes": 0, "replies": [] }, { "id": 303452, "author_name": "", "author_url": "", "post_date": "2018-03-26T07:59:47.043000", "content": "", "votes": 0, "replies": [] }, { "id": 303227, "author_name": "", "author_url": "", "post_date": "2018-03-25T20:01:49.400000", "content": "", "votes": 0, "replies": [] }, { "id": 301461, "author_name": "", "author_url": "", "post_date": "2018-03-22T19:34:19.173000", "content": "", "votes": 0, "replies": [] }, { "id": 293522, "author_name": "", "author_url": "", "post_date": "2018-03-10T03:31:38.247000", "content": "", "votes": 0, "replies": [] }, { "id": 292546, "author_name": "", "author_url": "", "post_date": "2018-03-08T06:12:40.377000", "content": "", "votes": 0, "replies": [ { "id": 293409, "author_name": "", "author_url": "", "post_date": "2018-03-09T19:34:17.383000", "content": "", "votes": 1, "replies": [] } ] }, { "id": 292005, "author_name": "", "author_url": "", "post_date": "2018-03-07T07:57:54.270000", "content": "", "votes": 0, "replies": [ { "id": 294058, "author_name": "", "author_url": "", "post_date": "2018-03-11T07:19:49.030000", "content": "", "votes": 0, "replies": [] }, { "id": 294676, "author_name": "", "author_url": "", "post_date": "2018-03-12T12:03:06.950000", "content": "", "votes": 0, "replies": [] }, { "id": 295129, "author_name": "", "author_url": "", "post_date": "2018-03-13T05:12:18.343000", "content": "", "votes": 0, "replies": [] } ] }, { "id": 290099, "author_name": "", "author_url": "", "post_date": "2018-03-02T23:29:51.303000", "content": "", "votes": 0, "replies": [] }, { "id": 289215, "author_name": "", "author_url": "", "post_date": "2018-03-01T06:24:50.940000", "content": "", "votes": 0, "replies": [] }, { "id": 281586, "author_name": "", "author_url": "", "post_date": "2018-02-12T21:12:53.807000", "content": "", "votes": 0, "replies": [ { "id": 282731, "author_name": "", "author_url": "", "post_date": "2018-02-14T13:32:03.137000", "content": "", "votes": 1, "replies": [] }, { "id": 285729, "author_name": "", "author_url": "", "post_date": "2018-02-20T22:41:10.147000", "content": "", "votes": 0, "replies": [] }, { "id": 291534, "author_name": "", "author_url": "", "post_date": "2018-03-06T10:02:56.763000", "content": "", "votes": 0, "replies": [] } ] }, { "id": 280402, "author_name": "", "author_url": "", "post_date": "2018-02-09T22:27:03.320000", "content": "", "votes": 0, "replies": [] }, { "id": 280159, "author_name": "", "author_url": "", "post_date": "2018-02-09T13:44:51.420000", "content": "", "votes": 0, "replies": [] }, { "id": 279666, "author_name": "", "author_url": "", "post_date": "2018-02-08T13:56:50.620000", "content": "", "votes": 0, "replies": [] }, { "id": 277409, "author_name": "", "author_url": "", "post_date": "2018-02-03T02:12:59.880000", "content": "", "votes": 0, "replies": [] }, { "id": 309640, "author_name": "", "author_url": "", "post_date": "2018-04-05T17:59:44.577000", "content": "", "votes": 0, "replies": [] }, { "id": 308512, "author_name": "", "author_url": "", "post_date": "2018-04-03T16:03:11.370000", "content": "", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "269233": "Post your (freely available) external data links here.", "269406": "Some potential resources with related datasets :\n\n**A comprehensive tutorial with selected use cases**\nhttp://www.andrewjanowczyk.com/use-case-1-nuclei-segmentation/ (dataset : http://andrewjanowczyk.com/wp-static/nuclei.tgz)\n\n**A Dataset and a Technique for Generalized Nuclear Segmentation in Histological Images**\nhttps://nucleisegmentationbenchmark.weebly.com/\n\n**2D images of 3T3 and U20S cells hand-segmented to show nuclear boundaries**\nhttp://murphylab.web.cmu.edu/data/#2DNuclei (dataset : http://murphylab.web.cmu.edu/data/2009_ISBI_Nuclei.html)\n\n**Digital Pathology : Nuclei Segmentation MICCAI 2017 Challenge**\nhttp://miccai.cloudapp.net/competitions/57 (currently not responding)\n\n**Mitochondria segmentation dataset**\nhttps://cvlab.epfl.ch/data/em\n", "288797": "I'll use fixed stage1_train masks https://github.com/lopuhin/kaggle-dsbowl-2018-dataset-fixes", "275397": "Think this will be really useful. \n[Link to dataset][1]\n\nBasically, the cells (brightfield) were first unstained and later stained (nuceli and cytoplasmic) using fluorescence microscopy. \n\nIt'll be both a training set for brightfield and flourenscent images. Especially useful for helping to locate the nuclei for brightfield samples in the dataset. \n\n![Data, training and testing pipelines for cell segmentation. ][2]\n\n**Abstract**\n\nDeep Convolutional Neural Networks (DCNN) have recently emerged as superior for many image segmentation tasks. The DCNN performance is however heavily dependent on the availability of large amounts of problem-specific training samples. Here we show that DCNNs trained on ground truth created automatically using fluorescently labeled cells, perform similar to manual annotations.\n\nReference\n\nSadanandan, SK, Ranefall, P, Le Guyader, S, Wählby, C (2017). Automated Training of Deep Convolutional Neural Networks for Cell Segmentation. Sci Rep, 7, 1:7860.\n\n\n [1]: http://cb.uu.se/~carolina/timelapse_data/\n [2]: https://media.springernature.com/lw900/springer-static/image/art%3A10.1038%2Fs41598-017-07599-6/MediaObjects/41598_2017_7599_Fig1_HTML.jpg", "269253": "__Adenocarcinomas with labelled nuclei dataset__\nhttps://warwick.ac.uk/fac/sci/dcs/research/tia/data/crchistolabelednucleihe/crchistophenotypes_2016_04_28.zip\n\nSource: K. Sirinukunwattana, S.E.A. Raza, Y.W Tsang, I.A. Cree, D.R.J. Snead, N.M. Rajpoot, ‘Locality Sensitive Deep Learning for Detection and Classification of Nuclei in Routine Colon Cancer Histology Images,’ IEEE Transactions on Medical Imaging, 2016\n \n__Other Relevant Papers__:\n\n1. Detection and Segmentation of Cell Nuclei in Virtual Microscopy Images: A Minimum-Model Approach https://www.nature.com/articles/srep00503", "272203": "virtual fluorescence microscopy:\n\nhttp://cbia.fi.muni.cz/projects/simulator.html\n\nhttp://cbia.fi.muni.cz/projects/cytopacq-a-simulation-toolbox_5.html\n\nhttp://cbia.fi.muni.cz/simulator/index.php", "271360": "https://cs.adelaide.edu.au/~carneiro/isbi14_challenge/dataset.html\n\nI find some train_stage1 images are from the cervical cytology images of isbi 14 challenge. I suspect they may be use in test_stage2. If they are not, I can all for isbi14 challenge images for training.\n\nSo I can I use these images?", "312659": "Stage 1 test data masks (generated from the csv file): https://github.com/yuanqing811/DSB2018_stage1_test", "284672": "Dear Organisers,\n\nI need an official word from Anne, Kyle or Allen about whether or not user-generated annotations to images already shared here on the thread are to be posted one week before the deadline.\n\nMy strategy is to use my existing model to pre-label the images and then manually touch up the missing masks or re-fine poor masks. As such my annotations are semi-automatic and are part of my methodology. \n\nI will likely abandon this strategy if I must share these masks as they are partly generated by my model. \n\nIf you could kindly respond in a timely manner that would be much appreciated.\n\nThanks.", "285466": "https://downloads.openmicroscopy.org/images/\n\nhttps://www.openmicroscopy.org/\n\nhttps://warwick.ac.uk/fac/sci/dcs/research/tia/data/", "271537": "Dataset at: \n\nhttps://github.com/sanuj/nuclei-net\n\nhttps://sanuj.github.io/nuclei-net-btp.html\n\nhttps://nucleisegmentationbenchmark.weebly.com/dataset.html\n\n---\n\nhttps://sites.google.com/site/aacruzr/image-datasets, see (Histology and Histopathology, etc)\n\nhttp://histo.anat.ubc.ca/HISTOLOGY/Basic%20Tissues/view.apml?\n\n----\nhuman altas\n\nhttps://www.proteinatlas.org/humanproteome\n", "282734": "If we change the train set (correct mistakes in the masks), do we have to post that data here as well?", "271270": "Broad Bioimage Benchmark Collection https://data.broadinstitute.org/bbbc/BBBC006/", "312803": "As I know I'm definitely not gonna win now, I make my revision of the training set and my annotation tool available at https://github.com/ibmua/data-science-bowl-2018-train-set . I think the training set is probably the best there is, as I went through each white-on-black sample and fixed it by hand. Annotation tool is also pretty useful, though, probably limitedly so now after the first stage is over. Maybe, for some future competitions, I donno.", "312111": "(https://cs.adelaide.edu.au/~carneiro/isbi14_challenge/dataset.html) PSB Crowdsourced Nuclei Annotation (https://becklab.hms.harvard.edu/software/psb-crowdsourced-nuclei-annotation-data-1) https://www.kaggle.com/voglinio/external-h-e-data-with-mask-annotations (made by Watt, converted from https://nucleisegmentationbenchmark.weebly.com/)\nhttp://www.andrewjanowczyk.com/use-case-1-nuclei-segmentation/\nhttps://bioimage.ucsb.edu/research/bio-segmentation", "308586": "We used these images (sources in the 'sources.txt' file)\nhttps://drive.google.com/file/d/1A_x-40lfh9FaBEsRv2zTjjvNOlcItnTX/view?usp=sharing", "308476": "https://zenodo.org/record/1175282\nhttp://cbia.fi.muni.cz/projects/cytopacq-a-simulation-toolbox_6.html", "307796": "Our team will probably use some pretrained models from the following list:\n\nhttps://github.com/Cadene/pretrained-models.pytorch\nhttps://keras.io/applications/\nhttps://github.com/fizyr/keras-models/releases/tag/v0.0.1\nhttps://mxnet.incubator.apache.org/api/python/gluon/model_zoo.html\n\nAdditionally we use some external datasets mentioned in this thread.", "307460": "PSB Crowdsourced Nuclei Annotation\n\nhttps://becklab.hms.harvard.edu/software/psb-crowdsourced-nuclei-annotation-data-1\n", "269385": "Are pre-trained models allowed in this competition?", "269291": "Is external data allowed in this competition?", "309363": "Pretrained model using the Coco dataset (https://github.com/matterport/Mask_RCNN/releases)\nFixed training dataset (https://github.com/lopuhin/kaggle-dsbowl-2018-dataset-fixes)\nTNBC_NucleiSegmentation (https://cs.adelaide.edu.au/~carneiro/isbi14_challenge/dataset.html)\nPSB Crowdsourced Nuclei Annotation (https://becklab.hms.harvard.edu/software/psb-crowdsourced-nuclei-annotation-data-1)\nhttps://www.kaggle.com/voglinio/external-h-e-data-with-mask-annotations (made by Watt, converted from https://nucleisegmentationbenchmark.weebly.com/)\n\n", "298355": " https://github.com/lopuhin/kaggle-dsbowl-2018-dataset-fixes", "276790": "If we change the images (cropping, rotating, flipping etc) from the training set to train our model, do we have to post that data here as well?", "273344": "https://ome.grc.nia.nih.gov/iicbu2008/", "269625": " In the AMIDA13 MICCAI grand competition, they achieved the **3rd highest** F1-score. \n\n**Cell Detection with Deep Convolutional Neural Network and Compressed Sensing**.\n\n[link to the arXiv paper][1]\n\n\n [1]: https://arxiv.org/pdf/1708.03307.pdf", "309275": "http://pretrained.ml\nhttps://gist.github.com/flyyufelix/7e2eafb149f72f4d38dd661882c554a6\nhttps://github.com/Cadene/pretrained-models.pytorch\nhttps://github.com/Zehaos/MobileNet\nhttps://github.com/aaron-xichen/pytorch-playground\nhttps://github.com/endymecy/awesome-deeplearning-resources\nhttps://github.com/fchollet/deep-learning-models\nhttps://github.com/fchollet/keras/blob/master/keras/applications/xception.py\nhttps://github.com/iro-cp/FCRN-DepthPrediction\nhttps://github.com/kentsommer/keras-inceptionV4\nhttps://github.com/mrharicot/monodepth\nhttps://github.com/myutwo150/keras-inception-resnet-v2\nhttps://github.com/nutszebra/resnext\nhttps://github.com/openimages/dataset\nhttps://github.com/raulmur/ORB_SLAM2\nhttps://github.com/shicai/SENet-Caffe\nhttps://github.com/taehoonlee/tensornets\nhttps://github.com/tensorflow/models/blob/master/slim/README.md\nhttps://github.com/titu1994/MobileNetworks\nhttps://keras.io/applications/\nhttps://softwaremill.com/counting-objects-with-faster-rcnn/\n---\nhttps://github.com/naoto0804/cross-domain-detection\nhttps://github.com/k3nt0w/FCN_via_keras\nhttps://github.com/imatge-upc/rsis\nhttps://skrish13.github.io/articles/2018-03/fair-cv-saga\n---\nhttps://github.com/pjreddie/darknet\nhttps://github.com/explosion/lightnet\nhttps://gitlab.com/EAVISE/lightnet\nhttps://github.com/yjmade/ios_camera_object_detection\nhttps://github.com/mrzl/ofxDarknet\nhttps://github.com/moovel/node-yolo\nhttps://github.com/leetenki/YOLOv2\n---\nhttps://www.tensorflow.org/hub/\nhttps://github.com/youansheng/pytorch-cv\nhttps://github.com/chainer/chainercv\nhttps://tryolabs.com/blog/2018/01/18/faster-r-cnn-down-the-rabbit-hole-of-modern-object-detection/\nhttps://github.com/facebookresearch/Detectron\nhttps://github.com/DeNA/Chainer_Mask_R-CNN\nhttps://adamspannbauer.github.io/2017/12/28/super-smash-cv/\nhttps://github.com/sfzhang15/RefineDet\nhttps://github.com/kuangliu/pytorch-retinanet\nhttps://github.com/fizyr/keras-retinanet\nhttps://github.com/szq0214/DSOD\nhttps://github.com/mfigurnov/sact\nhttps://github.com/allanzelener/YAD2K\nhttps://github.com/daijifeng001/R-FCN\nhttps://github.com/weichengkuo/DeepBox\nhttps://github.com/endernewton/tf-faster-rcnn\nhttp://vision.cs.utexas.edu/projects/pixelobjectness/\nhttps://imatge-upc.github.io/detection-2016-nipsws/\n---\nhttps://github.com/ternaus/robot-surgery-segmentation\nhttps://github.com/knorth55/chainer-fcis\nhttps://github.com/matterport/Mask_RCNN\nhttps://github.com/fyu/drn\nhttps://github.com/mitmul/chainer-pspnet\nhttps://github.com/TobyPDE/FRRN\nhttps://github.com/msracver/FCIS\nhttps://blog.deepsense.ai/deep-learning-for-satellite-imagery-via-image-segmentation/\nhttps://github.com/isht7/pytorch-deeplab-resnet\nhttps://github.com/kmaninis/OSVOS-caffe\nhttps://github.com/hszhao/PSPNet\nhttps://github.com/HyeonwooNoh/DeconvNet\nhttps://github.com/fregu856/segmentation\nhttps://github.com/e-lab/ENet-training\nhttp://www.mathgram.xyz/entry/keras/fcn\nhttp://stanford.edu/~rqi/pointnet/\nhttps://github.com/Engineering-Course/LIP_SSL\n\nand all models listed on this page\n", "278534": "", "270422": "23423", "411468": "amazing！", "313436": "We use the following package for data augmentation: \nhttps://github.com/NVIDIA/FastPhotoStyle/blob/master/USAGE.md\nJust realized that this package contains pre-trained model weights (https://github.com/suquark/PhotoWCTModels/blob/d8a4e35f2a994eb7c6f261b5e8172ae6397040f2/photo_wct.pth) that needs to be posted here. ", "313203": "I make my version of BBBC6 dataset turned into DSB style is available at https://drive.google.com/file/d/16KVLYeKBkk_Y0W4Fra_2AhjmW-p6JjZu/view?usp=sharing . Please, view images before use to understand how the data is combined in channels.", "312773": "Now, knowing what stage 2 data is like, I'm guessing, pretty much all datasets included here except BBBC ones are only gonna hurt performance on stage 2 data. Trying to fit them has only taken away effort from trying to fit the actual data for me. Stage 2 data doesn't even contain any gray black-on-white images. The gray-only solution I made in the first few days is actually 10x better for this data than the one I couldn't finish on time that was supposed to work on any colors and lighting conditions.", "312502": "https://nucleisegmentationbenchmark.weebly.com/", "312241": "Based on Mask RCNN with pre-trained model on COCO dataset from https://github.com/matterport/Mask_RCNN/releases", "312238": "I konw it's a bit late, we will use the external Data as follows:\n(1) TNBC_NucleiSegmentation: https://zenodo.org/record/1175282#.Ws2n_vkdhfA\n(2) https://github.com/lopuhin/kaggle-dsbowl-2018-dataset-fixes\n(3) pretrained on https://github.com/matterport/Mask_RCNN/releases", "312100": "Using pre-trained model from the MS Coco dataset, as part of the Matterport Mask RCNN implementation: reference https://github.com/matterport/Mask_RCNN/releases", "312066": "https://github.com/matterport/Mask_RCNN i use pretrained model on coco and imagenet", "311858": "https://github.com/lopuhin/kaggle-dsbowl-2018-dataset-fixes\nhttps://www.kaggle.com/voglinio/external-h-e-data-with-mask-annotations", "311412": "We used these images: https://www.kaggle.com/voglinio/external-h-e-data-with-mask-annotations", "311194": "Hasn't the external data deadline passed already?", "311193": "Using an additional data from our own microscope. It is anonimized and comes from few projects. Unfortunately only fluorescent images.\n\nhttps://www.dropbox.com/sh/i753b966e6zp7t4/AAB6KtMCgcmKzyfdxEzXVIyHa?dl=0", "310781": "Using pre-trained model from the MS Coco dataset, as part of the Matterport Mask RCNN implementation: reference https://github.com/matterport/Mask_RCNN/releases", "310744": "Deeplab v3, v3+ pretrained weights : https://github.com/tensorflow/models/tree/master/research/deeplab", "310701": "We used https://www2.eecs.berkeley.edu/Research/Projects/CS/vision/grouping/resources.html for synthetic images generation.\nhttps://zenodo.org/record/1174353#.WprukR9fg8o\nhttps://github.com/lopuhin/kaggle-dsbowl-2018-dataset-fixes", "310549": "https://data.broadinstitute.org/bbbc/BBBC006/, and models posted here ", "310501": "I have not seen these models linked yet, and I may be using them. \n\nhttps://github.com/tensorflow/models/blob/master/research/object_detection/g3doc/detection_model_zoo.md\n\nhttps://github.com/tensorflow/models/blob/master/research/deeplab/g3doc/model_zoo.md\n\nhttps://github.com/tensorflow/models/tree/master/research/slim", "309723": "Mask-RCNN model pretrained on imagenet: https://github.com/matterport/Mask_RCNN/releases", "309653": "https://cs.adelaide.edu.au/~carneiro/isbi14_challenge/dataset.html\nhttps://nucleisegmentationbenchmark.weebly.com/dataset.html\nhttps://ome.grc.nia.nih.gov/iicbu2008/\nhttp://rnabiology.ircm.qc.ca/RBPImage/index.php\nhttps://zenodo.org/record/1174353#.WsGiztP4-3V\nhttps://warwick.ac.uk/fac/sci/dcs/research/tia/data/crchistolabelednucleihe\nhttps://becklab.hms.harvard.edu/software/psb-crowdsourced-nuclei-annotation-data-1", "308983": " https://github.com/lopuhin/kaggle-dsbowl-2018-dataset-fixes", "308964": "Small dataset of images with no nuclei is available here https://www.kaggle.com/jakubczakon/improve-the-dataset-with-non-nuclei-images", "308889": "A Dataset and a Technique for Generalized Nuclear Segmentation in Histological Images\nhttps://nucleisegmentationbenchmark.weebly.com\n\nHand-segmented 2D Nuclear Images\nhttp://murphylab.web.cmu.edu/data/2009_ISBI_Nuclei.html\n\nA dataset for nuclei segmentation based on Breast Cancer patients\nhttps://zenodo.org/record/1175282\n\nMask-RCNN model pretrained on COCO\nhttps://github.com/matterport/Mask_RCNN/releases", "308776": "https://github.com/matterport/Mask_RCNN/releases \n1.we use the link to fix the wrong labels(https://github.com/lopuhin/kaggle-dsbowl-2018-dataset-fixes).\n\n2.https://www.kaggle.com/voglinio/external-h-e-data-with-mask-annotations\n\n3.https://nucleisegmentationbenchmark.weebly.com/\n\n4.https://becklab.hms.harvard.edu/software/psb-crowdsourced-nuclei-annotation-data-1\n\n5.https://cs.adelaide.edu.au/~carneiro/isbi14_challenge/\nBroad Institute Image Sets: https://data.broadinstitute.org/bbbc/image_sets.html\n\nISBI '14 Pap Smears: https://cs.adelaide.edu.au/~carneiro/isbi14_challenge/dataset.html\n\nBioimage Datasets: https://bioimage.ucsb.edu/research/bio-segmentation\n\nNuclear Segmentation for Deep Learning: http://www.andrewjanowczyk.com/use-case-1-nuclei-segmentation/\n\nMurphy Dataset: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC2901896/\n https://github.com/lopuhin/kaggle-dsbowl-2018-dataset-fixes \nhttps://www.kaggle.com/voglinio/external-h-e-data-with-mask-annotations \nhttps://nucleisegmentationbenchmark.weebly.com/:\n https://zenodo.org/record/1174353#.WprukR9fg8o\n", "308774": "https://downloads.openmicroscopy.org/images/\n\nhttps://www.openmicroscopy.org/\n\nhttps://warwick.ac.uk/fac/sci/dcs/research/tia/data/\n", "308773": "https://github.com/sanuj/nuclei-net\n\nhttps://sanuj.github.io/nuclei-net-btp.html\n\nhttps://nucleisegmentationbenchmark.weebly.com/dataset.html\n\nhttps://sites.google.com/site/aacruzr/image-datasets, see (Histology and Histopathology, etc)\n\nhttp://histo.anat.ubc.ca/HISTOLOGY/Basic%20Tissues/view.apml?\n\nhuman altas\n\nhttps://www.proteinatlas.org/humanproteome", "308771": "http://histo.anat.ubc.ca/HISTOLOGY/Basic%20Tissues/view.apml\nhttps://ome.grc.nia.nih.gov/iicbu2008/\nhttps://data.broadinstitute.org/bbbc/BBBC006/\n\nhttps://bioimage.ucsb.edu/research/bio-segmentation", "308738": "but I may use some of the following datasets on the final solution. They all have already been shared.\n\n| dataset name | link | |--------------|----------------| | crchistophenotypes_2016_04_28 | https://warwick.ac.uk/fac/sci/dcs/research/tia/data/crchistolabelednucleihe/ | | nuclei-net-master | https://github.com/sanuj/nuclei-net | | nucleisegmentationbenchmark | https://nucleisegmentationbenchmark.weebly.com/dataset.html | | isbi14_challenge | https://cs.adelaide.edu.au/~carneiro/isbi14_challenge/dataset.html | | TNBC_NucleiSegmentation | https://zenodo.org/record/1175282#.WsOx73XFJFs | | nuclei | http://andrewjanowczyk.com/wp-static/nuclei.tgz | | PSB_2015_ImageSize_400 | https://becklab.hms.harvard.edu/software/psb-crowdsourced-nuclei-annotation-data-1 | | cluster nuclei | https://imagej.nih.gov/ij/plugins/ihc-toolbox/index.html |\n\nDataset at:\n\nhttps://github.com/sanuj/nuclei-net\n\nhttps://sanuj.github.io/nuclei-net-btp.html\n\nhttps://nucleisegmentationbenchmark.weebly.com/dataset.html\n\nhttps://sites.google.com/site/aacruzr/image-datasets, see (Histology and Histopathology, etc)\n\nhttp://histo.anat.ubc.ca/HISTOLOGY/Basic%20Tissues/view.apml?\n\nhuman altas\n\nhttps://www.proteinatlas.org/humanproteome\n\nAlso COCO weights from https://github.com/matterport/Mask_RCNN/releases may be used.\n\nhttps://downloads.openmicroscopy.org/images/\n\nhttps://www.openmicroscopy.org/\n\nhttps://warwick.ac.uk/fac/sci/dcs/research/tia/data/\n\n\nSome potential resources with related datasets :\n\nA comprehensive tutorial with selected use cases http://www.andrewjanowczyk.com/use-case-1-nuclei-segmentation/ (dataset : http://andrewjanowczyk.com/wp-static/nuclei.tgz)\n\nA Dataset and a Technique for Generalized Nuclear Segmentation in Histological Images https://nucleisegmentationbenchmark.weebly.com/\n\n2D images of 3T3 and U20S cells hand-segmented to show nuclear boundaries http://murphylab.web.cmu.edu/data/#2DNuclei (dataset : http://murphylab.web.cmu.edu/data/2009_ISBI_Nuclei.html)\n\nDigital Pathology : Nuclei Segmentation MICCAI 2017 Challenge http://miccai.cloudapp.net/competitions/57 (currently not responding)\n\nMitochondria segmentation dataset https://cvlab.epfl.ch/data/em", "308707": "Using tensorflow slim pre-trained weights:\n\nhttps://github.com/tensorflow/models/tree/master/research/slim", "308673": "Anyone happen to know if our training data isn't partially taken from https://data.broadinstitute.org/bbbc/BBBC006/ ?", "308634": "https://tma.im/cgi-bin/home.pl\n\nand also\n\nhttps://openi.nlm.nih.gov/", "308572": "Nothing new in this post, but I may use some of the following datasets on the final solution. They all have already been shared.\n\n| dataset name | link | \n|--------------|----------------|\n| crchistophenotypes_2016_04_28 | https://warwick.ac.uk/fac/sci/dcs/research/tia/data/crchistolabelednucleihe/ |\n| nuclei-net-master | https://github.com/sanuj/nuclei-net |\n| nucleisegmentationbenchmark | https://nucleisegmentationbenchmark.weebly.com/dataset.html |\n| isbi14_challenge | https://cs.adelaide.edu.au/~carneiro/isbi14_challenge/dataset.html |\n| TNBC_NucleiSegmentation | https://zenodo.org/record/1175282#.WsOx73XFJFs |\n| nuclei | http://andrewjanowczyk.com/wp-static/nuclei.tgz |\n| PSB_2015_ImageSize_400 | https://becklab.hms.harvard.edu/software/psb-crowdsourced-nuclei-annotation-data-1 |\n| cluster nuclei | https://imagej.nih.gov/ij/plugins/ihc-toolbox/index.html |\n\nAlso COCO weights from https://github.com/matterport/Mask_RCNN/releases may be used.\n\nSupport for markdown tables will be appreciated.", "308464": "Mask-RCNN model pretrained on COCO: https://github.com/matterport/Mask_RCNN/releases \nDetectron (https://github.com/facebookresearch/Detectron) along with several weight files from the model zoo (https://github.com/facebookresearch/Detectron/blob/master/MODEL_ZOO.md).\nOf course Allen's classes.csv from https://www.kaggle.com/c/data-science-bowl-2018/discussion/48130.\n", "308426": "http://www.robots.ox.ac.uk/~vgg/research/mitosis/\nAutomated Labelling of Cell Cycle Phases", "308286": "We use the External Data, so we claim it here.\n\n1.we use the link to fix the wrong labels(https://github.com/lopuhin/kaggle-dsbowl-2018-dataset-fixes).\n\n2.https://www.kaggle.com/voglinio/external-h-e-data-with-mask-annotations\n\n3.https://nucleisegmentationbenchmark.weebly.com/\n\n4.https://becklab.hms.harvard.edu/software/psb-crowdsourced-nuclei-annotation-data-1\n\n5.https://cs.adelaide.edu.au/~carneiro/isbi14_challenge/\n\n", "308204": "[Some (mostly unannotated) phase contrast images of cells & nuclei from Google Images, filtered to \"Labeled for reuse with modification](https://www.google.com/search?q=phase+contrast+(cells%7Cnuclei%7Ccytology)&tbm=isch&tbs=sur:fmc). Phase contrast is probably the most difficult imaging modality to generalize to.", "308159": "How about pretrained models included in common ML libraries? For example, I'm using a pretrained VGG-19 from PyTorch/torchvision (`models.vgg19(pretrained=True)`) at the moment, though that may change. I thought I'd read somewhere that if it was included in a popular library it didn't need to be declared here (because it would already be available to everyone), but I'm not sure.\n\nJust in case, the torchvision models are [here](http://pytorch.org/docs/master/torchvision/models.html).", "308046": "Broad Institute Image Sets:\nhttps://data.broadinstitute.org/bbbc/image_sets.html\n\nISBI '14 Pap Smears:\nhttps://cs.adelaide.edu.au/~carneiro/isbi14_challenge/dataset.html\n\nBioimage Datasets:\nhttps://bioimage.ucsb.edu/research/bio-segmentation\n\nNuclear Segmentation for Deep Learning:\nhttp://www.andrewjanowczyk.com/use-case-1-nuclei-segmentation/\n\nMurphy Dataset:\nhttps://www.ncbi.nlm.nih.gov/pmc/articles/PMC2901896/", "307999": "Fixed Stage 1 Train Masks by lopuhin: \nhttps://github.com/lopuhin/kaggle-dsbowl-2018-dataset-fixes\n\nH&E Dataset by voglinio: \nhttps://www.kaggle.com/voglinio/external-h-e-data-with-mask-annotations\nBased on:\nhttps://nucleisegmentationbenchmark.weebly.com/\n\nA dataset for nuclei segmentation based on Breast Cancer patients:\nhttps://zenodo.org/record/1174353#.WprukR9fg8o", "307499": "https://warwick.ac.uk/fac/sci/dcs/research/tia/data/crchistolabelednucleihe\nhttp://www.andrewjanowczyk.com/use-case-1-nuclei-segmentation", "307488": "https://nucleisegmentationbenchmark.weebly.com/", "307417": "https://github.com/gigascience/paper-bray2017\n\n\nSupporting data for \"A dataset of images and morphological profiles of 30,000 small-molecule treatments using the Cell Painting assay\"\n\nhttp://www.cellimagelibrary.org/browse/celltype", "307275": " 1. Pretrained Weights: https://github.com/matterport/Mask_RCNN/releases\n 2. H&E Dataset by voglinio: https://www.kaggle.com/voglinio/external-h-e-data-with-mask-annotations\n 3. Fixed Stage 1 Train Masks by lopuhin: https://github.com/lopuhin/kaggle-dsbowl-2018-dataset-fixes", "306931": "https://zenodo.org/record/1174353#.WprukR9fg8o", "306863": "hence I assume that any programming language other than Python like java etc will also work right ? ", "306847": "Do we need to use specific programming languages like Python , R ? or We can use any other programming languages like java ? In the competition rules its mentioned we have to develop the \"algorithm\" they have not specified any language restrictions hence I assume that any programming language other than java will also work right ? ", "306830": "https://www.kaggle.com/voglinio/external-h-e-data-with-mask-annotations\nhttps://www.kaggle.com/bonlime/train-test-image-mosaic\nhttps://github.com/lopuhin/kaggle-dsbowl-2018-dataset-fixes", "306192": "https://imagej.nih.gov/ij/plugins/ihc-toolbox/index.html", "304897": "https://commons.wikimedia.org/w/index.php?title=Category:Histopathology", "304302": "https://kaggle2.blob.core.windows.net/forum-message-attachments/273531/8366/classes.csv", "303714": "https://data.broadinstitute.org/bbbc/image_sets.html", "303452": "I'll use fixed stage1_train masks https://github.com/lopuhin/kaggle-dsbowl-2018-dataset-fixes", "303227": "Pretrained models from torchvision as well as pretrained models linked in [faster-rcnn-pytorch][1]\n\n\n [1]: https://github.com/jwyang/faster-rcnn.pytorch", "301461": "ImageNet weights for Keras: https://keras.io/applications/", "293522": "http://rnabiology.ircm.qc.ca/RBPImage/index.php", "292546": "Hello, I have a question. If I use the data other people post here should I mention it? Won't this create an echo effect every time somebody publishes a new DB? Is it enough to say I will try to use all the data available here? If yes, it is said. Thanks", "292005": "I am new to Kaggle. Is there a way to upload fixed-up or external data onto a notebook, so that we can run our kernel against that new data? I can't seem to find the upload button. Many of the public kernels are accessing the provided data under \"../input\", so it'd be nice to be able to add/replace the data there. Sorry if this is something everyone already knows how to do.", "290099": "https://github.com/facebookresearch/Detectron", "289215": "Images from wikimedia \nhttps://commons.wikimedia.org/wiki/Category:Cytopathology_of_cancers", "281586": "Can I use the pre-trained model on COCO datasets for competition for prize?", "280402": "Found one more dataset from CellTrackingChallenge: http://www.celltrackingchallenge.net/datasets.html\n\nThere are plenty of time-lapse data:\n\n", "280159": "Tumor Proliferation Assessment Challenge 2016\nhttp://tupac.tue-image.nl", "279666": "50 H&E stained histopathology images, breast cancer. This are images with class (purple) that has low samples count in train set and included in test set. [http://bioimage.ucsb.edu/research/bio-segmentation][1]\n\n@inproceedings{\nDrelie08-298,\nauthor = {Elisa Drelie Gelasca and Jiyun Byun and Boguslaw Obara and B.S. Manjunath},\ntitle = {Evaluation and Benchmark for Biological Image Segmentation},\nbooktitle = {IEEE International Conference on Image Processing},\nlocation = {San Diego, CA},\nmonth = {Oct},\nyear = {2008},\nurl = {http://vision.ece.ucsb.edu/publications/elisa_ICIP08.pdf}\n} \n\n\n [1]: http://bioimage.ucsb.edu/research/bio-segmentation", "277409": "Hi Anne, I am planning to get nuclei images from google to supplement the training set. I have applied the search filter of \"Labeled or Reuse\" on Google. Can I use these images for training?", "309640": "PyTorch Mask RCNN and Coco pre-trained weights:\nhttps://github.com/multimodallearning/pytorch-mask-rcnn", "308512": "" } }