![\"\"](\"https://i.imgur.com/TMmDp6d.png\")
It seems there are some huge bboxes in some images. Is this normal?\n
It seems there are some huge bboxes in some images. Is this normal?\n
I think it is not normal. I've found a few similar, like the 2nd and the 3rd in this image. (The first seems normal)
\n\n![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F864684%2F470377695dd91a4c68e24e1e9208886d%2Fexample.jpg?generation=1588775544392115&alt=media\")
I have published my kernel right now, I show that there are less than 5% of such bboxes. Still this is interesting.
\n\nhttps://www.kaggle.com/artgor/object-detection-with-pytorch-lightning/
", "rawMarkdown": "I have published my kernel right now, I show that there are less than 5% of such bboxes. Still this is interesting.\n\nhttps://www.kaggle.com/artgor/object-detection-with-pytorch-lightning/", "votes": null }, { "id": "835992", "postDate": "05/06/2020 16:38:47", "content": "Thank for noticing this problem ! \n1- It's not normal\n2- I'll investigate the origin of the problem...
\n\nIs it happening on all locations ?
", "rawMarkdown": "Thank for noticing this problem ! \n1- It's not normal\n2- I'll investigate the origin of the problem... \n\nIs it happening on all locations ?", "votes": null }, { "id": "836009", "postDate": "05/06/2020 16:48:55", "content": "Thank you for the answer!
\n\nYes, they are in various locations:\n![\"\"](\"https://i.imgur.com/eJnB2bR.png\")
Ok I will check the original data to find why such boxes appear ! Thanks you for sharing the issue !
", "rawMarkdown": "Ok I will check the original data to find why such boxes appear ! Thanks you for sharing the issue !", "votes": null }, { "id": "836737", "postDate": "05/07/2020 07:41:13", "content": "For your information, the huge bbox comes from a weird bug of our annotation tool which assigned same id to different boxes, resulting in a super large \"merged box\". There are not too much very problematic edge cases like that but I will send to Kaggle an update version of such boxes, because we still have the individual \"segmentation mask\". Still there will be quite large boxes because in some images the wheat head is quite close to the camera, but at least it is a real life edgee case to handle and not a problem with the data :)
\n\nAnother issue I've noticed working on that there are some too small boxes are present in the file. I thought I filtered them before but apparently not. This small boxes often have a height or width below 10 and was too difficult to notice by a human reviewer on our tool. Most of the problems will be solved in the update and again it represents very few bounding boxes (apparently 170 on almost 150k).
\n\nI hope it help you with the competition and thank you again for your feedback ! This kind of competition is really new for our community, that's the reason there are few mistakes at the launch, but we really want to build the fairest competition possible !
", "rawMarkdown": "For your information, the huge bbox comes from a weird bug of our annotation tool which assigned same id to different boxes, resulting in a super large \"merged box\". There are not too much very problematic edge cases like that but I will send to Kaggle an update version of such boxes, because we still have the individual \"segmentation mask\". Still there will be quite large boxes because in some images the wheat head is quite close to the camera, but at least it is a real life edgee case to handle and not a problem with the data :)\n\nAnother issue I've noticed working on that there are some too small boxes are present in the file. I thought I filtered them before but apparently not. This small boxes often have a height or width below 10 and was too difficult to notice by a human reviewer on our tool. Most of the problems will be solved in the update and again it represents very few bounding boxes (apparently 170 on almost 150k). \n\nI hope it help you with the competition and thank you again for your feedback ! This kind of competition is really new for our community, that's the reason there are few mistakes at the launch, but we really want to build the fairest competition possible !", "votes": null }, { "id": "836802", "postDate": "05/07/2020 08:53:55", "content": "@bendvd is it possible to provide also segmentation masks that as you mentioned you already have?
", "rawMarkdown": "bendvd is it possible to provide also segmentation masks that as you mentioned you already have?", "votes": null }, { "id": "836837", "postDate": "05/07/2020 09:43:07", "content": "Thank you for your fast answer!
", "rawMarkdown": "Thank you for your fast answer!", "votes": null }, { "id": "837834", "postDate": "05/08/2020 04:41:24", "content": "I think you should make a threshold on the size of the boxes for example reducing the maximum size of a box .
", "rawMarkdown": "I think you should make a threshold on the size of the boxes for example reducing the maximum size of a box .", "votes": null }, { "id": "842062", "postDate": "05/11/2020 06:56:24", "content": "@bendvd Excuse me, About when do you think Data Souce will be updated ?
", "rawMarkdown": "bendvd Excuse me, About when do you think Data Souce will be updated ?", "votes": null }, { "id": "848630", "postDate": "05/15/2020 05:53:02", "content": "This is the original dataset without any corrections. Just in case someone needs it if an updated dataset is released.\nhttps://www.kaggle.com/vbookshelf/wheat-detection-comp-original-data
", "rawMarkdown": "This is the original dataset without any corrections. Just in case someone needs it if an updated dataset is released.\nhttps://www.kaggle.com/vbookshelf/wheat-detection-comp-original-data", "votes": null }, { "id": "850526", "postDate": "05/16/2020 18:07:13", "content": "@vbookshelf hi, is there any particular reason you saved the original version? I mean, didn't it have some issues like unexpected large bounding boxes, etc?
", "rawMarkdown": "vbookshelf hi, is there any particular reason you saved the original version? I mean, didn't it have some issues like unexpected large bounding boxes, etc?", "votes": null }, { "id": "850529", "postDate": "05/16/2020 18:10:12", "content": "@bendvd hi, is the dataset updated? If so, can you please point out what's the major difference in between? I downloaded the dataset a few days ago, and I didn't follow if I have the updated version or not. However, I found that there is some wheat head contains images without bounding box, is it normal? Though I read in the documentation about it, just wanted to re-confirm from you. Thanks.
", "rawMarkdown": "bendvd hi, is the dataset updated? If so, can you please point out what's the major difference in between? I downloaded the dataset a few days ago, and I didn't follow if I have the updated version or not. However, I found that there is some wheat head contains images without bounding box, is it normal? Though I read in the documentation about it, just wanted to re-confirm from you. Thanks.", "votes": null }, { "id": "850775", "postDate": "05/17/2020 02:25:02", "content": "Hi @ipythonx. I did this just as a precaution. With any change there's always the possibility of bugs and unintentional mistakes. My code is increasing in complexity and I didn't want to run the risk of any dataset changes breaking the code.
\n\nThat said, I haven't yet seen any notice in the forums that the original dataset has been updated. Personally, I think it should be left as is.
", "rawMarkdown": "Hi @ipythonx. I did this just as a precaution. With any change there's always the possibility of bugs and unintentional mistakes. My code is increasing in complexity and I didn't want to run the risk of any dataset changes breaking the code. \n\nThat said, I haven't yet seen any notice in the forums that the original dataset has been updated. Personally, I think it should be left as is.", "votes": null }, { "id": "850796", "postDate": "05/17/2020 03:06:46", "content": "@vbookshelf ok, thanks for the reply.\nAnyway, I really want to who downvoted both of our comments here, what's the cause 😄
", "rawMarkdown": "vbookshelf ok, thanks for the reply.\nAnyway, I really want to who downvoted both of our comments here, what's the cause 😄", "votes": null }, { "id": "864839", "postDate": "05/28/2020 08:12:54", "content": "@bendvd Any update? I dont want to spend my time for data cleaning.
", "rawMarkdown": "bendvd Any update? I dont want to spend my time for data cleaning.", "votes": null } ], "comments": [ { "id": 835838, "author_name": "pestipeti", "author_url": "", "post_date": "05/06/2020 14:43:20", "content": "I think it is not normal. I've found a few similar, like the 2nd and the 3rd in this image. (The first seems normal)
\n\n
I have published my kernel right now, I show that there are less than 5% of such bboxes. Still this is interesting.
\n\nhttps://www.kaggle.com/artgor/object-detection-with-pytorch-lightning/
", "votes": null, "replies": [] } ] }, { "id": 835992, "author_name": "bendvd", "author_url": "", "post_date": "05/06/2020 16:38:47", "content": "Thank for noticing this problem ! \n1- It's not normal\n2- I'll investigate the origin of the problem...
\n\nIs it happening on all locations ?
", "votes": null, "replies": [ { "id": 836009, "author_name": "artgor", "author_url": "", "post_date": "05/06/2020 16:48:55", "content": "Thank you for the answer!
\n\nYes, they are in various locations:\n
Ok I will check the original data to find why such boxes appear ! Thanks you for sharing the issue !
", "votes": null, "replies": [] }, { "id": 836737, "author_name": "bendvd", "author_url": "", "post_date": "05/07/2020 07:41:13", "content": "For your information, the huge bbox comes from a weird bug of our annotation tool which assigned same id to different boxes, resulting in a super large \"merged box\". There are not too much very problematic edge cases like that but I will send to Kaggle an update version of such boxes, because we still have the individual \"segmentation mask\". Still there will be quite large boxes because in some images the wheat head is quite close to the camera, but at least it is a real life edgee case to handle and not a problem with the data :)
\n\nAnother issue I've noticed working on that there are some too small boxes are present in the file. I thought I filtered them before but apparently not. This small boxes often have a height or width below 10 and was too difficult to notice by a human reviewer on our tool. Most of the problems will be solved in the update and again it represents very few bounding boxes (apparently 170 on almost 150k).
\n\nI hope it help you with the competition and thank you again for your feedback ! This kind of competition is really new for our community, that's the reason there are few mistakes at the launch, but we really want to build the fairest competition possible !
", "votes": null, "replies": [] }, { "id": 836837, "author_name": "artgor", "author_url": "", "post_date": "05/07/2020 09:43:07", "content": "Thank you for your fast answer!
", "votes": null, "replies": [] }, { "id": 842062, "author_name": "nakajima", "author_url": "", "post_date": "05/11/2020 06:56:24", "content": "@bendvd Excuse me, About when do you think Data Souce will be updated ?
", "votes": null, "replies": [] }, { "id": 850529, "author_name": "ipythonx", "author_url": "", "post_date": "05/16/2020 18:10:12", "content": "@bendvd hi, is the dataset updated? If so, can you please point out what's the major difference in between? I downloaded the dataset a few days ago, and I didn't follow if I have the updated version or not. However, I found that there is some wheat head contains images without bounding box, is it normal? Though I read in the documentation about it, just wanted to re-confirm from you. Thanks.
", "votes": null, "replies": [] }, { "id": 864839, "author_name": "lisosia", "author_url": "", "post_date": "05/28/2020 08:12:54", "content": "@bendvd Any update? I dont want to spend my time for data cleaning.
", "votes": null, "replies": [] } ] }, { "id": 836802, "author_name": "maksimovka", "author_url": "", "post_date": "05/07/2020 08:53:55", "content": "@bendvd is it possible to provide also segmentation masks that as you mentioned you already have?
", "votes": null, "replies": [] }, { "id": 837834, "author_name": "jarvisai7", "author_url": "", "post_date": "05/08/2020 04:41:24", "content": "I think you should make a threshold on the size of the boxes for example reducing the maximum size of a box .
", "votes": null, "replies": [] }, { "id": 848630, "author_name": "vbookshelf", "author_url": "", "post_date": "05/15/2020 05:53:02", "content": "This is the original dataset without any corrections. Just in case someone needs it if an updated dataset is released.\nhttps://www.kaggle.com/vbookshelf/wheat-detection-comp-original-data
", "votes": null, "replies": [ { "id": 850526, "author_name": "ipythonx", "author_url": "", "post_date": "05/16/2020 18:07:13", "content": "@vbookshelf hi, is there any particular reason you saved the original version? I mean, didn't it have some issues like unexpected large bounding boxes, etc?
", "votes": null, "replies": [] }, { "id": 850775, "author_name": "vbookshelf", "author_url": "", "post_date": "05/17/2020 02:25:02", "content": "Hi @ipythonx. I did this just as a precaution. With any change there's always the possibility of bugs and unintentional mistakes. My code is increasing in complexity and I didn't want to run the risk of any dataset changes breaking the code.
\n\nThat said, I haven't yet seen any notice in the forums that the original dataset has been updated. Personally, I think it should be left as is.
", "votes": null, "replies": [] }, { "id": 850796, "author_name": "ipythonx", "author_url": "", "post_date": "05/17/2020 03:06:46", "content": "@vbookshelf ok, thanks for the reply.\nAnyway, I really want to who downvoted both of our comments here, what's the cause 😄
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "835764": "It seems there are some huge bboxes in some images. Is this normal?\n", "835838": "I think it is not normal. I've found a few similar, like the 2nd and the 3rd in this image. (The first seems normal)\n\n", "835855": "I have published my kernel right now, I show that there are less than 5% of such bboxes. Still this is interesting.\n\nhttps://www.kaggle.com/artgor/object-detection-with-pytorch-lightning/", "835992": "Thank for noticing this problem ! \n1- It's not normal\n2- I'll investigate the origin of the problem... \n\nIs it happening on all locations ?", "836009": "Thank you for the answer!\n\nYes, they are in various locations:\n", "836124": "Ok I will check the original data to find why such boxes appear ! Thanks you for sharing the issue !", "836737": "For your information, the huge bbox comes from a weird bug of our annotation tool which assigned same id to different boxes, resulting in a super large \"merged box\". There are not too much very problematic edge cases like that but I will send to Kaggle an update version of such boxes, because we still have the individual \"segmentation mask\". Still there will be quite large boxes because in some images the wheat head is quite close to the camera, but at least it is a real life edgee case to handle and not a problem with the data :)\n\nAnother issue I've noticed working on that there are some too small boxes are present in the file. I thought I filtered them before but apparently not. This small boxes often have a height or width below 10 and was too difficult to notice by a human reviewer on our tool. Most of the problems will be solved in the update and again it represents very few bounding boxes (apparently 170 on almost 150k). \n\nI hope it help you with the competition and thank you again for your feedback ! This kind of competition is really new for our community, that's the reason there are few mistakes at the launch, but we really want to build the fairest competition possible !", "836802": "bendvd is it possible to provide also segmentation masks that as you mentioned you already have?", "836837": "Thank you for your fast answer!", "837834": "I think you should make a threshold on the size of the boxes for example reducing the maximum size of a box .", "842062": "bendvd Excuse me, About when do you think Data Souce will be updated ?", "848630": "This is the original dataset without any corrections. Just in case someone needs it if an updated dataset is released.\nhttps://www.kaggle.com/vbookshelf/wheat-detection-comp-original-data", "850526": "vbookshelf hi, is there any particular reason you saved the original version? I mean, didn't it have some issues like unexpected large bounding boxes, etc?", "850529": "bendvd hi, is the dataset updated? If so, can you please point out what's the major difference in between? I downloaded the dataset a few days ago, and I didn't follow if I have the updated version or not. However, I found that there is some wheat head contains images without bounding box, is it normal? Though I read in the documentation about it, just wanted to re-confirm from you. Thanks.", "850775": "Hi @ipythonx. I did this just as a precaution. With any change there's always the possibility of bugs and unintentional mistakes. My code is increasing in complexity and I didn't want to run the risk of any dataset changes breaking the code. \n\nThat said, I haven't yet seen any notice in the forums that the original dataset has been updated. Personally, I think it should be left as is.", "850796": "vbookshelf ok, thanks for the reply.\nAnyway, I really want to who downvoted both of our comments here, what's the cause 😄", "864839": "bendvd Any update? I dont want to spend my time for data cleaning." }, "source": "meta" }