{
  "id": 53337,
  "title": "Images for evaluating generalization",
  "url": "/competitions/data-science-bowl-2018/discussion/53337",
  "author_name": "",
  "post_date": "2018-03-29T14:29:05.928209500Z",
  "votes": 32,
  "comment_count": 9,
  "views": 0,
  "content": "<p>In stage 2, we have to detect nuclei in images which are in different domain of stage 1 images. So I googled and collected some possible extra images and did generalization test on them. Here I show the results of my model. If you have any other idea about external images for generalization test, please let me know it!\n<img src=\"https://storage.googleapis.com/kaggle-forum-message-attachments/305858/8901/summary.png\" alt=\"enter image description here\"></p>\n\n<p>links of images <br>\n<a href=\"https://s-media-cache-ak0.pinimg.com/originals/7c/17/09/7c17095c1ec6d9293765e45508719837.jpg\">https://s-media-cache-ak0.pinimg.com/originals/7c/17/09/7c17095c1ec6d9293765e45508719837.jpg</a> <br>\n<a href=\"https://thumbs.dreamstime.com/b/marble-chocolate-texture-7525822.jpg\">https://thumbs.dreamstime.com/b/marble-chocolate-texture-7525822.jpg</a> <br>\n<a href=\"http://jetspizza.com/dbphotos/display/24a6af7b5dd62183d15f56d33b4d4152/664/410\">http://jetspizza.com/dbphotos/display/24a6af7b5dd62183d15f56d33b4d4152/664/410</a> <br>\n<a href=\"http://corobuzz.com/wp-content/uploads/2016/03/df1.jpg\">http://corobuzz.com/wp-content/uploads/2016/03/df1.jpg</a>  </p>",
  "messages": [
    {
      "id": "305858",
      "postDate": "03/29/2018 14:29:05",
      "content": "<p>In stage 2, we have to detect nuclei in images which are in different domain of stage 1 images. So I googled and collected some possible extra images and did generalization test on them. Here I show the results of my model. If you have any other idea about external images for generalization test, please let me know it!\n<img src=\"https://storage.googleapis.com/kaggle-forum-message-attachments/305858/8901/summary.png\" alt=\"enter image description here\"></p>\n\n<p>links of images <br>\n<a href=\"https://s-media-cache-ak0.pinimg.com/originals/7c/17/09/7c17095c1ec6d9293765e45508719837.jpg\">https://s-media-cache-ak0.pinimg.com/originals/7c/17/09/7c17095c1ec6d9293765e45508719837.jpg</a> <br>\n<a href=\"https://thumbs.dreamstime.com/b/marble-chocolate-texture-7525822.jpg\">https://thumbs.dreamstime.com/b/marble-chocolate-texture-7525822.jpg</a> <br>\n<a href=\"http://jetspizza.com/dbphotos/display/24a6af7b5dd62183d15f56d33b4d4152/664/410\">http://jetspizza.com/dbphotos/display/24a6af7b5dd62183d15f56d33b4d4152/664/410</a> <br>\n<a href=\"http://corobuzz.com/wp-content/uploads/2016/03/df1.jpg\">http://corobuzz.com/wp-content/uploads/2016/03/df1.jpg</a>  </p>",
      "rawMarkdown": "In stage 2, we have to detect nuclei in images which are in different domain of stage 1 images. So I googled and collected some possible extra images and did generalization test on them. Here I show the results of my model. If you have any other idea about external images for generalization test, please let me know it!\n![enter image description here][1]\n\nlinks of images  \nhttps://s-media-cache-ak0.pinimg.com/originals/7c/17/09/7c17095c1ec6d9293765e45508719837.jpg  \nhttps://thumbs.dreamstime.com/b/marble-chocolate-texture-7525822.jpg  \nhttp://jetspizza.com/dbphotos/display/24a6af7b5dd62183d15f56d33b4d4152/664/410  \nhttp://corobuzz.com/wp-content/uploads/2016/03/df1.jpg  \n\n  [1]: https://kaggle2.blob.core.windows.net/forum-message-attachments/305858/8901/summary.png",
      "votes": null
    },
    {
      "id": "305888",
      "postDate": "03/29/2018 15:16:20",
      "content": "<p>This is great. Looks like solid validation set.</p>",
      "rawMarkdown": "This is great. Looks like solid validation set.",
      "votes": null
    },
    {
      "id": "305921",
      "postDate": "03/29/2018 16:14:12",
      "content": "<p>Awesome...How about Dalmatians?</p>",
      "rawMarkdown": "Awesome...How about Dalmatians?",
      "votes": null
    },
    {
      "id": "306584",
      "postDate": "03/30/2018 17:49:36",
      "content": "<p>😂</p>\n\n<p>I’m looking forward to your forthcoming Nature paper, “Muffin Adversaries for Biological Imaging!”</p>",
      "rawMarkdown": "😂\n\nI’m looking forward to your forthcoming Nature paper, “Muffin Adversaries for Biological Imaging!”",
      "votes": null
    },
    {
      "id": "306594",
      "postDate": "03/30/2018 18:05:39",
      "content": "<p>Eggs look totally valid 2nd stage test set :) - there was a sample like this in another Kaggle challenge for counting sea lions from aerial photos\n<a href=\"https://storage.googleapis.com/kaggle-forum-message-attachments/177392/6368/4659.jpg\">https://storage.googleapis.com/kaggle-forum-message-attachments/177392/6368/4659.jpg</a> </p>",
      "rawMarkdown": "Eggs look totally valid 2nd stage test set :) - there was a sample like this in another Kaggle challenge for counting sea lions from aerial photos\nhttps://kaggle2.blob.core.windows.net/forum-message-attachments/177392/6368/4659.jpg",
      "votes": null
    },
    {
      "id": "306811",
      "postDate": "03/31/2018 04:03:49",
      "content": "<p>Our deep learning models are in fact excellent ellipse detectors! We are still far from making cognitive learning machines! :(  </p>",
      "rawMarkdown": "Our deep learning models are in fact excellent ellipse detectors! We are still far from making cognitive learning machines! :(",
      "votes": null
    },
    {
      "id": "308201",
      "postDate": "04/03/2018 05:43:14",
      "content": "<p>Here are a few tougher-looking ones (transparency, overlap) from Google Images. They are not all appropriately licensed, so don't train on them if you want your model to be reusable after the competition, but they should be fine for generalizability checks.</p>\n\n<ul>\n<li><a href=\"https://www.google.com/search?tbm=isch&amp;q=transparent+balls\">transparent balls</a> - try filtering by color, under Tools</li>\n<li><a href=\"https://www.google.com/imgres?imgurl=http%3A%2F%2Ftheresabergdahl.com%2Fwp-content%2Fuploads%2F2016%2F07%2FColorful-Circles-Abstract-Vector-Background.jpg&amp;imgrefurl=http%3A%2F%2Ftheresabergdahl.com%2F2016%2F03%2Fgroup-singing-circles%2F&amp;docid=4mTJmjObMYMo5M&amp;tbnid=LtVOFj2WmTPFZM%3A\">this image</a> and the related images</li>\n<li><a href=\"https://www.google.com/search?tbm=isch&amp;q=glass+beads\">glass beads</a></li>\n<li><a href=\"https://www.google.com/search?tbm=isch&amp;q=leaves+floating+on+water+-maple\">leaves on water</a>, <a href=\"https://www.google.com/search?tbm=isch&amp;q=leaves+on+water+-maple\">leaves on water and water on leaves</a></li>\n<li><a href=\"https://www.google.com/search?tbm=isch&amp;q=center+pivot+irrigation+%28aerial%7Csatellite%29\">center-pivot irrigation</a></li>\n<li><a href=\"https://www.google.com/search?tbm=isch&amp;q=phase+contrast+%28cells%7Cnuclei%7Ccytology%29\">phase-contrast images of cells &amp; nuclei</a> - I'll also post this one, filtered to \"labeled for reuse with modification\", in the external data thread</li>\n</ul>\n\n<p>I'm also going to look for interesting things to take photos of over the next few days. I don't see a real need to get them in before the external data deadline, because I don't expect anybody to want to train on them, but if you do, speak up.</p>",
      "rawMarkdown": "Here are a few tougher-looking ones (transparency, overlap) from Google Images. They are not all appropriately licensed, so don't train on them if you want your model to be reusable after the competition, but they should be fine for generalizability checks.\n\n- [transparent balls](https://www.google.com/search?tbm=isch&amp;q=transparent+balls) - try filtering by color, under Tools\n- [this image](https://www.google.com/imgres?imgurl=http%3A%2F%2Ftheresabergdahl.com%2Fwp-content%2Fuploads%2F2016%2F07%2FColorful-Circles-Abstract-Vector-Background.jpg&amp;imgrefurl=http%3A%2F%2Ftheresabergdahl.com%2F2016%2F03%2Fgroup-singing-circles%2F&amp;docid=4mTJmjObMYMo5M&amp;tbnid=LtVOFj2WmTPFZM%3A) and the related images\n- [glass beads](https://www.google.com/search?tbm=isch&amp;q=glass+beads)\n- [leaves on water](https://www.google.com/search?tbm=isch&amp;q=leaves+floating+on+water+-maple), [leaves on water and water on leaves](https://www.google.com/search?tbm=isch&amp;q=leaves+on+water+-maple)\n- [center-pivot irrigation](https://www.google.com/search?tbm=isch&amp;q=center+pivot+irrigation+%28aerial%7Csatellite%29)\n- [phase-contrast images of cells &amp; nuclei](https://www.google.com/search?tbm=isch&amp;q=phase+contrast+%28cells%7Cnuclei%7Ccytology%29) - I'll also post this one, filtered to \"labeled for reuse with modification\", in the external data thread\n\nI'm also going to look for interesting things to take photos of over the next few days. I don't see a real need to get them in before the external data deadline, because I don't expect anybody to want to train on them, but if you do, speak up.",
      "votes": null
    },
    {
      "id": "309214",
      "postDate": "04/04/2018 20:19:11",
      "content": "<p>@OsciiArt No kidding, you make my day. I laughed out loudly in front of my being-trained-models :D  Please share more of your evaluations at the stage 2. </p>",
      "rawMarkdown": "OsciiArt No kidding, you make my day. I laughed out loudly in front of my being-trained-models :D  Please share more of your evaluations at the stage 2.",
      "votes": null
    },
    {
      "id": "309224",
      "postDate": "04/04/2018 20:40:59",
      "content": "<p>Your profile picture seems so appropriate...</p>",
      "rawMarkdown": "Your profile picture seems so appropriate...",
      "votes": null
    },
    {
      "id": "309343",
      "postDate": "04/05/2018 04:50:03",
      "content": "<p>Here are some photos I took. They're not all especially similar to the nucleus images in ellipse content, but I think that's a good thing: If you expect the stage 2 nuclei to all be elliptical, you can make sure you're not generalizing too much; if you don't, you can make sure you're generalizing enough.</p>",
      "rawMarkdown": "Here are some photos I took. They're not all especially similar to the nucleus images in ellipse content, but I think that's a good thing: If you expect the stage 2 nuclei to all be elliptical, you can make sure you're not generalizing too much; if you don't, you can make sure you're generalizing enough.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 305888,
      "author_name": "outliar",
      "author_url": "",
      "post_date": "03/29/2018 15:16:20",
      "content": "<p>This is great. Looks like solid validation set.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 305921,
      "author_name": "jamesrequa",
      "author_url": "",
      "post_date": "03/29/2018 16:14:12",
      "content": "<p>Awesome...How about Dalmatians?</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 306584,
      "author_name": "agoodman",
      "author_url": "",
      "post_date": "03/30/2018 17:49:36",
      "content": "<p>😂</p>\n\n<p>I’m looking forward to your forthcoming Nature paper, “Muffin Adversaries for Biological Imaging!”</p>",
      "votes": null,
      "replies": [
        {
          "id": 309224,
          "author_name": "bsp2020",
          "author_url": "",
          "post_date": "04/04/2018 20:40:59",
          "content": "<p>Your profile picture seems so appropriate...</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 306594,
      "author_name": "jandjenter",
      "author_url": "",
      "post_date": "03/30/2018 18:05:39",
      "content": "<p>Eggs look totally valid 2nd stage test set :) - there was a sample like this in another Kaggle challenge for counting sea lions from aerial photos\n<a href=\"https://storage.googleapis.com/kaggle-forum-message-attachments/177392/6368/4659.jpg\">https://storage.googleapis.com/kaggle-forum-message-attachments/177392/6368/4659.jpg</a> </p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 306811,
      "author_name": "sgalib",
      "author_url": "",
      "post_date": "03/31/2018 04:03:49",
      "content": "<p>Our deep learning models are in fact excellent ellipse detectors! We are still far from making cognitive learning machines! :(  </p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 308201,
      "author_name": "pointyointment",
      "author_url": "",
      "post_date": "04/03/2018 05:43:14",
      "content": "<p>Here are a few tougher-looking ones (transparency, overlap) from Google Images. They are not all appropriately licensed, so don't train on them if you want your model to be reusable after the competition, but they should be fine for generalizability checks.</p>\n\n<ul>\n<li><a href=\"https://www.google.com/search?tbm=isch&amp;q=transparent+balls\">transparent balls</a> - try filtering by color, under Tools</li>\n<li><a href=\"https://www.google.com/imgres?imgurl=http%3A%2F%2Ftheresabergdahl.com%2Fwp-content%2Fuploads%2F2016%2F07%2FColorful-Circles-Abstract-Vector-Background.jpg&amp;imgrefurl=http%3A%2F%2Ftheresabergdahl.com%2F2016%2F03%2Fgroup-singing-circles%2F&amp;docid=4mTJmjObMYMo5M&amp;tbnid=LtVOFj2WmTPFZM%3A\">this image</a> and the related images</li>\n<li><a href=\"https://www.google.com/search?tbm=isch&amp;q=glass+beads\">glass beads</a></li>\n<li><a href=\"https://www.google.com/search?tbm=isch&amp;q=leaves+floating+on+water+-maple\">leaves on water</a>, <a href=\"https://www.google.com/search?tbm=isch&amp;q=leaves+on+water+-maple\">leaves on water and water on leaves</a></li>\n<li><a href=\"https://www.google.com/search?tbm=isch&amp;q=center+pivot+irrigation+%28aerial%7Csatellite%29\">center-pivot irrigation</a></li>\n<li><a href=\"https://www.google.com/search?tbm=isch&amp;q=phase+contrast+%28cells%7Cnuclei%7Ccytology%29\">phase-contrast images of cells &amp; nuclei</a> - I'll also post this one, filtered to \"labeled for reuse with modification\", in the external data thread</li>\n</ul>\n\n<p>I'm also going to look for interesting things to take photos of over the next few days. I don't see a real need to get them in before the external data deadline, because I don't expect anybody to want to train on them, but if you do, speak up.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 309214,
      "author_name": "nicolefinnie",
      "author_url": "",
      "post_date": "04/04/2018 20:19:11",
      "content": "<p>@OsciiArt No kidding, you make my day. I laughed out loudly in front of my being-trained-models :D  Please share more of your evaluations at the stage 2. </p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 309343,
      "author_name": "pointyointment",
      "author_url": "",
      "post_date": "04/05/2018 04:50:03",
      "content": "<p>Here are some photos I took. They're not all especially similar to the nucleus images in ellipse content, but I think that's a good thing: If you expect the stage 2 nuclei to all be elliptical, you can make sure you're not generalizing too much; if you don't, you can make sure you're generalizing enough.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "305858": "In stage 2, we have to detect nuclei in images which are in different domain of stage 1 images. So I googled and collected some possible extra images and did generalization test on them. Here I show the results of my model. If you have any other idea about external images for generalization test, please let me know it!\n![enter image description here][1]\n\nlinks of images  \nhttps://s-media-cache-ak0.pinimg.com/originals/7c/17/09/7c17095c1ec6d9293765e45508719837.jpg  \nhttps://thumbs.dreamstime.com/b/marble-chocolate-texture-7525822.jpg  \nhttp://jetspizza.com/dbphotos/display/24a6af7b5dd62183d15f56d33b4d4152/664/410  \nhttp://corobuzz.com/wp-content/uploads/2016/03/df1.jpg  \n\n  [1]: https://kaggle2.blob.core.windows.net/forum-message-attachments/305858/8901/summary.png",
    "305888": "This is great. Looks like solid validation set.",
    "305921": "Awesome...How about Dalmatians?",
    "306584": "😂\n\nI’m looking forward to your forthcoming Nature paper, “Muffin Adversaries for Biological Imaging!”",
    "306594": "Eggs look totally valid 2nd stage test set :) - there was a sample like this in another Kaggle challenge for counting sea lions from aerial photos\nhttps://kaggle2.blob.core.windows.net/forum-message-attachments/177392/6368/4659.jpg",
    "306811": "Our deep learning models are in fact excellent ellipse detectors! We are still far from making cognitive learning machines! :(",
    "308201": "Here are a few tougher-looking ones (transparency, overlap) from Google Images. They are not all appropriately licensed, so don't train on them if you want your model to be reusable after the competition, but they should be fine for generalizability checks.\n\n- [transparent balls](https://www.google.com/search?tbm=isch&amp;q=transparent+balls) - try filtering by color, under Tools\n- [this image](https://www.google.com/imgres?imgurl=http%3A%2F%2Ftheresabergdahl.com%2Fwp-content%2Fuploads%2F2016%2F07%2FColorful-Circles-Abstract-Vector-Background.jpg&amp;imgrefurl=http%3A%2F%2Ftheresabergdahl.com%2F2016%2F03%2Fgroup-singing-circles%2F&amp;docid=4mTJmjObMYMo5M&amp;tbnid=LtVOFj2WmTPFZM%3A) and the related images\n- [glass beads](https://www.google.com/search?tbm=isch&amp;q=glass+beads)\n- [leaves on water](https://www.google.com/search?tbm=isch&amp;q=leaves+floating+on+water+-maple), [leaves on water and water on leaves](https://www.google.com/search?tbm=isch&amp;q=leaves+on+water+-maple)\n- [center-pivot irrigation](https://www.google.com/search?tbm=isch&amp;q=center+pivot+irrigation+%28aerial%7Csatellite%29)\n- [phase-contrast images of cells &amp; nuclei](https://www.google.com/search?tbm=isch&amp;q=phase+contrast+%28cells%7Cnuclei%7Ccytology%29) - I'll also post this one, filtered to \"labeled for reuse with modification\", in the external data thread\n\nI'm also going to look for interesting things to take photos of over the next few days. I don't see a real need to get them in before the external data deadline, because I don't expect anybody to want to train on them, but if you do, speak up.",
    "309214": "OsciiArt No kidding, you make my day. I laughed out loudly in front of my being-trained-models :D  Please share more of your evaluations at the stage 2.",
    "309224": "Your profile picture seems so appropriate...",
    "309343": "Here are some photos I took. They're not all especially similar to the nucleus images in ellipse content, but I think that's a good thing: If you expect the stage 2 nuclei to all be elliptical, you can make sure you're not generalizing too much; if you don't, you can make sure you're generalizing enough."
  },
  "source": "meta"
}