{
  "id": 70231,
  "title": "tSNEfied images",
  "url": "/competitions/human-protein-atlas-image-classification/discussion/70231",
  "author_name": "Tilii",
  "post_date": "2018-11-01T06:10:34.079000",
  "votes": 3,
  "comment_count": 0,
  "views": 0,
  "content": "<p>I was wondering how well the RGB images (yellow was not used) would embed by tSNE. Looked at a total of 2500 images (all 271 images with label <code>1</code>; the remaining images were from classes <code>0, 7 and 25</code> in the same proportion as in train data). Features were derived by VGG16, and after the embedding each image was resized to 128x128 and laid out in a 50x50 grid.</p>\n\n<p><img src=\"https://i.postimg.cc/fR1P5Zy4/cells-tsne-grid.jpg\" alt=\"enter image description here\"></p>\n\n<p>Since the image will probably be resized to fit the page, you may want to take a look at the attached high-res version.</p>",
  "messages": [
    {
      "id": 413555,
      "postDate": "2018-11-01T06:10:34.080Z",
      "content": "<p>I was wondering how well the RGB images (yellow was not used) would embed by tSNE. Looked at a total of 2500 images (all 271 images with label <code>1</code>; the remaining images were from classes <code>0, 7 and 25</code> in the same proportion as in train data). Features were derived by VGG16, and after the embedding each image was resized to 128x128 and laid out in a 50x50 grid.</p>\n\n<p><img src=\"https://i.postimg.cc/fR1P5Zy4/cells-tsne-grid.jpg\" alt=\"enter image description here\"></p>\n\n<p>Since the image will probably be resized to fit the page, you may want to take a look at the attached high-res version.</p>",
      "rawMarkdown": "I was wondering how well the RGB images (yellow was not used) would embed by tSNE. Looked at a total of 2500 images (all 271 images with label `1`; the remaining images were from classes `0, 7 and 25` in the same proportion as in train data). Features were derived by VGG16, and after the embedding each image was resized to 128x128 and laid out in a 50x50 grid.\n\n![enter image description here][1]\n\nSince the image will probably be resized to fit the page, you may want to take a look at the attached high-res version.\n\n  [1]: https://i.postimg.cc/fR1P5Zy4/cells-tsne-grid.jpg",
      "votes": 3
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "413555": "I was wondering how well the RGB images (yellow was not used) would embed by tSNE. Looked at a total of 2500 images (all 271 images with label `1`; the remaining images were from classes `0, 7 and 25` in the same proportion as in train data). Features were derived by VGG16, and after the embedding each image was resized to 128x128 and laid out in a 50x50 grid.\n\n![enter image description here][1]\n\nSince the image will probably be resized to fit the page, you may want to take a look at the attached high-res version.\n\n  [1]: https://i.postimg.cc/fR1P5Zy4/cells-tsne-grid.jpg"
  }
}