{
  "id": 50045,
  "title": "emdedding method for instance segmentation",
  "url": "/competitions/data-science-bowl-2018/discussion/50045",
  "author_name": "",
  "post_date": "2018-02-19T01:19:38.439208400Z",
  "votes": 3,
  "comment_count": 2,
  "views": 0,
  "content": "<p>anyone doing something like this? train a network to output:</p>\n\n<ul>\n<li><p>centers</p></li>\n<li><p>embedding of dim=N such that:  distance( embed(pixel1), embed(pixel2)) is small if pixel1 and pixel2 are of the same nuclei object</p></li>\n<li><p>using center pixels as seed, then cluster the rest of the pixels in the embedded space.</p></li>\n</ul>\n\n<p>as an example:</p>\n\n<p><a href=\"https://arxiv.org/pdf/1802.05591.pdf\">https://arxiv.org/pdf/1802.05591.pdf</a></p>",
  "messages": [
    {
      "id": "284979",
      "postDate": "02/19/2018 01:19:38",
      "content": "<p>anyone doing something like this? train a network to output:</p>\n\n<ul>\n<li><p>centers</p></li>\n<li><p>embedding of dim=N such that:  distance( embed(pixel1), embed(pixel2)) is small if pixel1 and pixel2 are of the same nuclei object</p></li>\n<li><p>using center pixels as seed, then cluster the rest of the pixels in the embedded space.</p></li>\n</ul>\n\n<p>as an example:</p>\n\n<p><a href=\"https://arxiv.org/pdf/1802.05591.pdf\">https://arxiv.org/pdf/1802.05591.pdf</a></p>",
      "rawMarkdown": "anyone doing something like this? train a network to output:\n\n- centers\n\n- embedding of dim=N such that:  distance( embed(pixel1), embed(pixel2)) is small if pixel1 and pixel2 are of the same nuclei object\n\n- using center pixels as seed, then cluster the rest of the pixels in the embedded space.\n\n\n\nas an example:\n\nhttps://arxiv.org/pdf/1802.05591.pdf",
      "votes": null
    },
    {
      "id": "285019",
      "postDate": "02/19/2018 04:37:05",
      "content": "<p>Good resourse.\nI also wondering</p>",
      "rawMarkdown": "Good resourse.\nI also wondering",
      "votes": null
    },
    {
      "id": "285906",
      "postDate": "02/21/2018 07:50:19",
      "content": "<p>I tried it. It seems to work well except in cases where there are many tiny objects near each other.</p>",
      "rawMarkdown": "I tried it. It seems to work well except in cases where there are many tiny objects near each other.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 285019,
      "author_name": "amxineohp",
      "author_url": "",
      "post_date": "02/19/2018 04:37:05",
      "content": "<p>Good resourse.\nI also wondering</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 285906,
      "author_name": "lolz0r",
      "author_url": "",
      "post_date": "02/21/2018 07:50:19",
      "content": "<p>I tried it. It seems to work well except in cases where there are many tiny objects near each other.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "284979": "anyone doing something like this? train a network to output:\n\n- centers\n\n- embedding of dim=N such that:  distance( embed(pixel1), embed(pixel2)) is small if pixel1 and pixel2 are of the same nuclei object\n\n- using center pixels as seed, then cluster the rest of the pixels in the embedded space.\n\n\n\nas an example:\n\nhttps://arxiv.org/pdf/1802.05591.pdf",
    "285019": "Good resourse.\nI also wondering",
    "285906": "I tried it. It seems to work well except in cases where there are many tiny objects near each other."
  },
  "source": "meta"
}