{
  "id": 52309,
  "title": "Mask RCNN pretrained weights.  coco vs imagenet vs ...",
  "url": "/competitions/data-science-bowl-2018/discussion/52309",
  "author_name": "",
  "post_date": "2018-03-18T17:27:25.417899600Z",
  "votes": 1,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Which pre-trained weights are most applicable to neclei segmentation (coco, imagenet, etc)?</p>",
  "messages": [
    {
      "id": "298068",
      "postDate": "03/18/2018 17:27:25",
      "content": "<p>Which pre-trained weights are most applicable to neclei segmentation (coco, imagenet, etc)?</p>",
      "rawMarkdown": "Which pre-trained weights are most applicable to neclei segmentation (coco, imagenet, etc)?",
      "votes": null
    },
    {
      "id": "298139",
      "postDate": "03/18/2018 22:48:12",
      "content": "<p>I used imagenet weights. If you crawl the discussion, though, there’s a split of using both—probably won’t impact your final solution too much, since high LB score will depend on ensembling and post processing.</p>\n\n<p>Use a well-trained model as a “tool” in your overall processing. </p>",
      "rawMarkdown": "I used imagenet weights. If you crawl the discussion, though, there’s a split of using both—probably won’t impact your final solution too much, since high LB score will depend on ensembling and post processing.\n\nUse a well-trained model as a “tool” in your overall processing.",
      "votes": null
    },
    {
      "id": "299033",
      "postDate": "03/20/2018 11:01:24",
      "content": "<p>Could you elaborate on what that means?</p>",
      "rawMarkdown": "Could you elaborate on what that means?",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 298139,
      "author_name": "spnrpa",
      "author_url": "",
      "post_date": "03/18/2018 22:48:12",
      "content": "<p>I used imagenet weights. If you crawl the discussion, though, there’s a split of using both—probably won’t impact your final solution too much, since high LB score will depend on ensembling and post processing.</p>\n\n<p>Use a well-trained model as a “tool” in your overall processing. </p>",
      "votes": null,
      "replies": [
        {
          "id": 299033,
          "author_name": "sushifan",
          "author_url": "",
          "post_date": "03/20/2018 11:01:24",
          "content": "<p>Could you elaborate on what that means?</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "298068": "Which pre-trained weights are most applicable to neclei segmentation (coco, imagenet, etc)?",
    "298139": "I used imagenet weights. If you crawl the discussion, though, there’s a split of using both—probably won’t impact your final solution too much, since high LB score will depend on ensembling and post processing.\n\nUse a well-trained model as a “tool” in your overall processing.",
    "299033": "Could you elaborate on what that means?"
  },
  "source": "meta"
}