{
  "id": 48932,
  "title": "does perfect segmentation in one mask will give me 100% LB?",
  "url": "/competitions/data-science-bowl-2018/discussion/48932",
  "author_name": "",
  "post_date": "2018-02-04T18:23:48.733901400Z",
  "votes": 1,
  "comment_count": 2,
  "views": 0,
  "content": "<p>my question regards to the pixel extract evaluation and the nucleis that connected to eavh other. \nif i understands right. its not enough to segment a perfect image because i wont extract them right.\nam i correct?</p>",
  "messages": [
    {
      "id": "277887",
      "postDate": "02/04/2018 18:23:48",
      "content": "<p>my question regards to the pixel extract evaluation and the nucleis that connected to eavh other. \nif i understands right. its not enough to segment a perfect image because i wont extract them right.\nam i correct?</p>",
      "rawMarkdown": "my question regards to the pixel extract evaluation and the nucleis that connected to eavh other. \nif i understands right. its not enough to segment a perfect image because i wont extract them right.\nam i correct?",
      "votes": null
    },
    {
      "id": "277888",
      "postDate": "02/04/2018 18:29:50",
      "content": "<p>unless someone will create rle_encoding better that can recognize the diff nucleis. \ni think the rle_encoding change the approach of the game. maybe we dont need to do mask rcnn and bbox just encode better</p>",
      "rawMarkdown": "unless someone will create rle_encoding better that can recognize the diff nucleis. \ni think the rle_encoding change the approach of the game. maybe we dont need to do mask rcnn and bbox just encode better",
      "votes": null
    },
    {
      "id": "277943",
      "postDate": "02/04/2018 21:38:12",
      "content": "<p>Yes. But I don’t think the encoding or decoding matters outside of the one label per pixel requirement. However, I think post-processing is critical. Especially if you’re using an RCNN variant. You’ll really need to think carefully about non-maximum suppression, mask refinement, and how to classify overlapping pixels (i.e. it might be more complicated than argmax).</p>",
      "rawMarkdown": "Yes. But I don’t think the encoding or decoding matters outside of the one label per pixel requirement. However, I think post-processing is critical. Especially if you’re using an RCNN variant. You’ll really need to think carefully about non-maximum suppression, mask refinement, and how to classify overlapping pixels (i.e. it might be more complicated than argmax).",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 277888,
      "author_name": "aviwolfson",
      "author_url": "",
      "post_date": "02/04/2018 18:29:50",
      "content": "<p>unless someone will create rle_encoding better that can recognize the diff nucleis. \ni think the rle_encoding change the approach of the game. maybe we dont need to do mask rcnn and bbox just encode better</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 277943,
      "author_name": "agoodman",
      "author_url": "",
      "post_date": "02/04/2018 21:38:12",
      "content": "<p>Yes. But I don’t think the encoding or decoding matters outside of the one label per pixel requirement. However, I think post-processing is critical. Especially if you’re using an RCNN variant. You’ll really need to think carefully about non-maximum suppression, mask refinement, and how to classify overlapping pixels (i.e. it might be more complicated than argmax).</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "277887": "my question regards to the pixel extract evaluation and the nucleis that connected to eavh other. \nif i understands right. its not enough to segment a perfect image because i wont extract them right.\nam i correct?",
    "277888": "unless someone will create rle_encoding better that can recognize the diff nucleis. \ni think the rle_encoding change the approach of the game. maybe we dont need to do mask rcnn and bbox just encode better",
    "277943": "Yes. But I don’t think the encoding or decoding matters outside of the one label per pixel requirement. However, I think post-processing is critical. Especially if you’re using an RCNN variant. You’ll really need to think carefully about non-maximum suppression, mask refinement, and how to classify overlapping pixels (i.e. it might be more complicated than argmax)."
  },
  "source": "meta"
}