{
  "id": 145412,
  "title": "Any thoughts on removing the extra 100 images in the training set that do not have masks?",
  "url": "/competitions/prostate-cancer-grade-assessment/discussion/145412",
  "author_name": "gao-hongnan",
  "post_date": "2020-04-23T04:07:55.027000",
  "votes": -1,
  "comment_count": 3,
  "views": 0,
  "content": "<p>The organizer did mentioned that : <strong>Not all training images have label masks.</strong></p>\n\n<p>So I did a rough <a href=\"https://www.kaggle.com/reighns/sanity-check-on-original-vs-masked-images\">sanity check</a> here and there seems to be 100 images that do not have corresponding masks annotated.  Is it a good idea to remove those 100 images?</p>",
  "messages": [
    {
      "id": 817972,
      "postDate": "2020-04-23T15:11:43.527Z",
      "content": "<p><a href=\"/reighns\">@reighns</a> Have you seen <a href=\"https://www.kaggle.com/c/prostate-cancer-grade-assessment/discussion/145534\">SlidSeg</a> before? Could be useful on this front. </p>",
      "rawMarkdown": "@reighns Have you seen [SlidSeg](https://www.kaggle.com/c/prostate-cancer-grade-assessment/discussion/145534) before? Could be useful on this front. ",
      "votes": 1,
      "replies": [
        {
          "id": 818851,
          "postDate": "2020-04-24T06:44:40.553Z",
          "content": "<p><a href=\"/dannellyz\">@dannellyz</a> good one, thanks</p>",
          "rawMarkdown": "@dannellyz good one, thanks"
        }
      ]
    },
    {
      "id": 817472,
      "postDate": "2020-04-23T07:26:44.177Z",
      "content": "<p>I think remove them at the beginning might enough to provide you a decent score. But we could still use confident model which trained on data with mask to generate the corresponding label for them and treat them as new training data to new model.</p>",
      "rawMarkdown": "I think remove them at the beginning might enough to provide you a decent score. But we could still use confident model which trained on data with mask to generate the corresponding label for them and treat them as new training data to new model.",
      "votes": 1
    },
    {
      "id": 817317,
      "postDate": "2020-04-23T04:07:55.027Z",
      "content": "<p>The organizer did mentioned that : <strong>Not all training images have label masks.</strong></p>\n\n<p>So I did a rough <a href=\"https://www.kaggle.com/reighns/sanity-check-on-original-vs-masked-images\">sanity check</a> here and there seems to be 100 images that do not have corresponding masks annotated.  Is it a good idea to remove those 100 images?</p>",
      "rawMarkdown": "The organizer did mentioned that : **Not all training images have label masks.**\n\nSo I did a rough [sanity check](https://www.kaggle.com/reighns/sanity-check-on-original-vs-masked-images) here and there seems to be 100 images that do not have corresponding masks annotated.  Is it a good idea to remove those 100 images?",
      "votes": -1
    }
  ],
  "comments": [
    {
      "id": 817972,
      "author_name": "Zac Dannelly",
      "author_url": "",
      "post_date": "2020-04-23T15:11:43.527000",
      "content": "<p><a href=\"/reighns\">@reighns</a> Have you seen <a href=\"https://www.kaggle.com/c/prostate-cancer-grade-assessment/discussion/145534\">SlidSeg</a> before? Could be useful on this front. </p>",
      "votes": 1,
      "replies": [
        {
          "id": 818851,
          "author_name": "gao-hongnan",
          "author_url": "",
          "post_date": "2020-04-24T06:44:40.553000",
          "content": "<p><a href=\"/dannellyz\">@dannellyz</a> good one, thanks</p>",
          "votes": 0,
          "replies": []
        }
      ]
    },
    {
      "id": 817472,
      "author_name": "Tsai29",
      "author_url": "",
      "post_date": "2020-04-23T07:26:44.177000",
      "content": "<p>I think remove them at the beginning might enough to provide you a decent score. But we could still use confident model which trained on data with mask to generate the corresponding label for them and treat them as new training data to new model.</p>",
      "votes": 1,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "817972": "@reighns Have you seen [SlidSeg](https://www.kaggle.com/c/prostate-cancer-grade-assessment/discussion/145534) before? Could be useful on this front. ",
    "817472": "I think remove them at the beginning might enough to provide you a decent score. But we could still use confident model which trained on data with mask to generate the corresponding label for them and treat them as new training data to new model.",
    "817317": "The organizer did mentioned that : **Not all training images have label masks.**\n\nSo I did a rough [sanity check](https://www.kaggle.com/reighns/sanity-check-on-original-vs-masked-images) here and there seems to be 100 images that do not have corresponding masks annotated.  Is it a good idea to remove those 100 images?"
  }
}