{
  "id": 49436,
  "title": "Overlapping predictions",
  "url": "/competitions/data-science-bowl-2018/discussion/49436",
  "author_name": "",
  "post_date": "2018-02-11T09:50:52.995506Z",
  "votes": 1,
  "comment_count": 3,
  "views": 0,
  "content": "<p>Hi guys,</p>\n\n<p>How are you dealing with overlapping predictions?</p>\n\n<p>Now, when I trained my model (rcnn) and made predictions on a few test images I have overlapping predictions. Because of that, I merge the separate masks to a new \"united mask image\". But after the merging, I am sure I lose accuracy because those overlapping predictions become one united prediction.</p>\n\n<p>So my question is, how are you dealing with the overlappping predictions?</p>",
  "messages": [
    {
      "id": "280881",
      "postDate": "02/11/2018 09:50:52",
      "content": "<p>Hi guys,</p>\n\n<p>How are you dealing with overlapping predictions?</p>\n\n<p>Now, when I trained my model (rcnn) and made predictions on a few test images I have overlapping predictions. Because of that, I merge the separate masks to a new \"united mask image\". But after the merging, I am sure I lose accuracy because those overlapping predictions become one united prediction.</p>\n\n<p>So my question is, how are you dealing with the overlappping predictions?</p>",
      "rawMarkdown": "Hi guys,\n\nHow are you dealing with overlapping predictions?\n\nNow, when I trained my model (rcnn) and made predictions on a few test images I have overlapping predictions. Because of that, I merge the separate masks to a new \"united mask image\". But after the merging, I am sure I lose accuracy because those overlapping predictions become one united prediction.\n\nSo my question is, how are you dealing with the overlappping predictions?",
      "votes": null
    },
    {
      "id": "280962",
      "postDate": "02/11/2018 15:06:19",
      "content": "<p>I just sorted predicted masks from smaller to larger and removed duplicate pixels from the larger image. Not sure whether that's the best way to do it (or a good way to do it.) but it lets me create a solution I can submit.\nI did think about it deeply but seems like this is better than combining two masks since I'm guessing mispredicting pixels would happen a lot more often than mispredicting object.</p>",
      "rawMarkdown": "I just sorted predicted masks from smaller to larger and removed duplicate pixels from the larger image. Not sure whether that's the best way to do it (or a good way to do it.) but it lets me create a solution I can submit.\nI did think about it deeply but seems like this is better than combining two masks since I'm guessing mispredicting pixels would happen a lot more often than mispredicting object.",
      "votes": null
    },
    {
      "id": "280981",
      "postDate": "02/11/2018 16:28:23",
      "content": "<p>And have you compared this two \"naive\" methods?</p>",
      "rawMarkdown": "And have you compared this two \"naive\" methods?",
      "votes": null
    },
    {
      "id": "281106",
      "postDate": "02/12/2018 01:58:14",
      "content": "<p>Sometimes, model can potentially give better predictions with overlapping than human labeled data. But it can be an error in submission. Is this expected for organizer?</p>",
      "rawMarkdown": "Sometimes, model can potentially give better predictions with overlapping than human labeled data. But it can be an error in submission. Is this expected for organizer?",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 280962,
      "author_name": "bsp2020",
      "author_url": "",
      "post_date": "02/11/2018 15:06:19",
      "content": "<p>I just sorted predicted masks from smaller to larger and removed duplicate pixels from the larger image. Not sure whether that's the best way to do it (or a good way to do it.) but it lets me create a solution I can submit.\nI did think about it deeply but seems like this is better than combining two masks since I'm guessing mispredicting pixels would happen a lot more often than mispredicting object.</p>",
      "votes": null,
      "replies": [
        {
          "id": 280981,
          "author_name": "gaborvecsei",
          "author_url": "",
          "post_date": "02/11/2018 16:28:23",
          "content": "<p>And have you compared this two \"naive\" methods?</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 281106,
      "author_name": "akirasosa",
      "author_url": "",
      "post_date": "02/12/2018 01:58:14",
      "content": "<p>Sometimes, model can potentially give better predictions with overlapping than human labeled data. But it can be an error in submission. Is this expected for organizer?</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "280881": "Hi guys,\n\nHow are you dealing with overlapping predictions?\n\nNow, when I trained my model (rcnn) and made predictions on a few test images I have overlapping predictions. Because of that, I merge the separate masks to a new \"united mask image\". But after the merging, I am sure I lose accuracy because those overlapping predictions become one united prediction.\n\nSo my question is, how are you dealing with the overlappping predictions?",
    "280962": "I just sorted predicted masks from smaller to larger and removed duplicate pixels from the larger image. Not sure whether that's the best way to do it (or a good way to do it.) but it lets me create a solution I can submit.\nI did think about it deeply but seems like this is better than combining two masks since I'm guessing mispredicting pixels would happen a lot more often than mispredicting object.",
    "280981": "And have you compared this two \"naive\" methods?",
    "281106": "Sometimes, model can potentially give better predictions with overlapping than human labeled data. But it can be an error in submission. Is this expected for organizer?"
  },
  "source": "meta"
}