{
  "id": 92964,
  "title": "Evaluation Exception: The same pixel may not be assigned to two different objects.",
  "url": "/competitions/imaterialist-fashion-2019-FGVC6/discussion/92964",
  "author_name": "",
  "post_date": "2019-05-22T05:43:08.122641300Z",
  "votes": null,
  "comment_count": 3,
  "views": 0,
  "content": "<p>I encounter the above problem when I submit my result from one object detection model. \nBut I am confused that I have ensured that all the objects have a unique EncodedPixel with their corresponding ClassId \nSo what is the possible problem? \nI am new here. Any suggestion or question will be thankful.</p>\n\n<p>I am sorry that this question has been raised by others. But I don't know how to close this topic</p>",
  "messages": [
    {
      "id": "534954",
      "postDate": "05/22/2019 05:43:08",
      "content": "<p>I encounter the above problem when I submit my result from one object detection model. \nBut I am confused that I have ensured that all the objects have a unique EncodedPixel with their corresponding ClassId \nSo what is the possible problem? \nI am new here. Any suggestion or question will be thankful.</p>\n\n<p>I am sorry that this question has been raised by others. But I don't know how to close this topic</p>",
      "rawMarkdown": "I encounter the above problem when I submit my result from one object detection model. \nBut I am confused that I have ensured that all the objects have a unique EncodedPixel with their corresponding ClassId \nSo what is the possible problem? \nI am new here. Any suggestion or question will be thankful.\n\nI am sorry that this question has been raised by others. But I don't know how to close this topic",
      "votes": null
    },
    {
      "id": "536037",
      "postDate": "05/23/2019 21:11:13",
      "content": "<p><a href=\"/cinderellarobaker\">@cinderellarobaker</a> As far as I understand <em>unique EncodedPixel coding doesn't always mean that predicted masks are not overlapped</em>. You can check your submission to not have for the same classes overlapped masks - one of possible directions can be convert masks from <code>RLE</code> to <code>binary</code> encoding and check intersections.</p>",
      "rawMarkdown": "cinderellarobaker As far as I understand *unique EncodedPixel coding doesn't always mean that predicted masks are not overlapped*. You can check your submission to not have for the same classes overlapped masks - one of possible directions can be convert masks from `RLE` to `binary` encoding and check intersections.",
      "votes": null
    },
    {
      "id": "536167",
      "postDate": "05/24/2019 03:58:51",
      "content": "<p><a href=\"/alexgruzdev\">@alexgruzdev</a> is right. The unique EncodedPixel only applies to masks with the same ClassId.  </p>",
      "rawMarkdown": "alexgruzdev is right. The unique EncodedPixel only applies to masks with the same ClassId.",
      "votes": null
    },
    {
      "id": "537398",
      "postDate": "05/26/2019 23:58:21",
      "content": "<p><a href=\"/alexgruzdev\">@alexgruzdev</a> <a href=\"/makeitworkjml\">@makeitworkjml</a> Thanks for your clear explanation. it's helpful</p>",
      "rawMarkdown": "alexgruzdev @makeitworkjml Thanks for your clear explanation. it's helpful",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 536037,
      "author_name": "alexgruzdev",
      "author_url": "",
      "post_date": "05/23/2019 21:11:13",
      "content": "<p><a href=\"/cinderellarobaker\">@cinderellarobaker</a> As far as I understand <em>unique EncodedPixel coding doesn't always mean that predicted masks are not overlapped</em>. You can check your submission to not have for the same classes overlapped masks - one of possible directions can be convert masks from <code>RLE</code> to <code>binary</code> encoding and check intersections.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 536167,
      "author_name": "makeitworkjml",
      "author_url": "",
      "post_date": "05/24/2019 03:58:51",
      "content": "<p><a href=\"/alexgruzdev\">@alexgruzdev</a> is right. The unique EncodedPixel only applies to masks with the same ClassId.  </p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 537398,
      "author_name": "cinderellarobaker",
      "author_url": "",
      "post_date": "05/26/2019 23:58:21",
      "content": "<p><a href=\"/alexgruzdev\">@alexgruzdev</a> <a href=\"/makeitworkjml\">@makeitworkjml</a> Thanks for your clear explanation. it's helpful</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "534954": "I encounter the above problem when I submit my result from one object detection model. \nBut I am confused that I have ensured that all the objects have a unique EncodedPixel with their corresponding ClassId \nSo what is the possible problem? \nI am new here. Any suggestion or question will be thankful.\n\nI am sorry that this question has been raised by others. But I don't know how to close this topic",
    "536037": "cinderellarobaker As far as I understand *unique EncodedPixel coding doesn't always mean that predicted masks are not overlapped*. You can check your submission to not have for the same classes overlapped masks - one of possible directions can be convert masks from `RLE` to `binary` encoding and check intersections.",
    "536167": "alexgruzdev is right. The unique EncodedPixel only applies to masks with the same ClassId.",
    "537398": "alexgruzdev @makeitworkjml Thanks for your clear explanation. it's helpful"
  },
  "source": "meta"
}