{
  "id": 124486,
  "title": "Focal Loss for me :(",
  "url": "/competitions/pku-autonomous-driving/discussion/124486",
  "author_name": "",
  "post_date": "2020-01-04T11:48:25.237491800Z",
  "votes": 3,
  "comment_count": 5,
  "views": 0,
  "content": "<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F2034058%2Fad1547bf1e53a5bc5ec717498f8db1fe%2F81493304_491861998376548_2588493548538036224_n.jpg?generation=1578138486876114&amp;alt=media\" alt=\"\"></p>",
  "messages": [
    {
      "id": "710166",
      "postDate": "01/04/2020 11:48:25",
      "content": "<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F2034058%2Fad1547bf1e53a5bc5ec717498f8db1fe%2F81493304_491861998376548_2588493548538036224_n.jpg?generation=1578138486876114&amp;alt=media\" alt=\"\"></p>",
      "rawMarkdown": "![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F2034058%2Fad1547bf1e53a5bc5ec717498f8db1fe%2F81493304_491861998376548_2588493548538036224_n.jpg?generation=1578138486876114&amp;alt=media)",
      "votes": null
    },
    {
      "id": "710178",
      "postDate": "01/04/2020 12:12:56",
      "content": "<p>Haha.. pretty much the same as my situation.</p>",
      "rawMarkdown": "Haha.. pretty much the same as my situation.",
      "votes": null
    },
    {
      "id": "710181",
      "postDate": "01/04/2020 12:15:03",
      "content": "<p>I wonder it would be helpful if we add auxiliary heads for other labels except heatmap.</p>",
      "rawMarkdown": "I wonder it would be helpful if we add auxiliary heads for other labels except heatmap.",
      "votes": null
    },
    {
      "id": "710619",
      "postDate": "01/05/2020 01:38:57",
      "content": "<p>minimum loss may not guarantee to provide maximum competition metric. If you are calculating mAP after every epoch, that may give more realistic performance of the model.</p>",
      "rawMarkdown": "minimum loss may not guarantee to provide maximum competition metric. If you are calculating mAP after every epoch, that may give more realistic performance of the model.",
      "votes": null
    },
    {
      "id": "710690",
      "postDate": "01/05/2020 04:47:19",
      "content": "<p>best advice,thank you for your comment <a href=\"/sgalib\">@sgalib</a> </p>",
      "rawMarkdown": "best advice,thank you for your comment @sgalib",
      "votes": null
    },
    {
      "id": "712585",
      "postDate": "01/07/2020 12:06:03",
      "content": "<p>I also found that minimum loss doesn't equal to maximum competition metric.</p>",
      "rawMarkdown": "I also found that minimum loss doesn't equal to maximum competition metric.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 710178,
      "author_name": "xiejialun",
      "author_url": "",
      "post_date": "01/04/2020 12:12:56",
      "content": "<p>Haha.. pretty much the same as my situation.</p>",
      "votes": null,
      "replies": [
        {
          "id": 710181,
          "author_name": "xiejialun",
          "author_url": "",
          "post_date": "01/04/2020 12:15:03",
          "content": "<p>I wonder it would be helpful if we add auxiliary heads for other labels except heatmap.</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 710619,
      "author_name": "sgalib",
      "author_url": "",
      "post_date": "01/05/2020 01:38:57",
      "content": "<p>minimum loss may not guarantee to provide maximum competition metric. If you are calculating mAP after every epoch, that may give more realistic performance of the model.</p>",
      "votes": null,
      "replies": [
        {
          "id": 710690,
          "author_name": "mobassir",
          "author_url": "",
          "post_date": "01/05/2020 04:47:19",
          "content": "<p>best advice,thank you for your comment <a href=\"/sgalib\">@sgalib</a> </p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 712585,
      "author_name": "yudanqiu",
      "author_url": "",
      "post_date": "01/07/2020 12:06:03",
      "content": "<p>I also found that minimum loss doesn't equal to maximum competition metric.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "710166": "![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F2034058%2Fad1547bf1e53a5bc5ec717498f8db1fe%2F81493304_491861998376548_2588493548538036224_n.jpg?generation=1578138486876114&amp;alt=media)",
    "710178": "Haha.. pretty much the same as my situation.",
    "710181": "I wonder it would be helpful if we add auxiliary heads for other labels except heatmap.",
    "710619": "minimum loss may not guarantee to provide maximum competition metric. If you are calculating mAP after every epoch, that may give more realistic performance of the model.",
    "710690": "best advice,thank you for your comment @sgalib",
    "712585": "I also found that minimum loss doesn't equal to maximum competition metric."
  },
  "source": "meta"
}