{
  "id": 171617,
  "title": "Why best loss can't get good lb",
  "url": "/competitions/global-wheat-detection/discussion/171617",
  "author_name": "",
  "post_date": "2020-08-01T18:55:43.867848600Z",
  "votes": 1,
  "comment_count": 4,
  "views": 0,
  "content": "<p>I m able to reach valid loss of 338 and train loss of 329 with eff det but unable to see score moving .\nAny suggestions would be of help.\nDo I need to keep high iou thold. ? </p>",
  "messages": [
    {
      "id": "954471",
      "postDate": "08/01/2020 18:55:43",
      "content": "<p>I m able to reach valid loss of 338 and train loss of 329 with eff det but unable to see score moving .\nAny suggestions would be of help.\nDo I need to keep high iou thold. ? </p>",
      "rawMarkdown": "I m able to reach valid loss of 338 and train loss of 329 with eff det but unable to see score moving .\nAny suggestions would be of help.\nDo I need to keep high iou thold. ?",
      "votes": null
    },
    {
      "id": "954557",
      "postDate": "08/01/2020 20:52:26",
      "content": "<p>Overfitting might be one of the reasons?</p>",
      "rawMarkdown": "Overfitting might be one of the reasons?",
      "votes": null
    },
    {
      "id": "954705",
      "postDate": "08/02/2020 02:32:42",
      "content": "<p>The test data is from a different site (distribution) as per the paper this is why pseudo labeling is improving the score significantly in this competition.</p>",
      "rawMarkdown": "The test data is from a different site (distribution) as per the paper this is why pseudo labeling is improving the score significantly in this competition.",
      "votes": null
    },
    {
      "id": "954773",
      "postDate": "08/02/2020 04:51:49",
      "content": "<p>Prevent model from overfiting. Add dropout layer, use earlystopping, etc</p>",
      "rawMarkdown": "Prevent model from overfiting. Add dropout layer, use earlystopping, etc",
      "votes": null
    },
    {
      "id": "954784",
      "postDate": "08/02/2020 05:14:12",
      "content": "<p><a href=\"/ma7moud\">@ma7moud</a> \n could changing iou could be of help ?\nWhat is purpose of it,to avoid duplicate BBxs or replacing more bbxs with single one ?</p>",
      "rawMarkdown": "ma7moud \n could changing iou could be of help ?\nWhat is purpose of it,to avoid duplicate BBxs or replacing more bbxs with single one ?",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 954557,
      "author_name": "jonykarki",
      "author_url": "",
      "post_date": "08/01/2020 20:52:26",
      "content": "<p>Overfitting might be one of the reasons?</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 954705,
      "author_name": "ma7moud",
      "author_url": "",
      "post_date": "08/02/2020 02:32:42",
      "content": "<p>The test data is from a different site (distribution) as per the paper this is why pseudo labeling is improving the score significantly in this competition.</p>",
      "votes": null,
      "replies": [
        {
          "id": 954784,
          "author_name": "jaideepvalani",
          "author_url": "",
          "post_date": "08/02/2020 05:14:12",
          "content": "<p><a href=\"/ma7moud\">@ma7moud</a> \n could changing iou could be of help ?\nWhat is purpose of it,to avoid duplicate BBxs or replacing more bbxs with single one ?</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 954773,
      "author_name": "philippurwokoap",
      "author_url": "",
      "post_date": "08/02/2020 04:51:49",
      "content": "<p>Prevent model from overfiting. Add dropout layer, use earlystopping, etc</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "954471": "I m able to reach valid loss of 338 and train loss of 329 with eff det but unable to see score moving .\nAny suggestions would be of help.\nDo I need to keep high iou thold. ?",
    "954557": "Overfitting might be one of the reasons?",
    "954705": "The test data is from a different site (distribution) as per the paper this is why pseudo labeling is improving the score significantly in this competition.",
    "954773": "Prevent model from overfiting. Add dropout layer, use earlystopping, etc",
    "954784": "ma7moud \n could changing iou could be of help ?\nWhat is purpose of it,to avoid duplicate BBxs or replacing more bbxs with single one ?"
  },
  "source": "meta"
}