{
  "id": 284999,
  "title": "why x101  is not as good as R-50 in detectron2?",
  "url": "/competitions/sartorius-cell-instance-segmentation/discussion/284999",
  "author_name": "",
  "post_date": "2021-11-03T05:40:33.445888600Z",
  "votes": 3,
  "comment_count": 4,
  "views": 0,
  "content": "<p>x101 model has less loss but worse LB?</p>",
  "messages": [
    {
      "id": "1568885",
      "postDate": "11/03/2021 05:40:33",
      "content": "<p>x101 model has less loss but worse LB?</p>",
      "rawMarkdown": "x101 model has less loss but worse LB?",
      "votes": null
    },
    {
      "id": "1569372",
      "postDate": "11/03/2021 11:42:44",
      "content": "<p>Hi! I also wonder why, have you get any idea?</p>",
      "rawMarkdown": "Hi! I also wonder why, have you get any idea?",
      "votes": null
    },
    {
      "id": "1569418",
      "postDate": "11/03/2021 12:26:33",
      "content": "<p>perhaps  need classifier first?  three cell types , loss not consistent.</p>",
      "rawMarkdown": "perhaps  need classifier first?  three cell types , loss not consistent.",
      "votes": null
    },
    {
      "id": "1598375",
      "postDate": "11/28/2021 12:54:01",
      "content": "<p>In my case, with almost the same hyperparameters, x101 gave better results than r50.</p>",
      "rawMarkdown": "In my case, with almost the same hyperparameters, x101 gave better results than r50.",
      "votes": null
    },
    {
      "id": "1600023",
      "postDate": "11/30/2021 03:58:15",
      "content": "<p>thx for your reply.  when I posted this,  I got x101 with better loss but lower LB without post processing.</p>\n<p>since then, due to GPU limit, I have not compared the two model by LB. </p>\n<p>may I should try again.</p>",
      "rawMarkdown": "thx for your reply.  when I posted this,  I got x101 with better loss but lower LB without post processing.\n\nsince then, due to GPU limit, I have not compared the two model by LB. \n\nmay I should try again.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1569372,
      "author_name": "xhanxux",
      "author_url": "",
      "post_date": "11/03/2021 11:42:44",
      "content": "<p>Hi! I also wonder why, have you get any idea?</p>",
      "votes": null,
      "replies": [
        {
          "id": 1569418,
          "author_name": "dragonzhang",
          "author_url": "",
          "post_date": "11/03/2021 12:26:33",
          "content": "<p>perhaps  need classifier first?  three cell types , loss not consistent.</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 1598375,
      "author_name": "yosukeyama",
      "author_url": "",
      "post_date": "11/28/2021 12:54:01",
      "content": "<p>In my case, with almost the same hyperparameters, x101 gave better results than r50.</p>",
      "votes": null,
      "replies": [
        {
          "id": 1600023,
          "author_name": "dragonzhang",
          "author_url": "",
          "post_date": "11/30/2021 03:58:15",
          "content": "<p>thx for your reply.  when I posted this,  I got x101 with better loss but lower LB without post processing.</p>\n<p>since then, due to GPU limit, I have not compared the two model by LB. </p>\n<p>may I should try again.</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "1568885": "x101 model has less loss but worse LB?",
    "1569372": "Hi! I also wonder why, have you get any idea?",
    "1569418": "perhaps  need classifier first?  three cell types , loss not consistent.",
    "1598375": "In my case, with almost the same hyperparameters, x101 gave better results than r50.",
    "1600023": "thx for your reply.  when I posted this,  I got x101 with better loss but lower LB without post processing.\n\nsince then, due to GPU limit, I have not compared the two model by LB. \n\nmay I should try again."
  },
  "source": "meta"
}