{
  "id": 232175,
  "title": "Will the small network get the better result?",
  "url": "/competitions/hubmap-kidney-segmentation/discussion/232175",
  "author_name": "",
  "post_date": "2021-04-12T13:45:09.039804500Z",
  "votes": 3,
  "comment_count": 3,
  "views": 0,
  "content": "<p>I have tried unet with eb0~eb4, and the cv is eb2(0.928) &gt; eb4(0.926) &gt; eb1(0.925) &gt; eb0(0.924)&gt;eb3(0.923),and the eb2 can get the best LB,will the small network get the better result in this competition？<br>\n![<a href=\"https://drive.google.com/file/d/1FZn2tMoMYfDbKPFu3ltlAB2ysAxDMgl3/view?usp=sharing](url\" target=\"_blank\">https://drive.google.com/file/d/1FZn2tMoMYfDbKPFu3ltlAB2ysAxDMgl3/view?usp=sharing](url</a> to embed)</p>",
  "messages": [
    {
      "id": "1271335",
      "postDate": "04/12/2021 13:45:09",
      "content": "<p>I have tried unet with eb0~eb4, and the cv is eb2(0.928) &gt; eb4(0.926) &gt; eb1(0.925) &gt; eb0(0.924)&gt;eb3(0.923),and the eb2 can get the best LB,will the small network get the better result in this competition？<br>\n![<a href=\"https://drive.google.com/file/d/1FZn2tMoMYfDbKPFu3ltlAB2ysAxDMgl3/view?usp=sharing](url\" target=\"_blank\">https://drive.google.com/file/d/1FZn2tMoMYfDbKPFu3ltlAB2ysAxDMgl3/view?usp=sharing](url</a> to embed)</p>",
      "rawMarkdown": "I have tried unet with eb0~eb4, and the cv is eb2(0.928) > eb4(0.926) > eb1(0.925) > eb0(0.924)>eb3(0.923),and the eb2 can get the best LB,will the small network get the better result in this competition？\n![https://drive.google.com/file/d/1FZn2tMoMYfDbKPFu3ltlAB2ysAxDMgl3/view?usp=sharing](url to embed)",
      "votes": null
    },
    {
      "id": "1271480",
      "postDate": "04/12/2021 16:43:50",
      "content": "<p>For me eb0 is getting the better results, i'm still trying to improve eb2 but i don't think i will keep working on bigger models than eb2</p>",
      "rawMarkdown": "For me eb0 is getting the better results, i'm still trying to improve eb2 but i don't think i will keep working on bigger models than eb2",
      "votes": null
    },
    {
      "id": "1272354",
      "postDate": "04/13/2021 12:16:11",
      "content": "<p>I used b7 and performed well.</p>",
      "rawMarkdown": "I used b7 and performed well.",
      "votes": null
    },
    {
      "id": "1275631",
      "postDate": "04/16/2021 14:08:42",
      "content": "<p>I used se_resnext101, se_resnext50 and resnet34, their performance is similar,  the difference is about 0.001 or 0.002 on CV and LB, .</p>",
      "rawMarkdown": "I used se_resnext101, se_resnext50 and resnet34, their performance is similar,  the difference is about 0.001 or 0.002 on CV and LB, .",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1271480,
      "author_name": "victorasso",
      "author_url": "",
      "post_date": "04/12/2021 16:43:50",
      "content": "<p>For me eb0 is getting the better results, i'm still trying to improve eb2 but i don't think i will keep working on bigger models than eb2</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1272354,
      "author_name": "southsakura",
      "author_url": "",
      "post_date": "04/13/2021 12:16:11",
      "content": "<p>I used b7 and performed well.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1275631,
      "author_name": "hesene",
      "author_url": "",
      "post_date": "04/16/2021 14:08:42",
      "content": "<p>I used se_resnext101, se_resnext50 and resnet34, their performance is similar,  the difference is about 0.001 or 0.002 on CV and LB, .</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1271335": "I have tried unet with eb0~eb4, and the cv is eb2(0.928) > eb4(0.926) > eb1(0.925) > eb0(0.924)>eb3(0.923),and the eb2 can get the best LB,will the small network get the better result in this competition？\n![https://drive.google.com/file/d/1FZn2tMoMYfDbKPFu3ltlAB2ysAxDMgl3/view?usp=sharing](url to embed)",
    "1271480": "For me eb0 is getting the better results, i'm still trying to improve eb2 but i don't think i will keep working on bigger models than eb2",
    "1272354": "I used b7 and performed well.",
    "1275631": "I used se_resnext101, se_resnext50 and resnet34, their performance is similar,  the difference is about 0.001 or 0.002 on CV and LB, ."
  },
  "source": "meta"
}