{
  "id": 370508,
  "title": "mmclassification benchmark (LB=0.20)",
  "url": "/competitions/rsna-breast-cancer-detection/discussion/370508",
  "author_name": "takuoko",
  "post_date": "2022-12-04T23:39:24.973000",
  "votes": 17,
  "comment_count": 3,
  "views": 0,
  "content": "<p>I share a baseline using my favorite tool, openmmlab. If you are interested, you can use mmclassification for this competition. You can change the backbone and augmentation settings simply by changing the config.</p>\n<p>settings:</p>\n<ul>\n<li>mmclassification <a href=\"https://github.com/open-mmlab/mmclassification\" target=\"_blank\">https://github.com/open-mmlab/mmclassification</a></li>\n<li>train(9913 patients) / val(2000 patients)</li>\n<li>effnet-b3</li>\n<li>img size=224</li>\n<li>LB=0.20, val without thresholding(0.148), val with thresholding(0.299)</li>\n</ul>\n<p>train repo: <a href=\"https://github.com/okotaku/clshub/tree/main/configs/projects/rsna2022\" target=\"_blank\">https://github.com/okotaku/clshub/tree/main/configs/projects/rsna2022</a></p>\n<p>How to train:</p>\n<pre><code>1. prepare data based on README\n2. then run following commands\n\ndocker compose up -d clshub\ndocker compose exec clshub mim train mmcls configs/projects/rsna2022/efficientnet/efficientnet-b3_2xb8_rsna2022.py --gpus 2 --launcher pytorch\n</code></pre>\n<p>inference notebook: <a href=\"https://www.kaggle.com/code/takuok/fork-of-rsna-2022-baseline-effnetb3-a39288\" target=\"_blank\">https://www.kaggle.com/code/takuok/fork-of-rsna-2022-baseline-effnetb3-a39288</a></p>\n<p>other resources<br>\nsplit: <a href=\"https://www.kaggle.com/code/takuok/rsna2022-split-data?scriptVersionId=112610114\" target=\"_blank\">https://www.kaggle.com/code/takuok/rsna2022-split-data?scriptVersionId=112610114</a><br>\nweights: <a href=\"https://www.kaggle.com/datasets/takuok/rsna2022weights\" target=\"_blank\">https://www.kaggle.com/datasets/takuok/rsna2022weights</a><br>\nprepare tools: <a href=\"https://www.kaggle.com/code/takuok/pytorch-libs/notebook\" target=\"_blank\">https://www.kaggle.com/code/takuok/pytorch-libs/notebook</a><br>\n                        <a href=\"https://www.kaggle.com/code/takuok/rsna2022-git/notebook\" target=\"_blank\">https://www.kaggle.com/code/takuok/rsna2022-git/notebook</a><br>\n                        <a href=\"https://www.kaggle.com/code/takuok/fork-of-rsna2022-libs/notebook\" target=\"_blank\">https://www.kaggle.com/code/takuok/fork-of-rsna2022-libs/notebook</a><br>\n                        <a href=\"https://www.kaggle.com/code/takuok/rsna2022-libs/notebook\" target=\"_blank\">https://www.kaggle.com/code/takuok/rsna2022-libs/notebook</a></p>\n<h1>Acknowledge</h1>\n<p>I am a member of Z by HP Data Science Global Ambassadors. Special Thanks to Z by HP for sponsoring me a Z8G4 Workstation with dual A6000 GPU and a ZBook with RTX5000 GPU.</p>",
  "messages": [
    {
      "id": 2055252,
      "postDate": "2022-12-04T23:39:24.973Z",
      "content": "<p>I share a baseline using my favorite tool, openmmlab. If you are interested, you can use mmclassification for this competition. You can change the backbone and augmentation settings simply by changing the config.</p>\n<p>settings:</p>\n<ul>\n<li>mmclassification <a href=\"https://github.com/open-mmlab/mmclassification\" target=\"_blank\">https://github.com/open-mmlab/mmclassification</a></li>\n<li>train(9913 patients) / val(2000 patients)</li>\n<li>effnet-b3</li>\n<li>img size=224</li>\n<li>LB=0.20, val without thresholding(0.148), val with thresholding(0.299)</li>\n</ul>\n<p>train repo: <a href=\"https://github.com/okotaku/clshub/tree/main/configs/projects/rsna2022\" target=\"_blank\">https://github.com/okotaku/clshub/tree/main/configs/projects/rsna2022</a></p>\n<p>How to train:</p>\n<pre><code>1. prepare data based on README\n2. then run following commands\n\ndocker compose up -d clshub\ndocker compose exec clshub mim train mmcls configs/projects/rsna2022/efficientnet/efficientnet-b3_2xb8_rsna2022.py --gpus 2 --launcher pytorch\n</code></pre>\n<p>inference notebook: <a href=\"https://www.kaggle.com/code/takuok/fork-of-rsna-2022-baseline-effnetb3-a39288\" target=\"_blank\">https://www.kaggle.com/code/takuok/fork-of-rsna-2022-baseline-effnetb3-a39288</a></p>\n<p>other resources<br>\nsplit: <a href=\"https://www.kaggle.com/code/takuok/rsna2022-split-data?scriptVersionId=112610114\" target=\"_blank\">https://www.kaggle.com/code/takuok/rsna2022-split-data?scriptVersionId=112610114</a><br>\nweights: <a href=\"https://www.kaggle.com/datasets/takuok/rsna2022weights\" target=\"_blank\">https://www.kaggle.com/datasets/takuok/rsna2022weights</a><br>\nprepare tools: <a href=\"https://www.kaggle.com/code/takuok/pytorch-libs/notebook\" target=\"_blank\">https://www.kaggle.com/code/takuok/pytorch-libs/notebook</a><br>\n                        <a href=\"https://www.kaggle.com/code/takuok/rsna2022-git/notebook\" target=\"_blank\">https://www.kaggle.com/code/takuok/rsna2022-git/notebook</a><br>\n                        <a href=\"https://www.kaggle.com/code/takuok/fork-of-rsna2022-libs/notebook\" target=\"_blank\">https://www.kaggle.com/code/takuok/fork-of-rsna2022-libs/notebook</a><br>\n                        <a href=\"https://www.kaggle.com/code/takuok/rsna2022-libs/notebook\" target=\"_blank\">https://www.kaggle.com/code/takuok/rsna2022-libs/notebook</a></p>\n<h1>Acknowledge</h1>\n<p>I am a member of Z by HP Data Science Global Ambassadors. Special Thanks to Z by HP for sponsoring me a Z8G4 Workstation with dual A6000 GPU and a ZBook with RTX5000 GPU.</p>",
      "rawMarkdown": "I share a baseline using my favorite tool, openmmlab. If you are interested, you can use mmclassification for this competition. You can change the backbone and augmentation settings simply by changing the config.\n\nsettings:\n- mmclassification https://github.com/open-mmlab/mmclassification\n- train(9913 patients) / val(2000 patients)\n- effnet-b3\n- img size=224\n- LB=0.20, val without thresholding(0.148), val with thresholding(0.299)\n\ntrain repo: https://github.com/okotaku/clshub/tree/main/configs/projects/rsna2022\n\nHow to train:\n\n```\n1. prepare data based on README\n2. then run following commands\n\ndocker compose up -d clshub\ndocker compose exec clshub mim train mmcls configs/projects/rsna2022/efficientnet/efficientnet-b3_2xb8_rsna2022.py --gpus 2 --launcher pytorch\n```\n\ninference notebook: https://www.kaggle.com/code/takuok/fork-of-rsna-2022-baseline-effnetb3-a39288\n\nother resources\nsplit: https://www.kaggle.com/code/takuok/rsna2022-split-data?scriptVersionId=112610114\nweights: https://www.kaggle.com/datasets/takuok/rsna2022weights\nprepare tools: https://www.kaggle.com/code/takuok/pytorch-libs/notebook\n                        https://www.kaggle.com/code/takuok/rsna2022-git/notebook\n                        https://www.kaggle.com/code/takuok/fork-of-rsna2022-libs/notebook\n                        https://www.kaggle.com/code/takuok/rsna2022-libs/notebook\n\n# Acknowledge\nI am a member of Z by HP Data Science Global Ambassadors. Special Thanks to Z by HP for sponsoring me a Z8G4 Workstation with dual A6000 GPU and a ZBook with RTX5000 GPU.",
      "votes": 17
    },
    {
      "id": 2061418,
      "postDate": "2022-12-11T05:55:20.810Z",
      "content": "<p>Hi takuoko, as you said you use img size=224, but in code I saw self.image_size = 512?</p>",
      "rawMarkdown": "Hi takuoko, as you said you use img size=224, but in code I saw self.image_size = 512?",
      "replies": [
        {
          "id": 2064672,
          "postDate": "2022-12-14T02:50:09.347Z",
          "content": "<p><code>self.image_size = 512</code> is based on <a href=\"https://www.kaggle.com/datasets/tmyok1984/rsna2022-jpg-512\" target=\"_blank\">this dataset</a>.<br>\nMy flow is <code>original size -&gt; 512 -&gt; 224</code>.<br>\nImage size can be checked in <a href=\"https://github.com/okotaku/clshub/blob/main/configs/_base_/datasets/rsna2022/bs32_224.py#L24\" target=\"_blank\">config</a>.</p>",
          "rawMarkdown": "`self.image_size = 512` is based on [this dataset](https://www.kaggle.com/datasets/tmyok1984/rsna2022-jpg-512).\nMy flow is `original size -> 512 -> 224`.\nImage size can be checked in [config](https://github.com/okotaku/clshub/blob/main/configs/_base_/datasets/rsna2022/bs32_224.py#L24).",
          "votes": 1
        }
      ]
    },
    {
      "id": 2055318,
      "postDate": "2022-12-05T01:48:41.783Z",
      "content": "<p>thanks for sharing.  well informed.</p>",
      "rawMarkdown": "thanks for sharing.  well informed."
    }
  ],
  "comments": [
    {
      "id": 2061418,
      "author_name": "Mr.Fire",
      "author_url": "",
      "post_date": "2022-12-11T05:55:20.810000",
      "content": "<p>Hi takuoko, as you said you use img size=224, but in code I saw self.image_size = 512?</p>",
      "votes": 0,
      "replies": [
        {
          "id": 2064672,
          "author_name": "takuoko",
          "author_url": "",
          "post_date": "2022-12-14T02:50:09.347000",
          "content": "<p><code>self.image_size = 512</code> is based on <a href=\"https://www.kaggle.com/datasets/tmyok1984/rsna2022-jpg-512\" target=\"_blank\">this dataset</a>.<br>\nMy flow is <code>original size -&gt; 512 -&gt; 224</code>.<br>\nImage size can be checked in <a href=\"https://github.com/okotaku/clshub/blob/main/configs/_base_/datasets/rsna2022/bs32_224.py#L24\" target=\"_blank\">config</a>.</p>",
          "votes": 1,
          "replies": []
        }
      ]
    },
    {
      "id": 2055318,
      "author_name": "dragon zhang",
      "author_url": "",
      "post_date": "2022-12-05T01:48:41.783000",
      "content": "<p>thanks for sharing.  well informed.</p>",
      "votes": 0,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "2055252": "I share a baseline using my favorite tool, openmmlab. If you are interested, you can use mmclassification for this competition. You can change the backbone and augmentation settings simply by changing the config.\n\nsettings:\n- mmclassification https://github.com/open-mmlab/mmclassification\n- train(9913 patients) / val(2000 patients)\n- effnet-b3\n- img size=224\n- LB=0.20, val without thresholding(0.148), val with thresholding(0.299)\n\ntrain repo: https://github.com/okotaku/clshub/tree/main/configs/projects/rsna2022\n\nHow to train:\n\n```\n1. prepare data based on README\n2. then run following commands\n\ndocker compose up -d clshub\ndocker compose exec clshub mim train mmcls configs/projects/rsna2022/efficientnet/efficientnet-b3_2xb8_rsna2022.py --gpus 2 --launcher pytorch\n```\n\ninference notebook: https://www.kaggle.com/code/takuok/fork-of-rsna-2022-baseline-effnetb3-a39288\n\nother resources\nsplit: https://www.kaggle.com/code/takuok/rsna2022-split-data?scriptVersionId=112610114\nweights: https://www.kaggle.com/datasets/takuok/rsna2022weights\nprepare tools: https://www.kaggle.com/code/takuok/pytorch-libs/notebook\n                        https://www.kaggle.com/code/takuok/rsna2022-git/notebook\n                        https://www.kaggle.com/code/takuok/fork-of-rsna2022-libs/notebook\n                        https://www.kaggle.com/code/takuok/rsna2022-libs/notebook\n\n# Acknowledge\nI am a member of Z by HP Data Science Global Ambassadors. Special Thanks to Z by HP for sponsoring me a Z8G4 Workstation with dual A6000 GPU and a ZBook with RTX5000 GPU.",
    "2061418": "Hi takuoko, as you said you use img size=224, but in code I saw self.image_size = 512?",
    "2055318": "thanks for sharing.  well informed."
  }
}