{
  "id": 341192,
  "title": "2nd place baseline solution ",
  "url": "/competitions/unifesp-x-ray-body-part-classifier/writeups/jgk-2nd-place-baseline-solution",
  "author_name": "",
  "post_date": "2022-08-01T18:01:51.726025700Z",
  "votes": 5,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Hi,All!<br>\nWe all did a good job the last few months, congratulations on the end of the competition.</p>\n<p>My baseline, SingleLabelMulticlass:</p>\n<ol>\n<li><p>FastAi version 2.0.Thanks, <a href=\"https://www.kaggle.com/jhoward\" target=\"_blank\">@jhoward</a> <br>\nit's really cool stuff!</p></li>\n<li><p>Albumentations.Of this I used:GaussNoise,ImageCompression,GaussianBlur,HorizontalFlip,RandomBrightnessContrast,OneOf([RandomBrightnessContrast(), FancyPCA(), HueSaturationValue()]),ToGray,ShiftScaleRotate.</p></li>\n</ol>\n<p>3.Normalizing the data \"Like ImageNet\".</p>\n<p>4.Callbacks: [EarlyStoppingCallback(monitor='valid_loss')], fastai.callback.fp16.</p>\n<p>5.Loss Function-tried different variants-the best was default.</p>\n<p>6.CNN and optimizer. 1submit:resnet50 adam optimizer. 2submit:resnet152 ranger optimizer (with learn.fit_flat_cos)</p>\n<p>7.Learning rate:default [learn.lr_find()]</p>\n<p>8.Batch size-tried different variants-the best was default (64)</p>\n<p>That's all,have a nice day!</p>",
  "messages": [
    {
      "id": "1880436",
      "postDate": "08/01/2022 18:01:51",
      "content": "<p>Hi,All!<br>\nWe all did a good job the last few months, congratulations on the end of the competition.</p>\n<p>My baseline, SingleLabelMulticlass:</p>\n<ol>\n<li><p>FastAi version 2.0.Thanks, <a href=\"https://www.kaggle.com/jhoward\" target=\"_blank\">@jhoward</a> <br>\nit's really cool stuff!</p></li>\n<li><p>Albumentations.Of this I used:GaussNoise,ImageCompression,GaussianBlur,HorizontalFlip,RandomBrightnessContrast,OneOf([RandomBrightnessContrast(), FancyPCA(), HueSaturationValue()]),ToGray,ShiftScaleRotate.</p></li>\n</ol>\n<p>3.Normalizing the data \"Like ImageNet\".</p>\n<p>4.Callbacks: [EarlyStoppingCallback(monitor='valid_loss')], fastai.callback.fp16.</p>\n<p>5.Loss Function-tried different variants-the best was default.</p>\n<p>6.CNN and optimizer. 1submit:resnet50 adam optimizer. 2submit:resnet152 ranger optimizer (with learn.fit_flat_cos)</p>\n<p>7.Learning rate:default [learn.lr_find()]</p>\n<p>8.Batch size-tried different variants-the best was default (64)</p>\n<p>That's all,have a nice day!</p>",
      "rawMarkdown": "Hi,All!\nWe all did a good job the last few months, congratulations on the end of the competition.\n\nMy baseline, SingleLabelMulticlass:\n\n1. FastAi version 2.0.Thanks, @jhoward \nit's really cool stuff!\n\n2. Albumentations.Of this I used:GaussNoise,ImageCompression,GaussianBlur,HorizontalFlip,RandomBrightnessContrast,OneOf([RandomBrightnessContrast(), FancyPCA(), HueSaturationValue()]),ToGray,ShiftScaleRotate.\n\n3.Normalizing the data \"Like ImageNet\".\n\n4.Callbacks: [EarlyStoppingCallback(monitor='valid_loss')], fastai.callback.fp16.\n\n5.Loss Function-tried different variants-the best was default.\n\n6.CNN and optimizer. 1submit:resnet50 adam optimizer. 2submit:resnet152 ranger optimizer (with learn.fit_flat_cos)\n\n7.Learning rate:default [learn.lr_find()]\n\n8.Batch size-tried different variants-the best was default (64)\n\nThat's all,have a nice day!",
      "votes": null
    },
    {
      "id": "1882323",
      "postDate": "08/03/2022 07:52:33",
      "content": "<p>Will you consider to publish the full notebook? Thanks.</p>",
      "rawMarkdown": "Will you consider to publish the full notebook? Thanks.",
      "votes": null
    },
    {
      "id": "2694767",
      "postDate": "03/13/2024 09:08:31",
      "content": "<p>What did you use as a base model? Pretrained? How'd you handle greyscale if so? </p>\n<ul>\n<li>Is your code available anywhere?</li>\n</ul>",
      "rawMarkdown": "What did you use as a base model? Pretrained? How'd you handle greyscale if so? \n+ Is your code available anywhere?",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1882323,
      "author_name": "alexanderyyy",
      "author_url": "",
      "post_date": "08/03/2022 07:52:33",
      "content": "<p>Will you consider to publish the full notebook? Thanks.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 2694767,
      "author_name": "danofer",
      "author_url": "",
      "post_date": "03/13/2024 09:08:31",
      "content": "<p>What did you use as a base model? Pretrained? How'd you handle greyscale if so? </p>\n<ul>\n<li>Is your code available anywhere?</li>\n</ul>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1880436": "Hi,All!\nWe all did a good job the last few months, congratulations on the end of the competition.\n\nMy baseline, SingleLabelMulticlass:\n\n1. FastAi version 2.0.Thanks, @jhoward \nit's really cool stuff!\n\n2. Albumentations.Of this I used:GaussNoise,ImageCompression,GaussianBlur,HorizontalFlip,RandomBrightnessContrast,OneOf([RandomBrightnessContrast(), FancyPCA(), HueSaturationValue()]),ToGray,ShiftScaleRotate.\n\n3.Normalizing the data \"Like ImageNet\".\n\n4.Callbacks: [EarlyStoppingCallback(monitor='valid_loss')], fastai.callback.fp16.\n\n5.Loss Function-tried different variants-the best was default.\n\n6.CNN and optimizer. 1submit:resnet50 adam optimizer. 2submit:resnet152 ranger optimizer (with learn.fit_flat_cos)\n\n7.Learning rate:default [learn.lr_find()]\n\n8.Batch size-tried different variants-the best was default (64)\n\nThat's all,have a nice day!",
    "1882323": "Will you consider to publish the full notebook? Thanks.",
    "2694767": "What did you use as a base model? Pretrained? How'd you handle greyscale if so? \n+ Is your code available anywhere?"
  },
  "source": "meta"
}