{
  "id": 145140,
  "title": "Very simple fastai2 training baseline",
  "url": "/competitions/prostate-cancer-grade-assessment/discussion/145140",
  "author_name": "ilovescience",
  "post_date": "2020-04-22T01:48:04.896000",
  "votes": 10,
  "comment_count": 2,
  "views": 0,
  "content": "<p>I resized all the images (from third level of the multi-level tiff) to 256x256 (over <a href=\"https://www.kaggle.com/tanlikesmath/panda-challenge-resized-dataset\">here</a>) and trained a simple ResNet101 on these images. Even though the 256x256 images are terrible, I was surprised that the accuracy was 40%-50% and the Quadratically Weighted Kappa was 0.4-0.5. </p>\n\n<p>Check out my kernel <a href=\"https://www.kaggle.com/tanlikesmath/fastai2-training-baseline\">here</a>!</p>\n\n<p>I hope it's a useful baseline!</p>\n\n<p>Definitely, by improving the dataset, and using patch-based classification, the QWK would definitely improve.</p>",
  "messages": [
    {
      "id": 815969,
      "postDate": "2020-04-22T01:48:04.897Z",
      "content": "<p>I resized all the images (from third level of the multi-level tiff) to 256x256 (over <a href=\"https://www.kaggle.com/tanlikesmath/panda-challenge-resized-dataset\">here</a>) and trained a simple ResNet101 on these images. Even though the 256x256 images are terrible, I was surprised that the accuracy was 40%-50% and the Quadratically Weighted Kappa was 0.4-0.5. </p>\n\n<p>Check out my kernel <a href=\"https://www.kaggle.com/tanlikesmath/fastai2-training-baseline\">here</a>!</p>\n\n<p>I hope it's a useful baseline!</p>\n\n<p>Definitely, by improving the dataset, and using patch-based classification, the QWK would definitely improve.</p>",
      "rawMarkdown": "I resized all the images (from third level of the multi-level tiff) to 256x256 (over [here](https://www.kaggle.com/tanlikesmath/panda-challenge-resized-dataset)) and trained a simple ResNet101 on these images. Even though the 256x256 images are terrible, I was surprised that the accuracy was 40%-50% and the Quadratically Weighted Kappa was 0.4-0.5. \n\nCheck out my kernel [here](https://www.kaggle.com/tanlikesmath/fastai2-training-baseline)!\n\nI hope it's a useful baseline!\n\nDefinitely, by improving the dataset, and using patch-based classification, the QWK would definitely improve.",
      "votes": 10
    },
    {
      "id": 823289,
      "postDate": "2020-04-27T14:39:27.593Z",
      "content": "<p>Looks Good! GOOD JOB </p>",
      "rawMarkdown": "Looks Good! GOOD JOB "
    },
    {
      "id": 816213,
      "postDate": "2020-04-22T07:22:44.583Z",
      "content": "<p>Thanks for sharing :)</p>",
      "rawMarkdown": "Thanks for sharing :)"
    }
  ],
  "comments": [
    {
      "id": 823289,
      "author_name": "Sukanthen SS",
      "author_url": "",
      "post_date": "2020-04-27T14:39:27.593000",
      "content": "<p>Looks Good! GOOD JOB </p>",
      "votes": 0,
      "replies": []
    },
    {
      "id": 816213,
      "author_name": "Alberto Maria Falletta",
      "author_url": "",
      "post_date": "2020-04-22T07:22:44.583000",
      "content": "<p>Thanks for sharing :)</p>",
      "votes": 0,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "815969": "I resized all the images (from third level of the multi-level tiff) to 256x256 (over [here](https://www.kaggle.com/tanlikesmath/panda-challenge-resized-dataset)) and trained a simple ResNet101 on these images. Even though the 256x256 images are terrible, I was surprised that the accuracy was 40%-50% and the Quadratically Weighted Kappa was 0.4-0.5. \n\nCheck out my kernel [here](https://www.kaggle.com/tanlikesmath/fastai2-training-baseline)!\n\nI hope it's a useful baseline!\n\nDefinitely, by improving the dataset, and using patch-based classification, the QWK would definitely improve.",
    "823289": "Looks Good! GOOD JOB ",
    "816213": "Thanks for sharing :)"
  }
}