{
  "id": 218489,
  "title": " Momentum Contrast | Unsupervised learning | Explore Further",
  "url": "/competitions/ranzcr-clip-catheter-line-classification/discussion/218489",
  "author_name": "",
  "post_date": "2021-02-10T21:09:45.125452300Z",
  "votes": 5,
  "comment_count": 1,
  "views": 0,
  "content": "<h3>Extracting the features(or visual) representations by using self-supervised training</h3>\n<p>Facebook-Research has worked on x-ray datasets to build a self-supervised learning method like MoCo and there is a recent paper on covid-19 <a href=\"https://arxiv.org/abs/2101.04909\" target=\"_blank\">paper</a>, which uses this method for prognosis. </p>\n<ul>\n<li>They have released pretrained models trained on DenseNet, I tried to use the pretrained model, here is <a href=\"https://www.kaggle.com/shivanandmn/self-supervise-momentum-contrast-great-idea?scriptVersionId=54013004\" target=\"_blank\">notebook</a>.</li>\n<li>Improvement ideas - train MoCo using ranzcr-dataset mostly on better performing like EfficientNet models, cross-validation and test-time augmentation. </li>\n</ul>\n<h6># Please provide your ideas to explore this further… 🎉</h6>",
  "messages": [
    {
      "id": "1195502",
      "postDate": "02/10/2021 21:09:45",
      "content": "<h3>Extracting the features(or visual) representations by using self-supervised training</h3>\n<p>Facebook-Research has worked on x-ray datasets to build a self-supervised learning method like MoCo and there is a recent paper on covid-19 <a href=\"https://arxiv.org/abs/2101.04909\" target=\"_blank\">paper</a>, which uses this method for prognosis. </p>\n<ul>\n<li>They have released pretrained models trained on DenseNet, I tried to use the pretrained model, here is <a href=\"https://www.kaggle.com/shivanandmn/self-supervise-momentum-contrast-great-idea?scriptVersionId=54013004\" target=\"_blank\">notebook</a>.</li>\n<li>Improvement ideas - train MoCo using ranzcr-dataset mostly on better performing like EfficientNet models, cross-validation and test-time augmentation. </li>\n</ul>\n<h6># Please provide your ideas to explore this further… 🎉</h6>",
      "rawMarkdown": "### Extracting the features(or visual) representations by using self-supervised training\nFacebook-Research has worked on x-ray datasets to build a self-supervised learning method like MoCo and there is a recent paper on covid-19 [paper](https://arxiv.org/abs/2101.04909), which uses this method for prognosis. \n - They have released pretrained models trained on DenseNet, I tried to use the pretrained model, here is [notebook](https://www.kaggle.com/shivanandmn/self-supervise-momentum-contrast-great-idea?scriptVersionId=54013004).\n - Improvement ideas - train MoCo using ranzcr-dataset mostly on better performing like EfficientNet models, cross-validation and test-time augmentation. \n####### Please provide your ideas to explore this further... 🎉",
      "votes": null
    },
    {
      "id": "1229597",
      "postDate": "03/07/2021 13:22:30",
      "content": "<p>Looks interesting Good work! 👍</p>",
      "rawMarkdown": "Looks interesting Good work! 👍",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1229597,
      "author_name": "anku5hk",
      "author_url": "",
      "post_date": "03/07/2021 13:22:30",
      "content": "<p>Looks interesting Good work! 👍</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1195502": "### Extracting the features(or visual) representations by using self-supervised training\nFacebook-Research has worked on x-ray datasets to build a self-supervised learning method like MoCo and there is a recent paper on covid-19 [paper](https://arxiv.org/abs/2101.04909), which uses this method for prognosis. \n - They have released pretrained models trained on DenseNet, I tried to use the pretrained model, here is [notebook](https://www.kaggle.com/shivanandmn/self-supervise-momentum-contrast-great-idea?scriptVersionId=54013004).\n - Improvement ideas - train MoCo using ranzcr-dataset mostly on better performing like EfficientNet models, cross-validation and test-time augmentation. \n####### Please provide your ideas to explore this further... 🎉",
    "1229597": "Looks interesting Good work! 👍"
  },
  "source": "meta"
}