{
  "id": 115379,
  "title": "dynamic window by predicting convolution weights at inference",
  "url": "/competitions/rsna-intracranial-hemorrhage-detection/discussion/115379",
  "author_name": "hengck23",
  "post_date": "2019-11-02T12:45:35.355000",
  "votes": 6,
  "comment_count": 3,
  "views": 0,
  "content": "<p>just an idea</p>\n\n<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F113660%2F5b4082f116ea4f8f8c204b008c9f6683%2FSelection_090.png?generation=1572698710842880&amp;alt=media\" alt=\"\">\n<img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F113660%2F9d15e00985b9eda6aeb6c337806cf8b8%2FSelection_089.png?generation=1572698729592327&amp;alt=media\" alt=\"\"></p>",
  "messages": [
    {
      "id": 663626,
      "postDate": "2019-11-02T12:45:35.357Z",
      "content": "<p>just an idea</p>\n\n<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F113660%2F5b4082f116ea4f8f8c204b008c9f6683%2FSelection_090.png?generation=1572698710842880&amp;alt=media\" alt=\"\">\n<img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F113660%2F9d15e00985b9eda6aeb6c337806cf8b8%2FSelection_089.png?generation=1572698729592327&amp;alt=media\" alt=\"\"></p>",
      "rawMarkdown": "just an idea\n\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F113660%2F5b4082f116ea4f8f8c204b008c9f6683%2FSelection_090.png?generation=1572698710842880&amp;alt=media)\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F113660%2F9d15e00985b9eda6aeb6c337806cf8b8%2FSelection_089.png?generation=1572698729592327&amp;alt=media)\n",
      "votes": 6
    },
    {
      "id": 663628,
      "postDate": "2019-11-02T12:46:23.603Z",
      "content": "<p>you can think of it as a LSTM to predict parameters for some attention layer in the network</p>",
      "rawMarkdown": "you can think of it as a LSTM to predict parameters for some attention layer in the network",
      "votes": 3
    },
    {
      "id": 664461,
      "postDate": "2019-11-03T17:46:19.703Z",
      "content": "<p>The paper referenced in the image should be this one: <a href=\"https://arxiv.org/abs/1901.10430\">https://arxiv.org/abs/1901.10430</a></p>",
      "rawMarkdown": "The paper referenced in the image should be this one: https://arxiv.org/abs/1901.10430",
      "votes": 1
    },
    {
      "id": 664191,
      "postDate": "2019-11-03T09:56:54.077Z",
      "content": "<p>Got a paper over dynamic convolution. Might be helpful to understand dynamic convolution first.\n<a href=\"https://arxiv.org/pdf/1807.03132.pdf\">https://arxiv.org/pdf/1807.03132.pdf</a></p>",
      "rawMarkdown": "Got a paper over dynamic convolution. Might be helpful to understand dynamic convolution first.\nhttps://arxiv.org/pdf/1807.03132.pdf\n"
    }
  ],
  "comments": [
    {
      "id": 663628,
      "author_name": "hengck23",
      "author_url": "",
      "post_date": "2019-11-02T12:46:23.603000",
      "content": "<p>you can think of it as a LSTM to predict parameters for some attention layer in the network</p>",
      "votes": 3,
      "replies": []
    },
    {
      "id": 664461,
      "author_name": "Mirco Nani",
      "author_url": "",
      "post_date": "2019-11-03T17:46:19.703000",
      "content": "<p>The paper referenced in the image should be this one: <a href=\"https://arxiv.org/abs/1901.10430\">https://arxiv.org/abs/1901.10430</a></p>",
      "votes": 1,
      "replies": []
    },
    {
      "id": 664191,
      "author_name": "NeerajSharma",
      "author_url": "",
      "post_date": "2019-11-03T09:56:54.077000",
      "content": "<p>Got a paper over dynamic convolution. Might be helpful to understand dynamic convolution first.\n<a href=\"https://arxiv.org/pdf/1807.03132.pdf\">https://arxiv.org/pdf/1807.03132.pdf</a></p>",
      "votes": 0,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "663626": "just an idea\n\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F113660%2F5b4082f116ea4f8f8c204b008c9f6683%2FSelection_090.png?generation=1572698710842880&amp;alt=media)\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F113660%2F9d15e00985b9eda6aeb6c337806cf8b8%2FSelection_089.png?generation=1572698729592327&amp;alt=media)\n",
    "663628": "you can think of it as a LSTM to predict parameters for some attention layer in the network",
    "664461": "The paper referenced in the image should be this one: https://arxiv.org/abs/1901.10430",
    "664191": "Got a paper over dynamic convolution. Might be helpful to understand dynamic convolution first.\nhttps://arxiv.org/pdf/1807.03132.pdf\n"
  }
}