{
  "id": 132666,
  "title": "Frame-by-frame training ?",
  "url": "/competitions/deepfake-detection-challenge/discussion/132666",
  "author_name": "ColdMirror",
  "post_date": "2020-02-27T07:13:10.500000",
  "votes": 1,
  "comment_count": 2,
  "views": 0,
  "content": "<p>In the training stage, I extracted 20 frames from one video, and 'stack' them to train. (eg. input: [20*3, 224, 224], label: 1) \nOne video has one label, is it correct? And the speed is very slow. (About 5h at 1 1080ti)\nOne frame have one label or One video one label?👀 </p>",
  "messages": [
    {
      "id": 757851,
      "postDate": "2020-02-27T07:13:10.500Z",
      "content": "<p>In the training stage, I extracted 20 frames from one video, and 'stack' them to train. (eg. input: [20*3, 224, 224], label: 1) \nOne video has one label, is it correct? And the speed is very slow. (About 5h at 1 1080ti)\nOne frame have one label or One video one label?👀 </p>",
      "rawMarkdown": "In the training stage, I extracted 20 frames from one video, and 'stack' them to train. (eg. input: [20*3, 224, 224], label: 1) \nOne video has one label, is it correct? And the speed is very slow. (About 5h at 1 1080ti)\nOne frame have one label or One video one label?👀 ",
      "votes": 1
    },
    {
      "id": 759532,
      "postDate": "2020-02-29T06:43:18.103Z",
      "content": "<p>Apparently, the labels in metadata correspond to videos, in your words, one video one label. <br>\nHowever, how use choose to use this label totally depends on your choice, and it might not be correct to assign the label of the whole video to every-single-frame in the video.</p>",
      "rawMarkdown": "Apparently, the labels in metadata correspond to videos, in your words, one video one label.  \nHowever, how use choose to use this label totally depends on your choice, and it might not be correct to assign the label of the whole video to every-single-frame in the video."
    },
    {
      "id": 757882,
      "postDate": "2020-02-27T07:58:35.513Z",
      "rawMarkdown": "",
      "isDeleted": true
    }
  ],
  "comments": [
    {
      "id": 759532,
      "author_name": "cmhzc",
      "author_url": "",
      "post_date": "2020-02-29T06:43:18.103000",
      "content": "<p>Apparently, the labels in metadata correspond to videos, in your words, one video one label. <br>\nHowever, how use choose to use this label totally depends on your choice, and it might not be correct to assign the label of the whole video to every-single-frame in the video.</p>",
      "votes": 0,
      "replies": []
    },
    {
      "id": 757882,
      "author_name": "",
      "author_url": "",
      "post_date": "2020-02-27T07:58:35.513000",
      "content": "",
      "votes": 0,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "757851": "In the training stage, I extracted 20 frames from one video, and 'stack' them to train. (eg. input: [20*3, 224, 224], label: 1) \nOne video has one label, is it correct? And the speed is very slow. (About 5h at 1 1080ti)\nOne frame have one label or One video one label?👀 ",
    "759532": "Apparently, the labels in metadata correspond to videos, in your words, one video one label.  \nHowever, how use choose to use this label totally depends on your choice, and it might not be correct to assign the label of the whole video to every-single-frame in the video.",
    "757882": ""
  }
}