{
  "id": 123026,
  "title": "Submission file question on loss",
  "url": "/competitions/deepfake-detection-challenge/discussion/123026",
  "author_name": "",
  "post_date": "2019-12-24T07:59:20.441925300Z",
  "votes": null,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Still a newbie so excuse me if this is an obvious question.\nSay a video is fake, and we detect 200 of 300 fake frames, 100 we miss.\nDo we submit a 1, saying we are 100% confident it is fake, or do we submit a 200/300 which is 0.666666 fake?</p>\n\n<p>If 100% confident the video is real, do we submit 0, or follow same logic as above.\nIf we submit a simple binary 1,0 will this penalise me?</p>",
  "messages": [
    {
      "id": "702055",
      "postDate": "12/24/2019 07:59:20",
      "content": "<p>Still a newbie so excuse me if this is an obvious question.\nSay a video is fake, and we detect 200 of 300 fake frames, 100 we miss.\nDo we submit a 1, saying we are 100% confident it is fake, or do we submit a 200/300 which is 0.666666 fake?</p>\n\n<p>If 100% confident the video is real, do we submit 0, or follow same logic as above.\nIf we submit a simple binary 1,0 will this penalise me?</p>",
      "rawMarkdown": "Still a newbie so excuse me if this is an obvious question.\nSay a video is fake, and we detect 200 of 300 fake frames, 100 we miss.\nDo we submit a 1, saying we are 100% confident it is fake, or do we submit a 200/300 which is 0.666666 fake?\n\nIf 100% confident the video is real, do we submit 0, or follow same logic as above.\nIf we submit a simple binary 1,0 will this penalise me?",
      "votes": null
    },
    {
      "id": "702060",
      "postDate": "12/24/2019 08:06:06",
      "content": "<p>We are trying to predict if an entire video contains any altered frames or audio so we don't need to make predictions at the frame level but rather across the whole video clip. </p>\n\n<p>You can output simple 1, 0 predictions but because of how the metric is formed if you guess 1 and the answer is really zero you will be massively penalized for a confident, but incorrect answer, instead what you should output is something like the probability that you think a video is real or fake. </p>",
      "rawMarkdown": "We are trying to predict if an entire video contains any altered frames or audio so we don't need to make predictions at the frame level but rather across the whole video clip. \n\nYou can output simple 1, 0 predictions but because of how the metric is formed if you guess 1 and the answer is really zero you will be massively penalized for a confident, but incorrect answer, instead what you should output is something like the probability that you think a video is real or fake.",
      "votes": null
    },
    {
      "id": "702074",
      "postDate": "12/24/2019 08:20:02",
      "content": "<p>Thanks for a clear explanation.</p>",
      "rawMarkdown": "Thanks for a clear explanation.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 702060,
      "author_name": "ryches",
      "author_url": "",
      "post_date": "12/24/2019 08:06:06",
      "content": "<p>We are trying to predict if an entire video contains any altered frames or audio so we don't need to make predictions at the frame level but rather across the whole video clip. </p>\n\n<p>You can output simple 1, 0 predictions but because of how the metric is formed if you guess 1 and the answer is really zero you will be massively penalized for a confident, but incorrect answer, instead what you should output is something like the probability that you think a video is real or fake. </p>",
      "votes": null,
      "replies": [
        {
          "id": 702074,
          "author_name": "gideonvos",
          "author_url": "",
          "post_date": "12/24/2019 08:20:02",
          "content": "<p>Thanks for a clear explanation.</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "702055": "Still a newbie so excuse me if this is an obvious question.\nSay a video is fake, and we detect 200 of 300 fake frames, 100 we miss.\nDo we submit a 1, saying we are 100% confident it is fake, or do we submit a 200/300 which is 0.666666 fake?\n\nIf 100% confident the video is real, do we submit 0, or follow same logic as above.\nIf we submit a simple binary 1,0 will this penalise me?",
    "702060": "We are trying to predict if an entire video contains any altered frames or audio so we don't need to make predictions at the frame level but rather across the whole video clip. \n\nYou can output simple 1, 0 predictions but because of how the metric is formed if you guess 1 and the answer is really zero you will be massively penalized for a confident, but incorrect answer, instead what you should output is something like the probability that you think a video is real or fake.",
    "702074": "Thanks for a clear explanation."
  },
  "source": "meta"
}