{
  "id": 128413,
  "title": "what is the Best LB score using only Face crops and no audio analysis",
  "url": "/competitions/deepfake-detection-challenge/discussion/128413",
  "author_name": "",
  "post_date": "2020-01-31T06:05:38.641133Z",
  "votes": 2,
  "comment_count": 6,
  "views": 0,
  "content": "",
  "messages": [
    {
      "id": "733440",
      "postDate": "01/31/2020 06:05:38",
      "content": "",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "734132",
      "postDate": "02/01/2020 00:27:36",
      "content": "<p>Our best score without audio:\n0.44276\nBTW I don't think using audio would boost score. Most videos' audio are not manipulated. </p>",
      "rawMarkdown": "Our best score without audio:\n0.44276\nBTW I don't think using audio would boost score. Most videos' audio are not manipulated.",
      "votes": null
    },
    {
      "id": "734159",
      "postDate": "02/01/2020 03:04:22",
      "content": "<p>Thank you very much for sharing your score and inputs.</p>",
      "rawMarkdown": "Thank you very much for sharing your score and inputs.",
      "votes": null
    },
    {
      "id": "734224",
      "postDate": "02/01/2020 05:44:42",
      "content": "<p>But we don't know if it is true about test videos. </p>",
      "rawMarkdown": "But we don't know if it is true about test videos.",
      "votes": null
    },
    {
      "id": "734702",
      "postDate": "02/01/2020 21:12:16",
      "content": "<p>I'm actually very interested in knowing how to use the audio for this dataset because not all deepfakes would have modded audio. Therefore ground truth for audio is quite challenging to create. This will indirectly lead to a lot of noise in the model being built.</p>",
      "rawMarkdown": "I'm actually very interested in knowing how to use the audio for this dataset because not all deepfakes would have modded audio. Therefore ground truth for audio is quite challenging to create. This will indirectly lead to a lot of noise in the model being built.",
      "votes": null
    },
    {
      "id": "734707",
      "postDate": "02/01/2020 21:23:53",
      "content": "<p><a href=\"/akashnandi\">@akashnandi</a> here is a dataset already pre-calculated the ground truth for audio. <a href=\"https://www.kaggle.com/basharallabadi/dfdc-video-audio-labels\">link</a></p>",
      "rawMarkdown": "akashnandi here is a dataset already pre-calculated the ground truth for audio. [link](https://www.kaggle.com/basharallabadi/dfdc-video-audio-labels)",
      "votes": null
    },
    {
      "id": "734834",
      "postDate": "02/02/2020 04:14:12",
      "content": "<p>Thanks, I'll have a look.</p>",
      "rawMarkdown": "Thanks, I'll have a look.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 734132,
      "author_name": "unkownhihi",
      "author_url": "",
      "post_date": "02/01/2020 00:27:36",
      "content": "<p>Our best score without audio:\n0.44276\nBTW I don't think using audio would boost score. Most videos' audio are not manipulated. </p>",
      "votes": null,
      "replies": [
        {
          "id": 734159,
          "author_name": "reachkishore",
          "author_url": "",
          "post_date": "02/01/2020 03:04:22",
          "content": "<p>Thank you very much for sharing your score and inputs.</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 734224,
          "author_name": "ankitsainiankit",
          "author_url": "",
          "post_date": "02/01/2020 05:44:42",
          "content": "<p>But we don't know if it is true about test videos. </p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 734702,
          "author_name": "akashnandi",
          "author_url": "",
          "post_date": "02/01/2020 21:12:16",
          "content": "<p>I'm actually very interested in knowing how to use the audio for this dataset because not all deepfakes would have modded audio. Therefore ground truth for audio is quite challenging to create. This will indirectly lead to a lot of noise in the model being built.</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 734707,
          "author_name": "unkownhihi",
          "author_url": "",
          "post_date": "02/01/2020 21:23:53",
          "content": "<p><a href=\"/akashnandi\">@akashnandi</a> here is a dataset already pre-calculated the ground truth for audio. <a href=\"https://www.kaggle.com/basharallabadi/dfdc-video-audio-labels\">link</a></p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 734834,
          "author_name": "akashnandi",
          "author_url": "",
          "post_date": "02/02/2020 04:14:12",
          "content": "<p>Thanks, I'll have a look.</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "733440": "",
    "734132": "Our best score without audio:\n0.44276\nBTW I don't think using audio would boost score. Most videos' audio are not manipulated.",
    "734159": "Thank you very much for sharing your score and inputs.",
    "734224": "But we don't know if it is true about test videos.",
    "734702": "I'm actually very interested in knowing how to use the audio for this dataset because not all deepfakes would have modded audio. Therefore ground truth for audio is quite challenging to create. This will indirectly lead to a lot of noise in the model being built.",
    "734707": "akashnandi here is a dataset already pre-calculated the ground truth for audio. [link](https://www.kaggle.com/basharallabadi/dfdc-video-audio-labels)",
    "734834": "Thanks, I'll have a look."
  },
  "source": "meta"
}