{
  "id": 131150,
  "title": "Is it necessary to align detected faces rather than only cropping them?",
  "url": "/competitions/deepfake-detection-challenge/discussion/131150",
  "author_name": "",
  "post_date": "2020-02-18T14:46:11.417777700Z",
  "votes": null,
  "comment_count": 3,
  "views": 0,
  "content": "<p>I am trying to construct a training set by cropping faces in videos.\nConsidering the huge data volumn, to align faces takes significant amount of computation.\nI wonder if it is necessary to align cropped faces.</p>\n\n<p>Thanks for the help!</p>",
  "messages": [
    {
      "id": "749295",
      "postDate": "02/18/2020 14:46:11",
      "content": "<p>I am trying to construct a training set by cropping faces in videos.\nConsidering the huge data volumn, to align faces takes significant amount of computation.\nI wonder if it is necessary to align cropped faces.</p>\n\n<p>Thanks for the help!</p>",
      "rawMarkdown": "I am trying to construct a training set by cropping faces in videos.\nConsidering the huge data volumn, to align faces takes significant amount of computation.\nI wonder if it is necessary to align cropped faces.\n\nThanks for the help!",
      "votes": null
    },
    {
      "id": "749738",
      "postDate": "02/18/2020 22:08:13",
      "content": "<p>I guess it depends on the model, but I assume you're referring to something like a 2D CNN -&gt; LSTM? One would imagine if you used a GlobalAveragePooling layer prior to the LSTM the exact location of the faces probably wouldn't matter, but I've not experimented...</p>",
      "rawMarkdown": "I guess it depends on the model, but I assume you're referring to something like a 2D CNN -&gt; LSTM? One would imagine if you used a GlobalAveragePooling layer prior to the LSTM the exact location of the faces probably wouldn't matter, but I've not experimented...",
      "votes": null
    },
    {
      "id": "750663",
      "postDate": "02/19/2020 15:31:47",
      "content": "<p>Based on the literature I'm exposed to alignment gives a small accuracy pump ... I would put it on the TODO list but not the most important thing you'd need to worry about now ... </p>",
      "rawMarkdown": "Based on the literature I'm exposed to alignment gives a small accuracy pump ... I would put it on the TODO list but not the most important thing you'd need to worry about now ...",
      "votes": null
    },
    {
      "id": "754401",
      "postDate": "02/23/2020 14:02:09",
      "content": "<p>Got it, thanks!</p>",
      "rawMarkdown": "Got it, thanks!",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 749738,
      "author_name": "jamesphoward",
      "author_url": "",
      "post_date": "02/18/2020 22:08:13",
      "content": "<p>I guess it depends on the model, but I assume you're referring to something like a 2D CNN -&gt; LSTM? One would imagine if you used a GlobalAveragePooling layer prior to the LSTM the exact location of the faces probably wouldn't matter, but I've not experimented...</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 750663,
      "author_name": "ma7moud",
      "author_url": "",
      "post_date": "02/19/2020 15:31:47",
      "content": "<p>Based on the literature I'm exposed to alignment gives a small accuracy pump ... I would put it on the TODO list but not the most important thing you'd need to worry about now ... </p>",
      "votes": null,
      "replies": [
        {
          "id": 754401,
          "author_name": "fuzhuolin",
          "author_url": "",
          "post_date": "02/23/2020 14:02:09",
          "content": "<p>Got it, thanks!</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "749295": "I am trying to construct a training set by cropping faces in videos.\nConsidering the huge data volumn, to align faces takes significant amount of computation.\nI wonder if it is necessary to align cropped faces.\n\nThanks for the help!",
    "749738": "I guess it depends on the model, but I assume you're referring to something like a 2D CNN -&gt; LSTM? One would imagine if you used a GlobalAveragePooling layer prior to the LSTM the exact location of the faces probably wouldn't matter, but I've not experimented...",
    "750663": "Based on the literature I'm exposed to alignment gives a small accuracy pump ... I would put it on the TODO list but not the most important thing you'd need to worry about now ...",
    "754401": "Got it, thanks!"
  },
  "source": "meta"
}