{
  "id": 131273,
  "title": "resize and detect or the other way around?",
  "url": "/competitions/deepfake-detection-challenge/discussion/131273",
  "author_name": "",
  "post_date": "2020-02-19T04:25:24.823903Z",
  "votes": 2,
  "comment_count": 4,
  "views": 0,
  "content": "<p>Hi guys, I just got interested in this competition so I have a simple question about how people are dealing with this problem.</p>\n\n<p>I saw wonderful kernels and discussions that gave me help to understand this competition.\nBut I'm not sure about one thing. What image size are you using to train the model? maybe (224,244)?\nResizing the frame into half the size will give me smaller faces such (100, 100, 3)\nWithout resizing, I will get larger images but with more time to extract frames.\nshould I start with smaller faces and gradually increase size? any advice would be appreciated</p>\n\n<p>I haven't trained nor submitted yet, but I'm curious how people are dealing with this.\nthanks and have a nice day!</p>",
  "messages": [
    {
      "id": "750022",
      "postDate": "02/19/2020 04:25:24",
      "content": "<p>Hi guys, I just got interested in this competition so I have a simple question about how people are dealing with this problem.</p>\n\n<p>I saw wonderful kernels and discussions that gave me help to understand this competition.\nBut I'm not sure about one thing. What image size are you using to train the model? maybe (224,244)?\nResizing the frame into half the size will give me smaller faces such (100, 100, 3)\nWithout resizing, I will get larger images but with more time to extract frames.\nshould I start with smaller faces and gradually increase size? any advice would be appreciated</p>\n\n<p>I haven't trained nor submitted yet, but I'm curious how people are dealing with this.\nthanks and have a nice day!</p>",
      "rawMarkdown": "Hi guys, I just got interested in this competition so I have a simple question about how people are dealing with this problem.\n\nI saw wonderful kernels and discussions that gave me help to understand this competition.\nBut I'm not sure about one thing. What image size are you using to train the model? maybe (224,244)?\nResizing the frame into half the size will give me smaller faces such (100, 100, 3)\nWithout resizing, I will get larger images but with more time to extract frames.\nshould I start with smaller faces and gradually increase size? any advice would be appreciated\n\nI haven't trained nor submitted yet, but I'm curious how people are dealing with this.\nthanks and have a nice day!",
      "votes": null
    },
    {
      "id": "750026",
      "postDate": "02/19/2020 04:29:43",
      "content": "<p>I am doing it without resizing frame. I am using 240,240 as face size to shape to but with a margin of 40.</p>",
      "rawMarkdown": "I am doing it without resizing frame. I am using 240,240 as face size to shape to but with a margin of 40.",
      "votes": null
    },
    {
      "id": "750027",
      "postDate": "02/19/2020 04:31:03",
      "content": "<p>I see, than how long does it take to extract all the faces?</p>",
      "rawMarkdown": "I see, than how long does it take to extract all the faces?",
      "votes": null
    },
    {
      "id": "750029",
      "postDate": "02/19/2020 04:32:55",
      "content": "<p>30 minutes per chunk(on T4 aws instance)</p>",
      "rawMarkdown": "30 minutes per chunk(on T4 aws instance)",
      "votes": null
    },
    {
      "id": "750031",
      "postDate": "02/19/2020 04:34:58",
      "content": "<p>that's pretty fast, thanks for sharing!</p>",
      "rawMarkdown": "that's pretty fast, thanks for sharing!",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 750026,
      "author_name": "unkownhihi",
      "author_url": "",
      "post_date": "02/19/2020 04:29:43",
      "content": "<p>I am doing it without resizing frame. I am using 240,240 as face size to shape to but with a margin of 40.</p>",
      "votes": null,
      "replies": [
        {
          "id": 750027,
          "author_name": "yangsaewon",
          "author_url": "",
          "post_date": "02/19/2020 04:31:03",
          "content": "<p>I see, than how long does it take to extract all the faces?</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 750029,
          "author_name": "unkownhihi",
          "author_url": "",
          "post_date": "02/19/2020 04:32:55",
          "content": "<p>30 minutes per chunk(on T4 aws instance)</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 750031,
          "author_name": "yangsaewon",
          "author_url": "",
          "post_date": "02/19/2020 04:34:58",
          "content": "<p>that's pretty fast, thanks for sharing!</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "750022": "Hi guys, I just got interested in this competition so I have a simple question about how people are dealing with this problem.\n\nI saw wonderful kernels and discussions that gave me help to understand this competition.\nBut I'm not sure about one thing. What image size are you using to train the model? maybe (224,244)?\nResizing the frame into half the size will give me smaller faces such (100, 100, 3)\nWithout resizing, I will get larger images but with more time to extract frames.\nshould I start with smaller faces and gradually increase size? any advice would be appreciated\n\nI haven't trained nor submitted yet, but I'm curious how people are dealing with this.\nthanks and have a nice day!",
    "750026": "I am doing it without resizing frame. I am using 240,240 as face size to shape to but with a margin of 40.",
    "750027": "I see, than how long does it take to extract all the faces?",
    "750029": "30 minutes per chunk(on T4 aws instance)",
    "750031": "that's pretty fast, thanks for sharing!"
  },
  "source": "meta"
}