{
  "id": 122140,
  "title": "average hardware for this competition + optimization of data to run on at least a 1070",
  "url": "/competitions/deepfake-detection-challenge/discussion/122140",
  "author_name": "",
  "post_date": "2019-12-18T02:19:56.749897100Z",
  "votes": null,
  "comment_count": 2,
  "views": 0,
  "content": "<p>hi people, i want to know if someone is going to run it on a at least a 1070 (8gb vram)</p>\n\n<p>i am not sure if this is going to be enough for all this videos in fullhd, i am waiting for the weekend to start to build some models and run it on this kind of hardware</p>\n\n<p>i am thinking of building a module to cut the faces of the video, like a roi of 300x300px , or something around it, and train on my machine</p>\n\n<p>if you have any suggestion on how to improve it to run in this kind of hardware, let me know, and let share experiencies about it</p>\n\n<p>thx</p>",
  "messages": [
    {
      "id": "697496",
      "postDate": "12/18/2019 02:19:56",
      "content": "<p>hi people, i want to know if someone is going to run it on a at least a 1070 (8gb vram)</p>\n\n<p>i am not sure if this is going to be enough for all this videos in fullhd, i am waiting for the weekend to start to build some models and run it on this kind of hardware</p>\n\n<p>i am thinking of building a module to cut the faces of the video, like a roi of 300x300px , or something around it, and train on my machine</p>\n\n<p>if you have any suggestion on how to improve it to run in this kind of hardware, let me know, and let share experiencies about it</p>\n\n<p>thx</p>",
      "rawMarkdown": "hi people, i want to know if someone is going to run it on a at least a 1070 (8gb vram)\n\ni am not sure if this is going to be enough for all this videos in fullhd, i am waiting for the weekend to start to build some models and run it on this kind of hardware\n\ni am thinking of building a module to cut the faces of the video, like a roi of 300x300px , or something around it, and train on my machine\n\nif you have any suggestion on how to improve it to run in this kind of hardware, let me know, and let share experiencies about it\n\nthx",
      "votes": null
    },
    {
      "id": "697794",
      "postDate": "12/18/2019 12:18:22",
      "content": "<p>If you look at something like MesoNet, it uses a very small neural network with only 27k parameters. (This was done on purpose because they don't want the NN to learn high-level concepts.) So your solution doesn't necessarily need to be a huge neural network.</p>",
      "rawMarkdown": "If you look at something like MesoNet, it uses a very small neural network with only 27k parameters. (This was done on purpose because they don't want the NN to learn high-level concepts.) So your solution doesn't necessarily need to be a huge neural network.",
      "votes": null
    },
    {
      "id": "698103",
      "postDate": "12/18/2019 20:02:48",
      "content": "<p>ohh nice, i am going to take a loot at that thx <a href=\"/humananalog\">@humananalog</a> !</p>",
      "rawMarkdown": "ohh nice, i am going to take a loot at that thx @humananalog !",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 697794,
      "author_name": "humananalog",
      "author_url": "",
      "post_date": "12/18/2019 12:18:22",
      "content": "<p>If you look at something like MesoNet, it uses a very small neural network with only 27k parameters. (This was done on purpose because they don't want the NN to learn high-level concepts.) So your solution doesn't necessarily need to be a huge neural network.</p>",
      "votes": null,
      "replies": [
        {
          "id": 698103,
          "author_name": "satansw",
          "author_url": "",
          "post_date": "12/18/2019 20:02:48",
          "content": "<p>ohh nice, i am going to take a loot at that thx <a href=\"/humananalog\">@humananalog</a> !</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "697496": "hi people, i want to know if someone is going to run it on a at least a 1070 (8gb vram)\n\ni am not sure if this is going to be enough for all this videos in fullhd, i am waiting for the weekend to start to build some models and run it on this kind of hardware\n\ni am thinking of building a module to cut the faces of the video, like a roi of 300x300px , or something around it, and train on my machine\n\nif you have any suggestion on how to improve it to run in this kind of hardware, let me know, and let share experiencies about it\n\nthx",
    "697794": "If you look at something like MesoNet, it uses a very small neural network with only 27k parameters. (This was done on purpose because they don't want the NN to learn high-level concepts.) So your solution doesn't necessarily need to be a huge neural network.",
    "698103": "ohh nice, i am going to take a loot at that thx @humananalog !"
  },
  "source": "meta"
}