{
  "id": 133348,
  "title": "CNN-generated images are surprisingly easy to spot...for now",
  "url": "/competitions/deepfake-detection-challenge/discussion/133348",
  "author_name": "",
  "post_date": "2020-03-02T08:52:49.010434100Z",
  "votes": 7,
  "comment_count": 4,
  "views": 0,
  "content": "<p><em>Originally posted on Hacker News</em></p>\n\n<p><a href=\"https://peterwang512.github.io/CNNDetection/\">https://peterwang512.github.io/CNNDetection/</a></p>",
  "messages": [
    {
      "id": "761204",
      "postDate": "03/02/2020 08:52:49",
      "content": "<p><em>Originally posted on Hacker News</em></p>\n\n<p><a href=\"https://peterwang512.github.io/CNNDetection/\">https://peterwang512.github.io/CNNDetection/</a></p>",
      "rawMarkdown": "*Originally posted on Hacker News*\n\nhttps://peterwang512.github.io/CNNDetection/",
      "votes": null
    },
    {
      "id": "761261",
      "postDate": "03/02/2020 10:23:41",
      "content": "<p>I tried this but didn't see the same sort of spectral artifacts in the training data. It's possible that a subset of training data has this but I couldn't find any evidence of those \"CNN fingerprints\" in the data I looked at.</p>",
      "rawMarkdown": "I tried this but didn't see the same sort of spectral artifacts in the training data. It's possible that a subset of training data has this but I couldn't find any evidence of those \"CNN fingerprints\" in the data I looked at.",
      "votes": null
    },
    {
      "id": "761303",
      "postDate": "03/02/2020 11:24:20",
      "content": "<p>The authors in the paper mentioned they used two GANs ... one that was good and one that was less good ... this is why there is a disparity of Fake Quality.</p>",
      "rawMarkdown": "The authors in the paper mentioned they used two GANs ... one that was good and one that was less good ... this is why there is a disparity of Fake Quality.",
      "votes": null
    },
    {
      "id": "761680",
      "postDate": "03/02/2020 20:57:25",
      "content": "<p>Thanks for the link! </p>",
      "rawMarkdown": "Thanks for the link!",
      "votes": null
    },
    {
      "id": "1273947",
      "postDate": "04/14/2021 19:43:28",
      "content": "<p>Hey , i'm beginnner in ML<br>\ni found this article for my project but i found it hard to use data set they have provided (i'm working on colab it is excceeding the memory )<br>\ncan any one share your repo of your notebook  who have done it 🤒and how you cropped the data set </p>\n<p>any insights would be helpful<br>\nThank you</p>",
      "rawMarkdown": "Hey , i'm beginnner in ML\ni found this article for my project but i found it hard to use data set they have provided (i'm working on colab it is excceeding the memory )\ncan any one share your repo of your notebook  who have done it 🤒and how you cropped the data set \n\nany insights would be helpful\nThank you",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1273947,
      "author_name": "rishikram",
      "author_url": "",
      "post_date": "04/14/2021 19:43:28",
      "content": "<p>Hey , i'm beginnner in ML<br>\ni found this article for my project but i found it hard to use data set they have provided (i'm working on colab it is excceeding the memory )<br>\ncan any one share your repo of your notebook  who have done it 🤒and how you cropped the data set </p>\n<p>any insights would be helpful<br>\nThank you</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 761261,
      "author_name": "humananalog",
      "author_url": "",
      "post_date": "03/02/2020 10:23:41",
      "content": "<p>I tried this but didn't see the same sort of spectral artifacts in the training data. It's possible that a subset of training data has this but I couldn't find any evidence of those \"CNN fingerprints\" in the data I looked at.</p>",
      "votes": null,
      "replies": [
        {
          "id": 761303,
          "author_name": "ma7moud",
          "author_url": "",
          "post_date": "03/02/2020 11:24:20",
          "content": "<p>The authors in the paper mentioned they used two GANs ... one that was good and one that was less good ... this is why there is a disparity of Fake Quality.</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 761680,
      "author_name": "yassinealouini",
      "author_url": "",
      "post_date": "03/02/2020 20:57:25",
      "content": "<p>Thanks for the link! </p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "761204": "*Originally posted on Hacker News*\n\nhttps://peterwang512.github.io/CNNDetection/",
    "761261": "I tried this but didn't see the same sort of spectral artifacts in the training data. It's possible that a subset of training data has this but I couldn't find any evidence of those \"CNN fingerprints\" in the data I looked at.",
    "761303": "The authors in the paper mentioned they used two GANs ... one that was good and one that was less good ... this is why there is a disparity of Fake Quality.",
    "761680": "Thanks for the link!",
    "1273947": "Hey , i'm beginnner in ML\ni found this article for my project but i found it hard to use data set they have provided (i'm working on colab it is excceeding the memory )\ncan any one share your repo of your notebook  who have done it 🤒and how you cropped the data set \n\nany insights would be helpful\nThank you"
  },
  "source": "meta"
}