{
  "id": 435424,
  "title": "Enhancing the Accuracy of Deep Fake Detection Model",
  "url": "/competitions/deepfake-detection-challenge/discussion/435424",
  "author_name": "",
  "post_date": "2023-08-29T10:23:42.257213Z",
  "votes": null,
  "comment_count": 1,
  "views": 0,
  "content": "<p>Greetings Kaggle Community! I hope all of you are doing well.</p>\n<p>I'm currently working on a project on \"Deep Fake Detection Using CNN-LSTM\", and using the DFDC dataset on Kaggle. Since the competition has ended 3 years ago, therefore, I'm using the currently available dataset that contains 2 folders, i.e., \"train_sample_videos\" and \"test_videos\". I want to use CNN-LSTM architecture, and my target is to get accuracy past 90%. Can anyone help or suggest me how to improve model's performance? It'd be a great help. Thank You.</p>\n<p>Currently, I'm following the methodology mentioned below:</p>\n<ol>\n<li>Capturing Frames from Videos</li>\n<li>Making a separate working directory, and making 2 new folders named \"real\" &amp; \"fake\", and then using Keras' ImageDataGenerator class to make train and test splits.</li>\n<li>I'm using the metadata available in dataset to separate real and fake faces.</li>\n</ol>",
  "messages": [
    {
      "id": "2414030",
      "postDate": "08/29/2023 10:23:42",
      "content": "<p>Greetings Kaggle Community! I hope all of you are doing well.</p>\n<p>I'm currently working on a project on \"Deep Fake Detection Using CNN-LSTM\", and using the DFDC dataset on Kaggle. Since the competition has ended 3 years ago, therefore, I'm using the currently available dataset that contains 2 folders, i.e., \"train_sample_videos\" and \"test_videos\". I want to use CNN-LSTM architecture, and my target is to get accuracy past 90%. Can anyone help or suggest me how to improve model's performance? It'd be a great help. Thank You.</p>\n<p>Currently, I'm following the methodology mentioned below:</p>\n<ol>\n<li>Capturing Frames from Videos</li>\n<li>Making a separate working directory, and making 2 new folders named \"real\" &amp; \"fake\", and then using Keras' ImageDataGenerator class to make train and test splits.</li>\n<li>I'm using the metadata available in dataset to separate real and fake faces.</li>\n</ol>",
      "rawMarkdown": "Greetings Kaggle Community! I hope all of you are doing well.\n\nI'm currently working on a project on \"Deep Fake Detection Using CNN-LSTM\", and using the DFDC dataset on Kaggle. Since the competition has ended 3 years ago, therefore, I'm using the currently available dataset that contains 2 folders, i.e., \"train_sample_videos\" and \"test_videos\". I want to use CNN-LSTM architecture, and my target is to get accuracy past 90%. Can anyone help or suggest me how to improve model's performance? It'd be a great help. Thank You.\n\nCurrently, I'm following the methodology mentioned below:\n1. Capturing Frames from Videos\n2. Making a separate working directory, and making 2 new folders named \"real\" & \"fake\", and then using Keras' ImageDataGenerator class to make train and test splits.\n3. I'm using the metadata available in dataset to separate real and fake faces.",
      "votes": null
    },
    {
      "id": "2667710",
      "postDate": "02/25/2024 09:40:53",
      "content": "<p><a href=\"https://www.kaggle.com/azazurrehmanbutt\" target=\"_blank\">@azazurrehmanbutt</a> Have you worked in this? Have you improved the model's performance </p>",
      "rawMarkdown": "azazurrehmanbutt Have you worked in this? Have you improved the model's performance",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 2667710,
      "author_name": "ayeshafareed",
      "author_url": "",
      "post_date": "02/25/2024 09:40:53",
      "content": "<p><a href=\"https://www.kaggle.com/azazurrehmanbutt\" target=\"_blank\">@azazurrehmanbutt</a> Have you worked in this? Have you improved the model's performance </p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "2414030": "Greetings Kaggle Community! I hope all of you are doing well.\n\nI'm currently working on a project on \"Deep Fake Detection Using CNN-LSTM\", and using the DFDC dataset on Kaggle. Since the competition has ended 3 years ago, therefore, I'm using the currently available dataset that contains 2 folders, i.e., \"train_sample_videos\" and \"test_videos\". I want to use CNN-LSTM architecture, and my target is to get accuracy past 90%. Can anyone help or suggest me how to improve model's performance? It'd be a great help. Thank You.\n\nCurrently, I'm following the methodology mentioned below:\n1. Capturing Frames from Videos\n2. Making a separate working directory, and making 2 new folders named \"real\" & \"fake\", and then using Keras' ImageDataGenerator class to make train and test splits.\n3. I'm using the metadata available in dataset to separate real and fake faces.",
    "2667710": "azazurrehmanbutt Have you worked in this? Have you improved the model's performance"
  },
  "source": "meta"
}