{
  "id": 462064,
  "title": "Concerns and Suggestion Regarding Pose Influence on the Metrics",
  "url": "/competitions/audio-driven-talking-head-generation/discussion/462064",
  "author_name": "",
  "post_date": "2023-12-18T05:49:58.890626800Z",
  "votes": null,
  "comment_count": 1,
  "views": 0,
  "content": "<p>For the video setting, the pose has too much influence on the metric calculations. The calculations of PSNR, SSIM, L1trans and L1angle are all heavily dependent on the ground-truth poses. However, predicting poses from speech is not reliable and has many answers.  One or two Distribution-Based metrics (FID, Inception Score) should be replaced with or included.</p>",
  "messages": [
    {
      "id": "2565545",
      "postDate": "12/18/2023 05:49:58",
      "content": "<p>For the video setting, the pose has too much influence on the metric calculations. The calculations of PSNR, SSIM, L1trans and L1angle are all heavily dependent on the ground-truth poses. However, predicting poses from speech is not reliable and has many answers.  One or two Distribution-Based metrics (FID, Inception Score) should be replaced with or included.</p>",
      "rawMarkdown": "For the video setting, the pose has too much influence on the metric calculations. The calculations of PSNR, SSIM, L1trans and L1angle are all heavily dependent on the ground-truth poses. However, predicting poses from speech is not reliable and has many answers.  One or two Distribution-Based metrics (FID, Inception Score) should be replaced with or included.",
      "votes": null
    },
    {
      "id": "2565607",
      "postDate": "12/18/2023 06:54:07",
      "content": "<p>Thank you for your comments. The question has been brought up <a href=\"https://www.kaggle.com/competitions/audio-driven-talking-head-generation/discussion/452293\" target=\"_blank\">here</a> before, and we've provided the GT camera poses obtained from AD-NeRF as a reference, you could find it <a href=\"https://www.kaggle.com/competitions/audio-driven-talking-head-generation/data?select=Testing_Camera_Poses\" target=\"_blank\">here</a>. </p>",
      "rawMarkdown": "Thank you for your comments. The question has been brought up [here](https://www.kaggle.com/competitions/audio-driven-talking-head-generation/discussion/452293) before, and we've provided the GT camera poses obtained from AD-NeRF as a reference, you could find it [here](https://www.kaggle.com/competitions/audio-driven-talking-head-generation/data?select=Testing_Camera_Poses).",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 2565607,
      "author_name": "jingnangao",
      "author_url": "",
      "post_date": "12/18/2023 06:54:07",
      "content": "<p>Thank you for your comments. The question has been brought up <a href=\"https://www.kaggle.com/competitions/audio-driven-talking-head-generation/discussion/452293\" target=\"_blank\">here</a> before, and we've provided the GT camera poses obtained from AD-NeRF as a reference, you could find it <a href=\"https://www.kaggle.com/competitions/audio-driven-talking-head-generation/data?select=Testing_Camera_Poses\" target=\"_blank\">here</a>. </p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "2565545": "For the video setting, the pose has too much influence on the metric calculations. The calculations of PSNR, SSIM, L1trans and L1angle are all heavily dependent on the ground-truth poses. However, predicting poses from speech is not reliable and has many answers.  One or two Distribution-Based metrics (FID, Inception Score) should be replaced with or included.",
    "2565607": "Thank you for your comments. The question has been brought up [here](https://www.kaggle.com/competitions/audio-driven-talking-head-generation/discussion/452293) before, and we've provided the GT camera poses obtained from AD-NeRF as a reference, you could find it [here](https://www.kaggle.com/competitions/audio-driven-talking-head-generation/data?select=Testing_Camera_Poses)."
  },
  "source": "meta"
}