{
  "id": 685930,
  "title": "Missing files",
  "url": "/competitions/motion-s-hierarchical-text-to-motion-generation-for-sign-language/discussion/685930",
  "author_name": "",
  "post_date": "2026-03-29T18:18:18.174912400Z",
  "votes": null,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Where can I find {gloss}.bvh files for Individual sign motion for each gloss in the sentence? It was listed in the dataset structure, but not in the dataset. I saw all the other data files, except the {gloss}.bvh</p>\n<p>How to use .bvh files? Since .npy files have pre-extracted motion features and can help visualize, what other information can .bvh files provide?</p>\n<p>And where is evaluation_script.py, I didn't find it. Is it replaced by Motion-S_Evaluator_T2M. default ?</p>\n<p>Is Motion-S starter notebook V2  the baseline-notebook? </p>",
  "messages": [
    {
      "id": "3431248",
      "postDate": "03/29/2026 18:18:18",
      "content": "<p>Where can I find {gloss}.bvh files for Individual sign motion for each gloss in the sentence? It was listed in the dataset structure, but not in the dataset. I saw all the other data files, except the {gloss}.bvh</p>\n<p>How to use .bvh files? Since .npy files have pre-extracted motion features and can help visualize, what other information can .bvh files provide?</p>\n<p>And where is evaluation_script.py, I didn't find it. Is it replaced by Motion-S_Evaluator_T2M. default ?</p>\n<p>Is Motion-S starter notebook V2  the baseline-notebook? </p>",
      "rawMarkdown": "Where can I find {gloss}.bvh files for Individual sign motion for each gloss in the sentence? It was listed in the dataset structure, but not in the dataset. I saw all the other data files, except the {gloss}.bvh\n\nHow to use .bvh files? Since .npy files have pre-extracted motion features and can help visualize, what other information can .bvh files provide?\n\nAnd where is evaluation_script.py, I didn't find it. Is it replaced by Motion-S_Evaluator_T2M. default ?\n\nIs Motion-S starter notebook V2  the baseline-notebook?",
      "votes": null
    },
    {
      "id": "3437860",
      "postDate": "04/08/2026 08:15:38",
      "content": "<p>Hello <a href=\"https://www.kaggle.com/cxy1368\" target=\"_blank\">@cxy1368</a> , the individual sign motion for each gloss isnt available on its own since the final bvh motion for a single gloss sentence was stitched from multiple single bvh of motion files.  On the query of how useful the bvh files can be? the bvh are preprocess to a 668 dim feature vector folowing this stage:</p>\n<pre><code>Raw BVH Motion File\n    ↓\n[Load &amp; Parse] → Local Quaternions + Positions\n    ↓\n[Downsample] → Target FPS\n    ↓\n[Forward Kinematics] → Global Positions + Orientations\n    ↓\n[Floor Alignment] → Put character on ground\n    ↓\n[Forward Direction] → Extract &amp; smooth facing direction\n    ↓\n[Root Normalization] → Align all poses to face Z+\n    ↓\n[Feature Extraction] → Velocities, rotations, contacts\n    ↓\nFinal Feature Vector (N-1, D_POSE)\n</code></pre>\n<p>..we did not provide the evaluation script in the Motion-S_Evaluator_T2M since its measures only one of the eval scores which is the r_presicion score(text-motion alignment)..</p>\n<p>the latest version of  Motion-S starter notebook V2 is supposed to give a higher score than the baseline</p>",
      "rawMarkdown": "Hello @cxy1368 , the individual sign motion for each gloss isnt available on its own since the final bvh motion for a single gloss sentence was stitched from multiple single bvh of motion files.  On the query of how useful the bvh files can be? the bvh are preprocess to a 668 dim feature vector folowing this stage:\n\n```\nRaw BVH Motion File\n    ↓\n[Load & Parse] → Local Quaternions + Positions\n    ↓\n[Downsample] → Target FPS\n    ↓\n[Forward Kinematics] → Global Positions + Orientations\n    ↓\n[Floor Alignment] → Put character on ground\n    ↓\n[Forward Direction] → Extract & smooth facing direction\n    ↓\n[Root Normalization] → Align all poses to face Z+\n    ↓\n[Feature Extraction] → Velocities, rotations, contacts\n    ↓\nFinal Feature Vector (N-1, D_POSE)\n```\n..we did not provide the evaluation script in the Motion-S_Evaluator_T2M since its measures only one of the eval scores which is the r_presicion score(text-motion alignment)..\n\nthe latest version of  Motion-S starter notebook V2 is supposed to give a higher score than the baseline",
      "votes": null
    },
    {
      "id": "3447542",
      "postDate": "04/23/2026 09:39:30",
      "content": "<p>Are the 668-dim BVH feature vectors provided per training sample? </p>",
      "rawMarkdown": "Are the 668-dim BVH feature vectors provided per training sample?",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 3437860,
      "author_name": "antonygithinji",
      "author_url": "",
      "post_date": "04/08/2026 08:15:38",
      "content": "<p>Hello <a href=\"https://www.kaggle.com/cxy1368\" target=\"_blank\">@cxy1368</a> , the individual sign motion for each gloss isnt available on its own since the final bvh motion for a single gloss sentence was stitched from multiple single bvh of motion files.  On the query of how useful the bvh files can be? the bvh are preprocess to a 668 dim feature vector folowing this stage:</p>\n<pre><code>Raw BVH Motion File\n    ↓\n[Load &amp; Parse] → Local Quaternions + Positions\n    ↓\n[Downsample] → Target FPS\n    ↓\n[Forward Kinematics] → Global Positions + Orientations\n    ↓\n[Floor Alignment] → Put character on ground\n    ↓\n[Forward Direction] → Extract &amp; smooth facing direction\n    ↓\n[Root Normalization] → Align all poses to face Z+\n    ↓\n[Feature Extraction] → Velocities, rotations, contacts\n    ↓\nFinal Feature Vector (N-1, D_POSE)\n</code></pre>\n<p>..we did not provide the evaluation script in the Motion-S_Evaluator_T2M since its measures only one of the eval scores which is the r_presicion score(text-motion alignment)..</p>\n<p>the latest version of  Motion-S starter notebook V2 is supposed to give a higher score than the baseline</p>",
      "votes": null,
      "replies": [
        {
          "id": 3447542,
          "author_name": "paulmwaura",
          "author_url": "",
          "post_date": "04/23/2026 09:39:30",
          "content": "<p>Are the 668-dim BVH feature vectors provided per training sample? </p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "3431248": "Where can I find {gloss}.bvh files for Individual sign motion for each gloss in the sentence? It was listed in the dataset structure, but not in the dataset. I saw all the other data files, except the {gloss}.bvh\n\nHow to use .bvh files? Since .npy files have pre-extracted motion features and can help visualize, what other information can .bvh files provide?\n\nAnd where is evaluation_script.py, I didn't find it. Is it replaced by Motion-S_Evaluator_T2M. default ?\n\nIs Motion-S starter notebook V2  the baseline-notebook?",
    "3437860": "Hello @cxy1368 , the individual sign motion for each gloss isnt available on its own since the final bvh motion for a single gloss sentence was stitched from multiple single bvh of motion files.  On the query of how useful the bvh files can be? the bvh are preprocess to a 668 dim feature vector folowing this stage:\n\n```\nRaw BVH Motion File\n    ↓\n[Load & Parse] → Local Quaternions + Positions\n    ↓\n[Downsample] → Target FPS\n    ↓\n[Forward Kinematics] → Global Positions + Orientations\n    ↓\n[Floor Alignment] → Put character on ground\n    ↓\n[Forward Direction] → Extract & smooth facing direction\n    ↓\n[Root Normalization] → Align all poses to face Z+\n    ↓\n[Feature Extraction] → Velocities, rotations, contacts\n    ↓\nFinal Feature Vector (N-1, D_POSE)\n```\n..we did not provide the evaluation script in the Motion-S_Evaluator_T2M since its measures only one of the eval scores which is the r_presicion score(text-motion alignment)..\n\nthe latest version of  Motion-S starter notebook V2 is supposed to give a higher score than the baseline",
    "3447542": "Are the 668-dim BVH feature vectors provided per training sample?"
  },
  "source": "meta"
}