README
MoCapAnything V2 — Demo Data (sample)
The mini dataset that powers the MoCapAnything V2 demo: example input videos, per-species 1-frame reference features, character meshes with skinning for rendering, and reference outputs to compare against. Download it into ./demo/data of the code repo and the bundled examples run out of the box.
- 💻 Code: github.com/phongdaot/MocapAnything
- 🏋️ Weights: kehong/MoCapAnythingV2-weights
Contents
| Folder | What's inside |
|---|---|
inputs/ | Example videos: in-the-wild animals, rendered zoo clips, objects, dance clips (with audio) |
zoo1030/ | Mini animal subset: 1-frame reference features (bvh_pose/, npz_train_image_only/), reference BVHs, character meshes + skinning (characters_fix_facezplus/), reference images (front & side views) |
obj1k/ | Mini object subset with the same layout (characters/) |
expected_results/ | Reference *_final.mp4 outputs for the wild / zoo / obj examples — what a successful run should look like |
The reference features ship only frame 0 of each reference sequence (all the demo needs at inference — verified bit-identical to using the full features), which keeps this download small.
Usage
git clone https://github.com/phongdaot/MocapAnything.git && cd MocapAnything
hf download kehong/MoCapAnythingV2-data-sample --repo-type dataset --local-dir ./demo/data
hf download kehong/MoCapAnythingV2-weights --local-dir ./checkpoints
python inference/video2pose2rot.py --config demo/configs/demo_wild.yaml # then compare with expected_results/
python demo/app.py # interactive demo
Notes
- This is a small demo sample intended to make the code runnable end-to-end; the full training datasets are a separate (future) release.
- Wild examples whose species has no reference in this mini subset are skipped with a warning — they unlock with the full dataset.
Contributors
6 commits
Similar repos
README
MoCapAnything V2 — Demo Data (sample)
The mini dataset that powers the MoCapAnything V2 demo: example input videos, per-species 1-frame reference features, character meshes with skinning for rendering, and reference outputs to compare against. Download it into ./demo/data of the code repo and the bundled examples run out of the box.
- 💻 Code: github.com/phongdaot/MocapAnything
- 🏋️ Weights: kehong/MoCapAnythingV2-weights
Contents
| Folder | What's inside |
|---|---|
inputs/ | Example videos: in-the-wild animals, rendered zoo clips, objects, dance clips (with audio) |
zoo1030/ | Mini animal subset: 1-frame reference features (bvh_pose/, npz_train_image_only/), reference BVHs, character meshes + skinning (characters_fix_facezplus/), reference images (front & side views) |
obj1k/ | Mini object subset with the same layout (characters/) |
expected_results/ | Reference *_final.mp4 outputs for the wild / zoo / obj examples — what a successful run should look like |
The reference features ship only frame 0 of each reference sequence (all the demo needs at inference — verified bit-identical to using the full features), which keeps this download small.
Usage
git clone https://github.com/phongdaot/MocapAnything.git && cd MocapAnything
hf download kehong/MoCapAnythingV2-data-sample --repo-type dataset --local-dir ./demo/data
hf download kehong/MoCapAnythingV2-weights --local-dir ./checkpoints
python inference/video2pose2rot.py --config demo/configs/demo_wild.yaml # then compare with expected_results/
python demo/app.py # interactive demo
Notes
- This is a small demo sample intended to make the code runnable end-to-end; the full training datasets are a separate (future) release.
- Wild examples whose species has no reference in this mini subset are skipped with a warning — they unlock with the full dataset.
Similar repos
Linked in READMEs
Contributors
6 commits