The dataset viewer is not available because its heuristics could not detect any supported data files. You can try uploading some data files, or configuring the data files location manually.
Hand2Throw: hand silhouettes
The operator's hand during teleoperation, reduced to a plain white silhouette on black (no camera
pixels), for the two cameras used to record the teleop demos.
These are the inputs of the silhouette latent action models in hand2throw-lam.
Code and full write-up: https://github.com/erensckin/hand2throw
Each silhouette is the convex hull of the 21 MediaPipe hand landmarks, dilated by about 0.35 hand sizes, centre-cropped to a square and resized to 96×96. Frames where no hand was detected are black.
Files
human_cam1_silhouette.npz (webcam) and human_cam2_silhouette.npz (phone), each with one row per
robot step of all 205 teleop episodes (33,741 rows):
| Array | Shape | Content |
|---|---|---|
frames |
(N, 96, 96, 3) uint8 | Silhouette (identical in all three channels) |
hand |
(N, 4) | Tracker features used by teleop: palm x, y (from the image centre, in hand sizes), pinch ratio, hand present |
actions |
(N, 7) | The robot action at that step |
episode, frame_index |
(N,) | Row → episode and step in the teleop dataset |
phase |
(N,) | 0 = hand-driven, 1 = scripted throw |
Usage
Place both files in data/lam_cache/ of the GitHub repo and probe the silhouette models:
uv run hf download erensckin/hand2throw-hand-silhouettes --repo-type dataset --local-dir data/lam_cache
uv run hf download erensckin/hand2throw-lam --include "human_cam*_silhouette/*" --local-dir outputs/lam
uv run python scripts/lam.py probe --sources human_cam1_silhouette,human_cam2_silhouette
This reports, on held-out episodes (index % 5 == 0), how well the latent actions and the tracker features predict the robot's actions (R² per action dimension).
- Downloads last month
- 20