HAS-Bench Baselines
Pretrained baseline checkpoints for the Hide-and-Seek tactile dataset: a dual-head multimodal classifier with an object head and a weight head, trained with the public release pipeline (seed 42).
| File | Modalities |
|---|---|
m123/best_model.pt |
tactile point cloud + raw skin signals + virtual skin wrenches |
m12345/best_model.pt |
+ end-effector pose + wrist F/T |
Each directory also contains the last_metrics.json written at the end of training.
Usage
hf download TUM-ICS/HAS-Bench-baselines --local-dir checkpoints
Dataloaders and evaluation scripts: https://github.com/TUM-ICS/Tactile-Hide-and-Seek