YAML Metadata Warning:empty or missing yaml metadata in repo card
Check out the documentation for more information.
WEAVER HRI cube: two-stage world model artefacts
Backup of the files that tools/run_cube_pipeline.sh (github.com/LeeHakHo/WEAVER-HRI, branch hri-finetune) produces.
Restore by copying into the repo root: data/ and runs/ map onto the same paths.
Checkpoints are trainable-parameter-only; weaver.finetune_utils.load_finetuned recovers the frozen remainder from arnavkj1995/WEAVER.
| path | what |
|---|---|
| data/weaver_hri_cube_v3p | final dataset: 16-D planar state, 14-D action, SD3 latents, norm stats, layout, control files |
| data/hand_tracks, hand_actions, object_tracks | WiLoR+depth hand points, IDM-tagged hand commands, SAM3 cube positions |
| runs/idm/idm_skel | inverse-dynamics model used to tag hand actions |
| runs/weaver_cube_s1p | stage-1 states-only WEAVER, checkpoint 40000 (post-hoc val selected) |
| data/s1p | stage-1 predictions (val per checkpoint, train for 40000) |
| runs/decoder/dec_s1p_40000 | small skeleton decoder |
| runs/weaver_cube_s1p_s2 | stage-2 WEAVER renderer, checkpoint 20000 |
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support