Downloads · 30 days
2
18% of all-time downloads
arunos728/fastwam-egodex-uncond-1cam320-30k
fastwam-egodex-uncond-1cam320-30k is a machine learning model from arunos728. Use it for the machine learning task on the model card, and read the license before you ship it in a product.
FastWAM on the EgoDex human-hands corpus: 26 task dirs, 44,507 episodes / 12,230,294 frames, lerobot v2.1, AV1 180x320 @ 30 fps, 44-D action/state.
Downloads · 30 days
2
18% of all-time downloads
All-time downloads
11
Public
Repo size
12 GB
Likes
0
Public
Click a slice to open those files.
.pt12 GB · 100%
From the Hugging Face model README
FastWAM on the EgoDex human-hands corpus: 26 task dirs, 44,507 episodes / 12,230,294 frames, lerobot v2.1, AV1 180x320 @ 30 fps, 44-D action/state.
4 GPU x batch 16 x grad-accum 3 = global batch 192, 30k steps, lr 1e-4 with cosine decay (to 1e-6 by the end).
Trained at 192x320, not the 176x320 the recipe originally specified: the wan22 VAE compresses 16x, so a 176-tall canvas becomes an 11-tall latent and the DiT patch (2, 2) rejects the odd dimension. The real constraint is a multiple of 32. 192 is the nearest one above the native 180.
Weights only; checkpoints/state/ is not published.