Weights for an anonymous submission
Weights of the model described in the submission, for use with its supplementary code
(scripts/evaluate.py, scripts/train.py). Each file is a plain PyTorch state_dict container
loadable with torch.load(path, weights_only=True).
| File | Content |
|---|---|
stage1.pt |
Stage 1: LoRA adapters, prediction heads and FK conditioning (the frozen MoVieS backbone is loaded from its public release) |
stage2.pt |
Stage 2: recurrent ReSplat updater and bridge gates |
latent_stem.pt |
LatentStem (video-latent input) |
SHA256SUMS lists the checksums. Set stage1_checkpoint, stage2_checkpoint and latent_stem_checkpoint in
the code's configs/paths.yaml to the downloaded files.
The released third-party weights these files build on (MoVieS, ReSplat) are not redistributed here; see the code's README for their sources and licenses.