Note the pre-training data (EgoClip, Ego4D) in the model card

Trim the model card to essentials

Fix the repo id in the loader docstring and default

Sync util/

Update the checkpoint filename in the header comment

Remove the language section; the checkpoint is single-modality

Update the paper reference in config.json

Trim the model card: drop cross-model comparisons, cite the IJCV version

Add verification details to the model card

Add STP ViT-B/16 pre-trained weights (slim: model only + decoder-free encoder)

initial commit