Reproduce “Anticipative Video Transformer” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Rohit Girdhar, Kristen Grauman
AVT predicts future actions from ordered video frames using masked temporal attention over observed frames. Training adds intermediate feature-prediction and action-classification losses to final next-action classification. On EpicKitchens-100, AVT-b validation class-mean recall@5 rises from 13.1 to 14.4 with both extra losses; separately, the multimodal AVT++ ensemble scores 16.7 on the challenge action metric, versus 13.4 for ICL+SJTU.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.