Reproduce “Forget, Anticipate and Adapt: Test Time Training for Long Videos” 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
Rajat Modi, Sebastian Noel, Xin Liang, Yogesh Singh Rawat
FFN reduces repeated work in long-video test-time training. It restores features for an exiting frame, predicts the next frame, and adapts when surprise exceeds a threshold. On COCO-Videos, it scores 45.1 AP versus 37.6 for Online TTT-MAE, taking 0.7 versus 4.1 seconds per frame on an A100. On EpicTours, semantic segmentation scores are 49.3 versus 42.8 in tests up to three hours.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.