Reproduce “UTOPI: Efficient Egocentric Long-Video Understanding in AR via User-Guided Token Pre-Compression” 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
Ziqi Wang, Su Chen, Qiance Tang, Jieyu Lin, Ziyun Li, Barbara De Salvo, Sai Qian Zhang
UTOPI compresses egocentric video tokens before later questions arrive, using gaze to retain attended content and head motion to merge repeated views across frames. On EgoSchema, Qwen2.5VL-7B retains 5% of visual tokens and reaches 61.4% accuracy, versus 60.1% uncompressed. At the same budget on EgoEye, accuracy falls to 53.4% from 58.4%, limiting the no-loss result to EgoSchema.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.