Reproduce “REMAP: Regularized Matching and Partial Alignment of Video Embeddings” 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
Soumyadeep Chandra, Kaushik Roy
REMAP learns shared procedural steps across videos by combining visual similarity with temporal relationships and allowing some frames to remain unmatched through a virtual sink. On EgoProceL, it reports 64.2 F1 on EGTEA-GAZE+, versus 37.4 for RGWOT, but trails RGWOT on PC Assembly and PC Disassembly. Results improve on several benchmarks, but do not establish that partial alignment alone explains the gains.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.