Reproduce “Models as Tools: An Agentic Coordination Framework for Unified Multimodal Visual Tracking” 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
Wenrui Cai, Yuzhe Li, Qingjie Liu, Yunhong Wang
ACTrack coordinates specialized models for single-object video tracking, using MCITrack for routine matching, SAM3 for motion and instance evidence, and a vision-language model after sustained conflict. On LaSOT’s long-term RGB test split, ACTrack-B224 reaches 80.1 AUC versus 75.3 for matching alone, a 4.8-point gain. This benchmark ablation supports the configuration, not universal superiority or VLM-only causation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.