Reproduce “Match-Any-Events: Zero-Shot Motion-Robust Feature Matching Across Wide Baselines for Event Cameras” 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
Ruijun Zhang, Hang Su, Kostas Daniilidis, Ziyun Wang
Match-Any-Events is a framework enabling zero-shot wide-baseline feature matching for event cameras, supporting both event-to-event and event-to-image tasks. This approach improved event-to-event matching AUC@5° on the ECM dataset to 54.61, representing a 2.93x increase over prior event-based methods, and demonstrated robustness to diverse motion profiles.
Key insights
- The combination of logarithmically windowed event voxel representation and the Temporal Aggregation Transformer (TAg) effectively learns motion-robust, multi-timescale features from sparse event streams.
- Adaptive sparsity-aware token selection is crucial for achieving computational efficiency in event-based transformers, enabling scalable processing without sacrificing matching performance.
- Zero-shot wide-baseline generalization for event cameras is achievable by training on large-scale synthetic datasets augmented with diverse viewpoints and motion profiles.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.