Reproduce “Lightweight Event-based Optical Flow Estimation via Iterative Deblurring” 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
Yilun Wu, Federico Paredes-Vallés, Guido C. H. E. de Croon
IDNet proposes a lightweight event-based optical flow network that eliminates computationally expensive correlation volumes by leveraging an iterative deblurring scheme on event traces. The method achieves state-of-the-art accuracy on DSEC-Flow while drastically reducing memory footprint to 20MB and enabling real-time operation with 8ms latency on embedded systems like the NVIDIA Jetson Xavier NX.
Key insights
- Motion blur in raw event traces implicitly encodes the search direction for pixel correspondences, which can be iteratively reduced through deblurring to refine flow estimates.
- Iterative refinement and motion compensation of event data are crucial for accuracy, with each contributing to a significant reduction in End-Point Error (EPE).
- Learning-based forward propagation of flow (predicting future flow) for temporal coherence significantly outperforms direct flow transport, reducing artifacts and improving robustness in real-time scenarios.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.