arXiv 2003.12039
RAFT: Recurrent All-Pairs Field Transforms for Optical Flow
By Zachary Teed and Jia Deng
Published 2020-03-26
Mindmap
Browse the paper's core ideas, clusters, and relationships in a structured outline.
We introduce Recurrent All-Pairs Field Transforms (RAFT), a new deep network architecture for optical flow. RAFT extracts per-pixel features, builds multi-scale 4D correlation volumes for all pairs of pixels, and iteratively updates a flow field through a recurrent unit that performs lookups on the correlation volumes. RAFT achieves state-of-the-art performance. On KITTI, RAFT achieves an F1-all error of 5.10%, a 16…