WAFT: Warping-Alone Field Transforms for Optical Flow (2025)

>"Optical flow is a fundamental low-level vision task that estimates per-pixel 2D motion between video frames. It has many downstream applications, including 3D reconstruction and synthesis [...] action recognition [...] frame interpolation [...] and autonomous driving..."

[...]

>"WAFT is similar to RAFT but replaces cost volume with high-resolution warping, achieving better accuracy with lower memory cost."

[...]

>"Compared with existing methods, WAFT ranks 1st on Spring, Sintel, and KITTI benchmarks, achieves the best zero-shot generalization on KITTI, while being

1.3 − 4.1× faster than existing methods

that have competitive accuracy (e.g., 1.3× than Flowformer++, 4.1× than CCMR+)."