arxivcs.CV2026-06-27
SciFlow: Semantic Cross Interference for Self-Supervised Optical Flow Domain Generalization
Jamie Menjay Lin, Jisoo Jeong, Hong Cai, Kai Wang, Fatih Porikli
Motions of objects and scenes carry essential intelligence in video understanding, offering rich cues for interpreting dynamic settings and interactions. Due to the cost and scarcity of high-quality annotation or ground truth of pixel-wise optical flow, however, motion estimation…