The article discusses the challenge of predicting optical flow in occluded areas, which is when a point is imaged in the reference frame but not in the next. The authors propose a YOIO framework that uses spatiotemporal information provided by the frame pair to improve accuracy and efficiency. The framework consists of an initial flow estimator, a multiple global information extraction module, and a unified refinement module. The results show significant improvement in optical flow estimates in occluded regions in only one iteration. The method outperforms previous methods in terms of prediction accuracy and calculation time.
Publication date: 12 Jan 2024
Project Page: ?
Paper: https://arxiv.org/pdf/2401.05879
