Abstract: Contemporary deep learning approaches for optical flow estimation continue to face persistent challenges in model interpretability, generalization capacity, and deployment efficiency, ...