arxivcs.CV2026-07-23
Self-Supervised Learning of Structured Dynamics from Videos
Lukas Knobel, Andrew Zisserman, Yuki M. Asano
Understanding motion in video is a fundamental challenge for visual learning, as frame-to-frame change entangles two sources of dynamics: camera motion and object motion. This decomposition has remained underexplored in representation learning, partly because these factors are ti…