Structured Dynamics Model for Decoupling Camera and Object Motion
July 22, 2026
The Structured Dynamics Model (SDM) separates camera motion from object motion by predicting future features from frozen ViT representations. This approach recovers robust motion representations by explicitly decomposing the dominant source of temporal change from residual dynamics.
HOW THIS AFFECTS YOU
●
researcherYou can improve motion representation learning by explicitly decomposing camera and object dynamics in video.