We can now use object movement as a 'super-signal' to perfectly separate light from matter in computer vision.
April 15, 2026
Original Paper
LumiMotion: Improving Gaussian Relighting with Scene Dynamics
arXiv · 2604.10994
The Takeaway
LumiMotion turns the 'problem' of motion in a scene into a tool for solving the hardest problem in vision: disentangling an object's real color (albedo) from the light hitting it. By using scene dynamics as a supervisory signal, the model achieves superior relighting performance. Previously, relighting required complex, static setups or massive datasets. Now, 'movement' provides the ground truth. This unlocks realistic, dynamic relighting for AR/VR and movie production, allowing for near-perfect digital 're-lighting' of moving objects without any specialized hardware.
From the abstract
In 3D reconstruction, the problem of inverse rendering, namely recovering the illumination of the scene and the material properties, is fundamental. Existing Gaussian Splatting-based methods primarily target static scenes and often assume simplified or moderate lighting to avoid entangling shadows with surface appearance. This limits their ability to accurately separate lighting effects from material properties, particularly in real-world conditions. We address this limitation by leveraging dyna