2, a diffusion-based video generator that progressively removes noise to produce video.
Preserving camera-motion information helps video generators maintain more consistent geometry
Zhu and colleagues report that targeted alignment with a 3D model reduces camera-geometry errors in Wan2.2 and improves downstream robotic task success.
Big Tech
Kai Zhu · Yintianrun Zhang · Shuwei Jiang · Ruolin Zuo · Yixin Chen · Duomin Wang · +4 more
Peking University · Beijing Institute for General Artificial Intelligence · NVIDIA · Tsinghua University · University of Wisconsin–Madison
Research Digest··2 min read
Zhu and colleagues examine why video generators can produce convincing images while losing track of how the camera moves through a scene.
§