LEGO: A Lifting-Free Approach for Exocentric-to-Egocentric Video Generation
LEGO proposes a lifting-free method to generate egocentric videos from single exocentric recordings by replacing explicit scene reconstruction with a learned view synthesizer. This synthesizer renders the target egocentric view directly, conditioning a video diffusion model on this render and its confidence map instead of depth maps or point clouds. The approach prioritizes structural alignment over sharpness, leading to better results without retraining.