zero123
Discover advancements in converting single images to 3D models using zero-shot learning techniques. This approach mitigates the Janus problem by employing camera modeling and using synthetic datasets such as Objaverse. Recent updates, including Zero123-XL and integrations with Threestudio, improve view synthesis and 3D reconstruction. Explore live demos, optimized codebases for RTX GPUs, and training scripts. This project is a collaboration between Columbia University and Toyota Research Institute, utilizing technology from Stable Diffusion and NeRDi.
