Google's PointDiT Recovers 3D Shapes from Single Images

Google introduced PointDiT, a pixel-space diffusion Transformer that recovers complex 3D shapes from a single RGB image. By directly processing 3D point patches with DINOv3 tokens, it outperforms complex latent space models in monocular geometry estimation.

2026-07-08 ~ 2026-07-09 · 2 related posts

• PointDiT: Pixel-Space Diffusion Transformer for Monocular Geometry — google · 2026-07-08
• PointDiT Recovers 3D Shapes from Single Images — maier_ak · 2026-07-09