Multi-hypotheses Conditioned Point Cloud Diffusion for 3D Human Reconstruction from Occluded Images

While implicit function methods capture detailed clothed shapes, they require aligned shape priors and or are weak at inpainting occluded regions given an image input.