Skip to content

OccFusion

Render a full human body from partial views

Visit official site

What OccFusion is

The technique is described as a way to render a complete human image with 3D Gaussian splatting and a pretrained 2D diffusion model even when parts are occluded.

What you can do with it

  • Generate a complete body mask from a partial visibility mask
  • Optimize 3D body Gaussians with Score-Distillation Sampling
  • Improve rarely seen parts through context interpolation

Getting started

  • Prepare a partly visible human video input
  • Generate the full body mask in the initialization stage
  • Apply optimization and then refinement

Pros & cons

✓ What we like

  • Handles partly hidden human input
  • Evaluation sequences named include ZJU-MoCap and OcMotion

! What to watch out for

  • The pipeline starts from a partly visible human video
  • Rendering relies on a pretrained 2D diffusion model

FAQ

What are the three stages?

The material names initialization, optimization and refinement.

What input is needed?

A video with a partly visible human body is described as the input.

Last reviewed: 2026-11-15

More AI image generation tools

View all →

How we review