Skip to content

DiffusionMat

A diffusion framework for coarse-to-fine alpha matting

Visit official site

What DiffusionMat is

The framework is described as an image matting method that uses a diffusion model to refine a coarse trimap into a clean alpha matte through iterative denoising.

What you can do with it

  • Matte people in natural scenes
  • Segment and cut out objects in natural images
  • Work on translucent object matting

How it works

  • Noise is added to a trimap at the start
  • A pretrained diffusion model denoises step by step
  • A correction module adjusts each denoising output

Pros & cons

✓ What we like

  • A dedicated loss targets alpha edge accuracy
  • The correction module keeps structure consistent with the input

! What to watch out for

  • The process starts from a noisy trimap
  • Refinement runs step by step through denoising

FAQ

What is the key module?

A correction module that adjusts output at every denoising step.

What does the loss target?

Alpha edge accuracy and consistency in opaque and transparent areas.

Last reviewed: 2026-11-15

More AI image generation tools

View all →

How we review