Skip to content

Diffusion with Forward Models

Sample unseen signals through known forward models

Visit official site

What Diffusion with Forward Models is

The model is described as a denoising diffusion probabilistic approach that samples from unobserved signal distributions measured by a known differentiable forward model.

What you can do with it

  • Sample from partially observed unknown signals
  • Use a known forward model as the measurement
  • Apply the approach to computer vision tasks
  • Generate 3D scene distributions in inverse graphics

Application scenarios

  • Rebuild a 3D scene from a single 2D image
  • Apply the method to denoising work
  • Explore other computer vision tasks

Pros & cons

✓ What we like

  • Partially observed signals can be sampled directly
  • The material describes pricing as flexible

! What to watch out for

  • Measurement relies on a known forward model
  • The described scope is image processing and computer vision

FAQ

How are signals observed?

The material says signals are measured through a known differentiable forward model.

What does inverse graphics produce?

A distribution of 3D scenes consistent with a single 2D input image.

Last reviewed: 2026-11-15

More AI image generation tools

View all →

How we review