DDPM

Denoising Diffusion Probabilistic Models

Jonathan Ho, Ajay Jain, Pieter Abbeel · NeurIPS 2020

Read paper · arXiv:2006.11239

Build a generative model from a forward noising process and a learned reverse denoising process.

What you will implement

Implement closed-form forward noising, the Gaussian posterior, and an epsilon-parameterized reverse step; schedules and sampling loops are supplied.

• 6. Closed-Form Forward Noising

  Implement the DDPM forward marginal to sample any noise level directly.

  Implement q_sample only. Schedule construction is provided; the cosine schedule is an Improved DDPM extension.

• 8. Posterior and Reverse Denoising Step

  Implement the DDPM Gaussian posterior and convert a noise prediction into one reverse update.

  Implement q_posterior and p_sample_step. The sampling loop and the test noise predictor are provided.