DDPM
A foundational class of generative models that learns to create data by slowly reversing a process that gradually adds Gaussian noise to the original inputs.
Think of It Like This
Like slowly unscrambling a television static screen pixel by pixel until a perfectly clear image eventually emerges from the random noise.
Denoising Diffusion Probabilistic Models (DDPM) operate through a parameterized Markov chain. During training, they learn to predict and subtract the exact amount of noise added at each timestep. Although highly effective at producing diverse and realistic images, they suffer from extremely slow inference times due to the many required steps.