denoising diffusion implicit model
In plain wordsA faster way to run a diffusion image generator that skips most of the steps and, for the same starting noise, always gives the same picture.
A family of sampling procedures that reuse a DDPMdenoising diffusion probabilistic model, n.: The 2020 recipe that made diffusion image generators work well: add noise to pictures over a thousand small steps, then train a network to guess the noise so it can be removed. exactly as trained: its authors showed that a whole family of non-Markovian noising processes leads to the same training loss, so one trained diffusion modeldiffusion model, n.: An AI that learns to create images, sound or video by practicing noise removal: shown pictures buried under static, it learns to clean them up, then builds new ones starting from pure static. supports many samplers, including far faster ones, with no retraining. At each step the model's noise prediction is used to estimate the clean image, which is then re-noised to the next, possibly much lower, noise level; a parameter η interpolates between fully deterministic sampling (η = 0) and DDPM-like stochastic sampling (η = 1). Deterministic DDIM can take large jumps along a subsequence of timesteps, which cut sampling time by a factor of 10 to 50 in the original experiments, and it fixes a one-to-one correspondence between starting noise and image. That correspondence allows meaningful interpolation between images in noise space and 'DDIM inversion', which recovers the noise that would regenerate a given image, a common first step in editing. DDIM corresponds to an Euler step of the probability-flow ODE.
Origin
Introduced by Jiaming Song, Chenlin Meng and Stefano Ermon of Stanford in 'Denoising Diffusion Implicit Models', posted in October 2020 and published at ICLR 2021.
Related
- diffusion samplerdiffusion sampler, n.: The step-by-step routine a diffusion model follows to turn noise into a finished picture; the choice of sampler and number of steps trades speed against quality.
- denoising diffusion probabilistic modeldenoising diffusion probabilistic model, n.: The 2020 recipe that made diffusion image generators work well: add noise to pictures over a thousand small steps, then train a network to guess the noise so it can be removed.
- score-based generative modelscore-based generative model, n.: A generator that learns, for any noisy example, which direction would make it look more like real data — then starts from pure noise and follows those directions step by step.
- instruction-based image editinginstruction-based image editing, n.: Changing a picture by telling an AI what to change — 'make it night', 'remove the car', 'give her a red coat' — instead of using editing tools by hand.
- consistency modelconsistency model, n.: A generator trained to jump from noise straight to a finished image in one step, instead of the dozens of small steps a diffusion model normally takes.
Sources
- Song, J., Meng, C. & Ermon, S. (2021). Denoising Diffusion Implicit Models. ICLR 2021. arXiv:2010.02502. arxiv.org/abs/2010.02502
Mentioned in
diffusion samplerdiffusion sampler, n.: The step-by-step routine a diffusion model follows to turn noise into a finished picture; the choice of sampler and number of steps trades speed against quality., instruction-based image editinginstruction-based image editing, n.: Changing a picture by telling an AI what to change — 'make it night', 'remove the car', 'give her a red coat' — instead of using editing tools by hand., denoising diffusion probabilistic modeldenoising diffusion probabilistic model, n.: The 2020 recipe that made diffusion image generators work well: add noise to pictures over a thousand small steps, then train a network to guess the noise so it can be removed.