D2C: Diffusion-Denoising Models for Few-shot Conditional Generation

TL;DR

D2C introduces a diffusion-denoising VAE framework that enables high-quality, fast, few-shot conditional image generation and manipulation, outperforming existing models in accuracy and speed.

Contribution

The paper presents a novel D2C model combining diffusion-based priors and contrastive learning for efficient few-shot conditional image generation.

Findings

Outperforms state-of-the-art VAEs and diffusion models on conditional generation.

Generates images two orders of magnitude faster than StyleGAN2.

Achieves 50-60% human preference in image manipulation tasks.

Abstract

Conditional generative models of high-dimensional images have many applications, but supervision signals from conditions to images can be expensive to acquire. This paper describes Diffusion-Decoding models with Contrastive representations (D2C), a paradigm for training unconditional variational autoencoders (VAEs) for few-shot conditional image generation. D2C uses a learned diffusion-based prior over the latent representations to improve generation and contrastive self-supervised learning to improve representation quality. D2C can adapt to novel generation tasks conditioned on labels or manipulation constraints, by learning from as few as 100 labeled examples. On conditional generation from new labels, D2C achieves superior performance over state-of-the-art VAEs and diffusion models. On conditional image manipulation, D2C generations are two orders of magnitude faster to produce over StyleGAN2 ones and are preferred by 50% - 60% of the human evaluators in a double-blind study.