Jan, 2024

改善基于扩散的图像合成

TL;DRDiffusion models for image synthesis are improved using ConPreDiff, a method that incorporates context prediction, achieving superior results in unconditional image generation, text-to-image generation, and image inpainting.