IIDM: Image-to-Image Diffusion Model for Semantic Image Synthesis
This addresses the problem of generating high-quality images from semantic conditions for applications like computer vision and graphics, representing an incremental improvement over GAN-based methods.
The paper tackles semantic image synthesis by treating it as an image denoising task using a novel image-to-image diffusion model (IIDM), which outperforms existing state-of-the-art methods by clear margins.
Semantic image synthesis aims to generate high-quality images given semantic conditions, i.e. segmentation masks and style reference images. Existing methods widely adopt generative adversarial networks (GANs). GANs take all conditional inputs and directly synthesize images in a single forward step. In this paper, semantic image synthesis is treated as an image denoising task and is handled with a novel image-to-image diffusion model (IIDM). Specifically, the style reference is first contaminated with random noise and then progressively denoised by IIDM, guided by segmentation masks. Moreover, three techniques, refinement, color-transfer and model ensembles, are proposed to further boost the generation quality. They are plug-in inference modules and do not require additional training. Extensive experiments show that our IIDM outperforms existing state-of-the-art methods by clear margins. Further analysis is provided via detailed demonstrations. We have implemented IIDM based on the Jittor framework; code is available at https://github.com/ader47/jittor-jieke-semantic_images_synthesis.