SupResDiffGAN a new approach for the Super-Resolution task

2025-04-18

AI Summary

Generative adversarial networks (GANs) and diffusion models struggle to jointly achieve high inference efficiency and reconstruction fidelity in high-resolution image super-resolution. Method: This paper proposes a GAN-diffusion hybrid architecture that jointly models the generative process in latent space, incorporates an adaptive noise corruption mechanism to mitigate discriminator overfitting, and—uniquely—reduces the number of diffusion steps to single digits while preserving perceptual quality. Contribution/Results: The method outperforms SR3 and I²SB across PSNR, SSIM, and LPIPS metrics on standard benchmarks. It achieves near real-time inference speed without compromising perceptual quality, thereby bridging, for the first time, the long-standing speed–quality trade-off gap between diffusion models and GANs.

Technology Category

• Computer Vision: Generative Adversarial Networks (GANs) for Vision
• Machine Learning: Deep Generative Models & Autoencoders
• Natural Language Processing: Generation

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Social Networks and Social Media: Generative AI / large language models and their impact on social systems

Abstract

In this work, we present SupResDiffGAN, a novel hybrid architecture that combines the strengths of Generative Adversarial Networks (GANs) and diffusion models for super-resolution tasks. By leveraging latent space representations and reducing the number of diffusion steps, SupResDiffGAN achieves significantly faster inference times than other diffusion-based super-resolution models while maintaining competitive perceptual quality. To prevent discriminator overfitting, we propose adaptive noise corruption, ensuring a stable balance between the generator and the discriminator during training. Extensive experiments on benchmark datasets show that our approach outperforms traditional diffusion models such as SR3 and I$^2$SB in efficiency and image quality. This work bridges the performance gap between diffusion- and GAN-based methods, laying the foundation for real-time applications of diffusion models in high-resolution image generation.

Problem

• Combines GANs and diffusion models for super-resolution
• Reduces diffusion steps for faster inference times
• Prevents discriminator overfitting with adaptive noise

Innovation

• Hybrid GAN-diffusion architecture for super-resolution
• Reduced diffusion steps for faster inference
• Adaptive noise corruption prevents discriminator overfitting