MaCo-GAN: Manifold-Contrastive Adversarial Learning for Single Image Super-Resolution

2026-06-03

AI Summary

This work addresses the tendency of conventional GANs in single-image super-resolution to produce hallucinatory artifacts due to discriminators prioritizing overall naturalness over conditional consistency. To mitigate this issue, the authors propose MaCo-GAN, a novel framework that introduces manifold contrastive learning into super-resolution for the first time. Specifically, it replaces the standard adversarial loss with a supervised contrastive loss and incorporates a dynamic fake sample synthesizer that generates both real and synthetic samples preserving low-resolution consistency. This design establishes a minimax contrastive game grounded in manifold structure. Notably, the method achieves a significantly improved trade-off between perceptual quality and distortion by merely substituting the adversarial loss, consistently enhancing super-resolution performance across multiple benchmarks.

Abstract

Conventional Generative Adversarial Networks (GANs) for Single Image Super-Resolution (SISR) often struggle with hallucinated artifacts, largely because standard discriminators evaluate overall image naturalness rather than strict conditional realism. To address this, we propose MaCo-GAN, a novel manifold-contrastive GAN framework that replaces the conventional adversarial loss with a supervised contrastive objective. A core component of our method is a dynamic fake sample synthesizer that transforms ground truth (GT) data into a spectrum of challenging, perceptually plausible fake images that strictly maintain low-resolution (LR) correspondence. Utilizing these synthesized samples, we establish a robust contrastive minimax game: the generator is trained to attract its predictions toward on-manifold fakes (low distortion) and repel them from off-manifold fakes (high distortion), while the discriminator optimizes the exact opposite. By simply replacing the adversarial loss of a baseline SR model with our proposed objective, we demonstrate consistent improvements in the perception-distortion trade-off across various benchmarks. Extensive ablation studies validate the effectiveness of our framework and provide deep insights into the dynamics of this conditional contrastive game.

Problem

• Single Image Super-Resolution
• Generative Adversarial Networks
• hallucinated artifacts
• conditional realism
• perception-distortion trade-off

Innovation

• Manifold-Contrastive Learning
• Single Image Super-Resolution
• Dynamic Fake Synthesizer
• Perception-Distortion Trade-off
• Contrastive Minimax Game