GGT-100K: Generative Ground Truth for Generalizable Real-World Image Restoration

2026-05-29

AI Summary

This work addresses the scarcity of high-quality paired data for real-world image restoration, a challenge exacerbated by the inability of existing synthetic data to accurately model complex real-world degradations. To overcome this limitation, the authors propose leveraging multimodal foundation models—such as Nano-Banana-2—combined with a vision-language-model-driven adaptive prompting strategy to generate high-fidelity, content-faithful “Generative Ground Truth” (GGT) from real low-quality images. Through a multi-stage quality control pipeline, they construct GGT-100K, a large-scale real paired dataset comprising 103,707 training pairs and 500 test pairs. Experimental results demonstrate that models trained or fine-tuned on GGT-100K exhibit significantly improved generalization in real-world scenarios, with particularly pronounced gains for generative restoration architectures.

Abstract

Real-world image restoration (IR) is bottlenecked by the scarcity of high-quality paired training data. Synthetic datasets are abundant but often fail to model real-world degradations, while real-world paired datasets are expensive and difficult to capture. As a result, IR models trained on these datasets show limited generalization in real-world scenarios. In this work, we propose Generative Ground Truth (GGT) by using generative multimodal foundation models (MFMs) to produce high-quality (HQ) targets from real-world low-quality (LQ) images. We first conduct a systematic evaluation of nine state-of-the-art MFMs, including Nano-Banana-2 and GPT-Image-2, on images of various scenes and degradation types. The results demonstrate that Nano-Banana-2 with VLM-based adaptive prompting shows the highest capability to synthesize perceptually realistic and content-faithful HQ targets, which can serve as the GGT for the LQ input. We then employ Nano-Banana-2 to build a GGT synthesis pipeline, which involves multi-stage quality control to ensure data reliability, and construct GGT-100K, an LQ-HQ paired dataset comprising 103,707 training pairs and covering diverse scenes and complex real-world degradations. A test set of 500 image pairs is also established. Extensive experiments show that GGT-100K consistently improves the real-world generalization of a wide range of IR models, with particularly strong benefits for finetuning generative models for IR tasks. Our results suggest that MFMs can serve as practical tools for restoration-oriented data generation, and GGT-100K is a useful resource to expand the generalization boundaries of real-world IR models.

Problem

• real-world image restoration
• paired training data
• data scarcity
• generalization
• image degradation

Innovation

• Generative Ground Truth
• Multimodal Foundation Models
• Real-World Image Restoration
• Adaptive Prompting
• Synthetic Paired Dataset