Human Body Restoration with One-Step Diffusion Model and A New Benchmark

Human body restoration, as a specific application of image restoration, is widely applied in practice and plays a vital role across diverse fields. However, thorough research remains difficult, particularly due to the lack of benchmark datasets. In this study, we propose a high-quality dataset automated cropping and filtering (HQ-ACF) pipeline. This pipeline leverages existing object detection datasets and other unlabeled images to automatically crop and filter high-quality human images. Using this pipeline, we constructed a person-based restoration with sophisticated objects and natural activities (PERSONA) dataset, which includes training, validation, and test sets. The dataset significantly surpasses other human-related datasets in both quality and content richness. Finally, we propose OSDHuman, a novel one-step diffusion model for human body restoration. Specifically, we propose a high-fidelity image embedder (HFIE) as the prompt generator to better guide the model with low-quality human image information, effectively avoiding misleading prompts. Experimental results show that OSDHuman outperforms existing methods in both visual quality and quantitative metrics. The dataset and code are available at: https://github.com/gobunu/OSDHuman.

In everyday life, we often capture human photos that are blurry, noisy, or low in resolution, making it difficult for both viewers and machines to perceive details clearly. Fixing such flaws usually requires large collections of clean before-and-after examples, but no substantial public dataset of high-quality human images existed before our work. To address this, we developed an automated system that scans existing image repositories and retains only sharp, well-framed pictures of people. Using this tool, we built a new dataset called PERSONA, which surpasses prior human-related datasets in both scale and diversity, and is freely available to the public. We also propose OSDHuman, a fast, one-step AI model designed to restore degraded human images more efficiently than traditional multi-step approaches. OSDHuman uses a high-fidelity prompt generator to extract useful cues from low-quality images while filtering out noise, effectively guiding the restoration process. Our model demonstrates superior performance in facial clarity, clothing textures, and body structure accuracy across both synthetic and real-world scenarios. Due to its efficiency and low computational requirements, it has potential applications in smartphones, surveillance systems, and creative tools. We hope these contributions advance research in human image restoration and enhance everyday visual experiences.