NTIRE 2025 Challenge on Image Super-Resolution ($ imes$4): Methods and Results
🤖 AI Summary
This work organizes the NTIRE 2025 Image Super-Resolution (×4) Challenge, addressing the reconstruction of low-resolution images generated via bicubic downscaling. It introduces a novel dual-track evaluation framework: the *restoration* track quantifies pixel-level fidelity using PSNR, while the *perceptual* track assesses visual realism via LPIPS and related metrics—thereby jointly optimizing objective accuracy and subjective quality. Participating methods span state-of-the-art techniques, including efficient network architectures, frequency-domain modeling, and hybrid GAN-diffusion priors, all trained and evaluated uniformly on the DIV2K/LiveSR benchmarks under standardized protocols. The challenge attracted 286 researchers, with 25 teams submitting valid entries. Several approaches set new ×4 super-resolution records in both PSNR and LPIPS, establishing the most authoritative benchmark and delineating the current frontier for this task.
Application Category
📝 Abstract
This paper presents the NTIRE 2025 image super-resolution ($ imes$4) challenge, one of the associated competitions of the 10th NTIRE Workshop at CVPR 2025. The challenge aims to recover high-resolution (HR) images from low-resolution (LR) counterparts generated through bicubic downsampling with a $ imes$4 scaling factor. The objective is to develop effective network designs or solutions that achieve state-of-the-art SR performance. To reflect the dual objectives of image SR research, the challenge includes two sub-tracks: (1) a restoration track, emphasizes pixel-wise accuracy and ranks submissions based on PSNR; (2) a perceptual track, focuses on visual realism and ranks results by a perceptual score. A total of 286 participants registered for the competition, with 25 teams submitting valid entries. This report summarizes the challenge design, datasets, evaluation protocol, the main results, and methods of each team. The challenge serves as a benchmark to advance the state of the art and foster progress in image SR.
Problem
Research questions and friction points this paper is trying to address.
Recover high-resolution images from low-resolution bicubic downsampling
Develop state-of-the-art super-resolution network designs
Evaluate performance via restoration and perceptual quality tracks
Innovation
Methods, ideas, or system contributions that make the work stand out.
Bicubic downsampling for LR image generation
Dual-track evaluation: PSNR and perceptual score
Benchmarking state-of-the-art SR network designs
🔎 Similar Papers
No similar papers found.
