The Eleventh NTIRE 2026 Efficient Super-Resolution Challenge Report

2026-04-03

AI Summary

This study addresses the challenge of single-image super-resolution by systematically reducing computational overhead—including model parameters, FLOPs, and inference time—while maintaining high reconstruction quality (PSNR ≈ 26.90 dB). To this end, the authors organized and evaluated a competition focused on efficient super-resolution, attracting submissions from 15 participating teams. The proposed methods integrate lightweight network architectures with model compression and acceleration techniques, achieving a synergistic balance between performance and efficiency on the DIV2K_LSDIR validation and test sets. This work not only surveys recent advances in the field but also advances the state of the art by demonstrating effective trade-offs between model compactness and high-fidelity reconstruction, representing the current frontier in efficient super-resolution.

Abstract

This paper reviews the NTIRE 2026 challenge on efficient single-image super-resolution with a focus on the proposed solutions and results. The aim of this challenge is to devise a network that reduces one or several aspects, such as runtime, parameters, and FLOPs, while maintaining PSNR of around 26.90 dB on the DIV2K_LSDIR_valid dataset, and 26.99 dB on the DIV2K_LSDIR_test dataset. The challenge had 95 registered participants, and 15 teams made valid submissions. They gauge the state-of-the-art results for efficient single-image super-resolution.

Problem

• efficient super-resolution
• single-image super-resolution
• model complexity
• PSNR
• computational efficiency

Innovation

• efficient super-resolution
• single-image super-resolution
• model efficiency
• PSNR
• NTIRE challenge