The Eleventh NTIRE 2026 Efficient Super-Resolution Challenge Report

Forum topic · 小凯 · 2026-04-06

Summary

This paper presents the report of the NTIRE 2026 Challenge on Efficient Single-Image Super-Resolution, held as part of the New Trends in Image Restoration and Enhancement (NTIRE) workshop series. The challenge asked participants to design networks that reduce one or more efficiency metrics—runtime, parameter count, and FLOPs—while maintaining super-resolution quality of approximately 26.90 dB PSNR on the DIV2K_LSDIR_valid dataset and 26.99 dB PSNR on the DIV2K_LSDIR_test dataset. The competition attracted 95 registered participants, of which 15 teams made valid submissions. The report reviews the proposed solutions and results, benchmarking the state of the art in efficient single-image super-resolution and highlighting trade-offs between model efficiency and reconstruction fidelity.

Overview

Field: Computer Vision Authors: Bin Ren, Hang Guo, Yan Shu, et al. Published: 2026-04-03 arXiv: 2604.03198

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 the challenge is to devise a network that reduces one or several aspects, such as runtime, parameters, and FLOPs, while maintaining a 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. Together, these submissions gauge the state-of-the-art results for efficient single-image super-resolution.

Key Details

Task: Efficient single-image super-resolution
Objective: Minimize runtime, parameter count, and/or FLOPs while sustaining target PSNR (~26.90 dB on DIV2K_LSDIR_valid; ~26.99 dB on DIV2K_LSDIR_test)
Participation: 95 registered participants; 15 teams with valid submissions
Outcome: Benchmark of state-of-the-art efficient super-resolution methods

---

*Auto-collected on 2026-04-06*