Beyond Model Design: Data-Centric Training and Self-Ensemble for Gaussian Color Image Denoising

2026-04-13

AI Summary

This work addresses the task of Gaussian color image denoising at a high noise level (σ = 50) by proposing a data-centric optimization framework that operates without modifying the Restormer backbone architecture. By leveraging joint training across multiple datasets, a two-stage optimization strategy, ×8 geometric self-ensemble, and TLC-based local inference encapsulation, the approach fully exploits the model’s inherent capacity. Evaluated on the NTIRE 2026 validation set, the method achieves a PSNR of 30.762 dB and an SSIM of 0.861, representing a substantial improvement of 3.366 dB over the original Restormer and significantly surpassing the performance limits of the existing architecture.

Technology Category

Computer Vision: Learning & Optimization for CVSearch and Optimization: Learning to SearchMachine Learning: Optimization

Application Category

Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for rankingEconomics, Online Markets and Human Computation: LLM based quality controls for crowd workSemantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

This paper presents our solution to the NTIRE 2026 Image Denoising Challenge (Gaussian color image denoising at fixed noise level $σ= 50$). Rather than proposing a new restoration backbone, we revisit the performance boundary of the mature Restormer architecture from two complementary directions: stronger data-centric training and more complete Test-Time capability release. Starting from the public Restormer $σ\!=\!50$ baseline, we expand the standard multi-dataset training recipe with larger and more diverse public image corpora and organize optimization into two stages. At inference, we apply $\times 8$ geometric self-ensemble to further release model capacity. A TLC-style local inference wrapper is retained for implementation consistency; however, systematic ablation reveals its quantitative contribution to be negligible in this setting. On the challenge validation set of 100 images, our final submission achieves 30.762 dB PSNR and 0.861 SSIM, improving over the public Restormer $σ\!=\!50$ pretrained baseline by up to 3.366 dB PSNR. Ablation studies show that the dominant gain originates from the expanded training corpus and the two-stage optimization schedule, and self-ensemble provides marginal but consistent improvement.

Problem

• Gaussian color image denoising
• data-centric training
• self-ensemble
• Restormer
• NTIRE challenge

Innovation

• data-centric training
• self-ensemble
• two-stage optimization
• Restormer
• image denoising