Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

The Tenth NTIRE 2025 Image Denoising Challenge Report

Lei Sun, Hang Guo, Bin Ren, Luc Van Gool, Radu Timofte, Yawei Li, Xiangyu Kong, Hyunhee Park, Xiaoxuan Yu, Suejin Han, Hakjae Jeon, Jia Li, Hyung-Ju Chun, Donghun Ryou, Inju Ha, Bohyung Han, Jingyu Ma, Zhijuan Huang, Huiyuan Fu, Hongyuan Yu, Boqi Zhang, Jiawei Shi, Heng Zhang, Huadong Ma, Deepak Kumar Tyagi, Aman Kukretti, Gajender Sharma, Sriharsha Koundinya, Asim Manna, Jun Cheng, Shan Tan, Jun Liu, Jiangwei Hao, Jianping Luo, Jie Lu, Satya Narayan Tazi, Arnim Gautam, Aditi Pawar, Aishwarya Joshi, Akshay Dudhane, Praful Hambadre, Sachin Chaudhary, Santosh Kumar Vipparthi, Subrahmanyam Murala, Jiachen Tu, Nikhil Akalwadi, Vijayalaxmi Ashok Aralikatti, Dheeraj Damodar Hegde, G Gyaneshwar Rao, Jatin Kalal, Chaitra Desai, Ramesh Ashok Tabib, Uma Mudenagudi, Zhenyuan Lin, Yubo Dong, Weikun Li, Anqi Li, Ang Gao, Weijun Yuan, Zhan Li, Ruting Deng, Yihang Chen, Yifan Deng, Zhanglu Chen, Boyang Yao, Shuling Zheng, Feng Zhang, Zhiheng Fu, Anas M. Ali, Bilel Benjdira, Wadii Boulila, Jan Seny, Pei Zhou, Jianhua Hu, K. L. Eddie Law, Jaeho Lee, M. J. Aashik Rasool, Abdur Rehman, SMA Sharif, Seongwan Kim, Alexandru Brateanu, Raul Balmez, Ciprian Orhei, Cosmin Ancuti, Zeyu Xiao, Zhuoyuan Li, Ziqi Wang, Yanyan Wei, Fei Wang, Kun Li, Shengeng Tang, Yunkai Zhang, Weirun Zhou, Haoxuan Lu

TL;DR

The NTIRE 2025 Image Denoising Challenge investigates high-quality denoising under AWGN with a fixed noise level of $σ=50$, using DIV2K and LSDIR with PSNR/SSIM evaluation. The paper surveys a wide array of approaches, from hybrid transformer-CNN architectures and self-ensembling to diffusion-based and prompt-driven restoration methods, highlighting three key themes: hybrid models for global and local feature fusion, data-driven training strategies including self-supervised pretraining (Generalized Denoising Score Matching), and advanced inference techniques such as geometric self-ensembling and adaptive patch processing. The main contributions include a detailed benchmark of methods, novel training and architectural innovations (e.g., time-conditioned diffusion, EMAPM prompts, and CATB blocks), and practical insights into fairness constraints and dataset usage. Collectively, the results establish new state-of-the-art baselines for denoising under realistic noise and offer a diverse set of techniques for robust, scalable restoration in real-world imaging pipelines.

Abstract

This paper presents an overview of the NTIRE 2025 Image Denoising Challenge (σ = 50), highlighting the proposed methodologies and corresponding results. The primary objective is to develop a network architecture capable of achieving high-quality denoising performance, quantitatively evaluated using PSNR, without constraints on computational complexity or model size. The task assumes independent additive white Gaussian noise (AWGN) with a fixed noise level of 50. A total of 290 participants registered for the challenge, with 20 teams successfully submitting valid results, providing insights into the current state-of-the-art in image denoising.

The Tenth NTIRE 2025 Image Denoising Challenge Report

TL;DR

The NTIRE 2025 Image Denoising Challenge investigates high-quality denoising under AWGN with a fixed noise level of , using DIV2K and LSDIR with PSNR/SSIM evaluation. The paper surveys a wide array of approaches, from hybrid transformer-CNN architectures and self-ensembling to diffusion-based and prompt-driven restoration methods, highlighting three key themes: hybrid models for global and local feature fusion, data-driven training strategies including self-supervised pretraining (Generalized Denoising Score Matching), and advanced inference techniques such as geometric self-ensembling and adaptive patch processing. The main contributions include a detailed benchmark of methods, novel training and architectural innovations (e.g., time-conditioned diffusion, EMAPM prompts, and CATB blocks), and practical insights into fairness constraints and dataset usage. Collectively, the results establish new state-of-the-art baselines for denoising under realistic noise and offer a diverse set of techniques for robust, scalable restoration in real-world imaging pipelines.

Abstract

This paper presents an overview of the NTIRE 2025 Image Denoising Challenge (σ = 50), highlighting the proposed methodologies and corresponding results. The primary objective is to develop a network architecture capable of achieving high-quality denoising performance, quantitatively evaluated using PSNR, without constraints on computational complexity or model size. The task assumes independent additive white Gaussian noise (AWGN) with a fixed noise level of 50. A total of 290 participants registered for the challenge, with 20 teams successfully submitting valid results, providing insights into the current state-of-the-art in image denoising.

Paper Structure

This paper contains 54 sections, 29 equations, 18 figures, 2 tables, 2 algorithms.

Table of Contents

  1. Introduction
  2. NTIRE 2025 Image Denoising Challenge
  3. Dataset
  4. Tracks and Competition
  5. Challenge Results
  6. Participants
  7. Main Ideas and Architectures
  8. Fairness
  9. Challenge Methods and Teams
  10. Samsung MX (Mobile eXperience) Business & Samsung R&D Institute China - Beijing (SRC-B)
  11. Model Framework
  12. Dataset and Training Strategy
  13. SNUCV
  14. BuptMM
  15. HMiDenoise
  16. ...and 39 more sections

Figures (18)

  • Figure 1: Framework of the hybrid network proposed by Team SRC-B.
  • Figure 2: The overview of the deep ensemble pipeline proposed by Team SNUCV.
  • Figure 3: The model architecture of DDU proposed by Team BuptMM.
  • Figure 4: Model architecture of DB-HAT proposed by Team HMiDenoise.
  • Figure 5: Block Diagram for Image Denoising using Restormer architecture along with Hard data mining and Ensemble Techniques (Team Pixel Purifiers).
  • ...and 13 more figures