AIM 2025 Challenge on Real-World RAW Image Denoising

TL;DR

The paper presents the AIM 2025 Real-World RAW Image Denoising Challenge, targeting camera-agnostic denoising trained on synthetic noise using real low-light RAW data from five DSLR cameras. It establishes a comprehensive benchmark with indoor paired and outdoor in-the-wild scenes, evaluated via full-reference and no-reference metrics, and enforces strict efficiency constraints to reflect real-world deployment. The contributions include diverse noise-modeling and denoising approaches (random masking, frequency-modulated networks, CFA-aware convolutions, and attention-based architectures) and multiple datasets and preprocessing pipelines that emphasize noise realism and generalization. The results demonstrate that data synthesis quality and training strategies substantially drive performance, with methods leveraging advanced noise modeling and data scaling achieving the top rankings. Overall, the challenge advances robust, practical RAW denoising suitable for consumer photography and safety-critical applications like autonomous driving by promoting camera-agnostic solutions grounded in realistic noise synthesis.

Abstract

We introduce the AIM 2025 Real-World RAW Image Denoising Challenge, aiming to advance efficient and effective denoising techniques grounded in data synthesis. The competition is built upon a newly established evaluation benchmark featuring challenging low-light noisy images captured in the wild using five different DSLR cameras. Participants are tasked with developing novel noise synthesis pipelines, network architectures, and training methodologies to achieve high performance across different camera models. Winners are determined based on a combination of performance metrics, including full-reference measures (PSNR, SSIM, LPIPS), and non-reference ones (ARNIQA, TOPIQ). By pushing the boundaries of camera-agnostic low-light RAW image denoising trained on synthetic data, the competition promotes the development of robust and practical models aligned with the rapid progress in digital photography. We expect the competition outcomes to influence multiple domains, from image restoration to night-time autonomous driving.

Figures (11)

• Figure 1: Illustration of a classical RAW Image Denoising Pipeline li2025noise.
• Figure 2: Illustration of the baseline model for noise synthesis li2025noise.
• Figure 3: Data samples from the AIM 2025 RAW Image Denoising Challenge dataset.
• Figure 4: The model framework diagram of FrENet.
• Figure 5: Team VMCL-ISP. Overview of the proposed PMNNP (Restormer).