Learning Arbitrary-Scale RAW Image Downscaling with Wavelet-based Recurrent Reconstruction

2025-07-30

AI Summary

Existing learning-based image downscaling methods predominantly operate in the sRGB domain, leading to detail blurring and artifacts; while RAW images preserve photon-level information, dedicated differentiable downscaling frameworks for RAW data remain absent. To address this, we propose the first arbitrary-scale (including non-integer ratios) downscaling method tailored for the RAW domain. Our approach introduces a wavelet-based lossless transform and a cyclic reconstruction framework that decouples low-frequency arbitrary-scale downscaling (LASDM) from high-frequency prediction (HFPM), augmented by an energy-maximization loss to enforce structural consistency in high-frequency components. The method significantly improves detail fidelity and reconstruction accuracy, outperforming state-of-the-art approaches quantitatively and visually. Furthermore, we release Real-NIRD—the first real-world RAW dataset supporting non-integer downscaling—thereby establishing a benchmark and advancing research on differentiable downscaling in the RAW domain.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Hardware-aware ML
• Search and Optimization: Mixed Discrete/Continuous Search

Application Category

• Search and Retrieval-Augmented AI: Efficiency and scalability of Web search engines
• Responsible Web: Data and user privacy-enhancing technologies for the Web
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications

Abstract

Image downscaling is critical for efficient storage and transmission of high-resolution (HR) images. Existing learning-based methods focus on performing downscaling within the sRGB domain, which typically suffers from blurred details and unexpected artifacts. RAW images, with their unprocessed photonic information, offer greater flexibility but lack specialized downscaling frameworks. In this paper, we propose a wavelet-based recurrent reconstruction framework that leverages the information lossless attribute of wavelet transformation to fulfill the arbitrary-scale RAW image downscaling in a coarse-to-fine manner, in which the Low-Frequency Arbitrary-Scale Downscaling Module (LASDM) and the High-Frequency Prediction Module (HFPM) are proposed to preserve structural and textural integrity of the reconstructed low-resolution (LR) RAW images, alongside an energy-maximization loss to align high-frequency energy between HR and LR domain. Furthermore, we introduce the Realistic Non-Integer RAW Downscaling (Real-NIRD) dataset, featuring a non-integer downscaling factor of 1.3$ imes$, and incorporate it with publicly available datasets with integer factors (2$ imes$, 3$ imes$, 4$ imes$) for comprehensive benchmarking arbitrary-scale image downscaling purposes. Extensive experiments demonstrate that our method outperforms existing state-of-the-art competitors both quantitatively and visually. The code and dataset will be released at https://github.com/RenYangSCU/ASRD.

Problem

• Develops RAW image downscaling preserving structural integrity
• Addresses blurred details in sRGB domain downscaling
• Introduces dataset for arbitrary-scale downscaling benchmarking

Innovation

• Wavelet-based recurrent reconstruction for RAW downscaling
• Coarse-to-fine LASDM and HFPM modules
• Energy-maximization loss aligns HR-LR domains