RPBA-Net: An Interpretable Residual Pyramid Bilateral Affine Network for RAW-Domain ISP Enhancement

2026-05-05

AI Summary

This work addresses the challenges of fragmented modules, uninterpretable mappings, and limited deployability in RAW-domain image signal processing by proposing RPBA-Net, a unified framework for demosaicking and detail enhancement via residual affine basis reconstruction. The method employs a pyramid bilateral affine grid integrated with guidance-driven autoregressive adaptive slicing and cross-layer fusion to hierarchically model global tone recovery and local texture enhancement. Smoothness and cross-scale consistency regularizations are introduced to further improve reconstruction quality. RPBA-Net achieves significant gains in both fidelity and perceptual quality over existing RAW-to-sRGB methods while maintaining low model complexity, demonstrating strong potential for efficient deployment on mobile platforms.

Abstract

To address module fragmentation, uninterpretable mappings, and deployment constraints in RAW-domain demosaicing, color correction, and detail enhancement, this paper proposes RPBA-Net, an interpretable residual pyramid bilateral affine network for RAW-domain ISP enhancement. Given packed RAW as input, the method performs residual affine base reconstruction by estimating a base RGB representation and learning identity-guided residual affine corrections, thereby unifying demosaicing and enhancement. It further builds pyramid bilateral affine grids and combines guide-driven autoregressive adaptive slicing with adaptive cross-layer fusion to hierarchically model global tone restoration and local texture enhancement. In addition, smoothness, cross-scale consistency, and magnitude regularization terms are introduced to improve model stability, controllability, and structural interpretability. Extensive experiments demonstrate that RPBA-Net surpasses representative RAW-to-sRGB methods and achieves state-of-the-art performance in reconstruction fidelity and perceptual quality, while maintaining low model complexity and strong deployment potential for mobile and embedded platforms.

Problem

• RAW-domain ISP
• module fragmentation
• uninterpretable mappings
• deployment constraints
• image enhancement

Innovation

• interpretable network
• residual pyramid bilateral affine
• RAW-domain ISP
• adaptive cross-layer fusion
• affine correction