ParamISP: Learned Forward and Inverse ISPs using Camera Parameters

TL;DR

ParamISP tackles the challenge of robust forward and inverse ISP across varying camera parameters by conditioning a modular ISP pipeline on EXIF-derived parameters. It introduces ParamNet to convert optical parameters into a modulation vector that controls Canonical, Local, and Global ISP components (CanoNet, LocalNet, GlobalNet), while also employing non-linear equalization and random dropout to stabilize learning. The approach achieves state-of-the-art RAW and sRGB reconstruction, enabling applications such as RAW-space deblurring and HDR synthesis with a compact model footprint. By leveraging camera parameters without altering ISP pipelines, ParamISP facilitates reliable RAW-level processing and cross-camera transfers for diverse imaging tasks.

Abstract

RAW images are rarely shared mainly due to its excessive data size compared to their sRGB counterparts obtained by camera ISPs. Learning the forward and inverse processes of camera ISPs has been recently demonstrated, enabling physically-meaningful RAW-level image processing on input sRGB images. However, existing learning-based ISP methods fail to handle the large variations in the ISP processes with respect to camera parameters such as ISO and exposure time, and have limitations when used for various applications. In this paper, we propose ParamISP, a learning-based method for forward and inverse conversion between sRGB and RAW images, that adopts a novel neural-network module to utilize camera parameters, which is dubbed as ParamNet. Given the camera parameters provided in the EXIF data, ParamNet converts them into a feature vector to control the ISP networks. Extensive experiments demonstrate that ParamISP achieve superior RAW and sRGB reconstruction results compared to previous methods and it can be effectively used for a variety of applications such as deblurring dataset synthesis, raw deblurring, HDR reconstruction, and camera-to-camera transfer.

Figures (8)

• Figure 1: Impact of camera parameters on a camera ISP. Images (a) and (b) are taken by a Samsung Galaxy S22 with different camera parameters. SS and ISO indicate the shutter speed and sensor sensitivity, respectively. In (c), we visualize RGB histograms of each cropped patch of (a) and (b). Despite the same target scene, the captured images exhibit distinct histograms, implying complex ISP operations dependent on the camera parameters.
• Figure 2: Overview of the proposed ParamISP framework. The full pipeline is constructed by combining learnable networks (ParamNet, LocalNet, GlobalNet) with invertible canonical camera operations (CanoNet). CanoNet consists of differentiable operations without learnable weights, where WB and CST denote white balance and color space transform, respectively.
• Figure 3: Architecture of ParamNet. (a) Given camera optical parameters, ParamNet estimates optical parameter features used for modulating the LocalNet and GlobalNet. (b) In order to deal with different scales and non-linearly distributed values of optical parameters, we propose to use non-linear equalization that exploits multiple non-linear mapping functions.
• Figure 4: Detailed architecture of LocalNet and GlobalNet. In GlobalNet, $f_g$ and $f_q$ represent the gamma correction and quadratic transformation, respectively, while $c_n$ represents the $n$-th coefficients $G^n$ and $W^n$, as explained in \ref{['ssec:ispnetwork']}.
• Figure 5: sRGB-to-RAW reconstruction. We show error maps between reconstructed and GT RAW images.