High-quality paired RAW-to-sRGB datasets are essential for data-driven ISP, yet existing datasets often rely on cross-device collection, causing misalignment and failing to faithfully reflect real capture conditions. We therefore introduce LC-ISP, a large-scale benchmark built from the intrinsic gap between a camera's live preview and offline capture streams. LC-ISP contains 53.9k spatially aligned groups, each comprising live RAW/live sRGB and multi-frame fused RGB/capture sRGB, and covers day, night, indoor, outdoor, and HDR scenes. It also provides capture-stream intermediates for modular supervision. We further develop a compact two-stage baseline that uses these intermediates as auxiliary targets. Under a 2.5-GFLOP budget for 448×448 RAW inputs, it outperforms the strongest single-frame baseline by over 2 dB PSNR while improving texture, denoising, and color reproduction. LC-ISP supports realistic benchmarking, the design of supervision strategies, and the development of lightweight ISP models.
Figures & tables
Figure 1: Motivation: Our method leverages the inherent differences between preview live streams and offline capture streams to construct a large-scale benchmark. (a) Live stream: A lightweight, single-frame pipeline for real-time sRGB rendering. (b) Capture stream: Multi-frame fusion with exposure bracketing to reconstruct high-quality fused RGB and sRGB outputs. The bottom row compares their visual results, highlighting the quality gap between the two streams.
Dataset
Images
Illumination
HDR Scenes
Aligned Pairs
Environment
Patch Size
Bit Depth
ZRR [ 18 ]
48,043
Day
✗
✗
Outdoor
448 × 448
10
MAI [ 17 ]
24,100
Day
✗
✗
Outdoor
224 × 224
12
LC-ISP (Ours)
53,892
Day & Night
✓
✓
In/Outdoor
448 × 448
12
Table 1: Comparison of the proposed LC-ISP with existing ISP datasets.
Figure 2: Examples from the LC-ISP Dataset. The figure shows the dataset’s scene distribution and several groups of sample images, with each group arranged from left to right: single-frame RAW, multi-frame fused RGB, live sRGB, and capture sRGB. (a) HDR/SDR scenes with different dynamic ranges. (b) Day/Night scenes under varying illumination. (c) Outdoor/Indoor scenes in diverse environments.
Figure 3: Image-quality comparison on LC-ISP. The chart compares LC-ISP ground truth with ZRR [ 18 ] , demonstrating the superior overall image quality of LC-ISP.
Figure 4: The overall architecture of our Dual-Stage ISP framework, consisting of a restoration network and an enhancement network. Leveraging multi-frame fused RGB as prior information, the restoration network emulates the restored RGB of the capture stream. With a cleaner intermediate RGB, the pixel-adaptive enhancement block enables more flexible mapping, producing sRGB results that are perceptually closer to those of the capture stream.
Category
Model
PSNR ↑
SSIM ↑
MS-SSIM ↑
LPIPS ↓
GFLOPs
Base
PyNet [ 18 ]
23.28
0.7948
0.9091
0.2531
685.16
AWNet [ 10 ]
24.43
0.7970
0.9089
0.2561
131.96
MWISPNet [ 20 ]
24.72
0.8020
0.9120
0.2495
176.80
LiteISPNet [ 47 ]
24.73
0.8028
0.9122
0.2477
143.89
SYENet [ 13 ]
20.00
0.7178
0.8561
0.3773
10.20
MobileIE [ 39 ]
19.27
0.7111
0.8539
0.3937
4.96
Table 2: Quantitative results for single-frame ISP models of different scales on the LC-ISP dataset. GFLOPs are measured using 448×448 RAW inputs. Best results are in bold .
Figure 5: From top to bottom: results for medium, and lite models. Our approach preserves more details, maintains balanced colors, and suppresses noise more effectively. Please zoom in for better view.
Figure 6: (a) Performance across data ratios on LC-ISP and ZRR. (b) Few-shot adaptation against training from scratch.
Method
PSNR ↑
SSIM ↑
MS-SSIM ↑
LPIPS ↓
MWISPNet
21.37
0.7388
0.8745
0.3624
Ours w/o Li
22.78
0.7693
0.8894
0.2980
Ours (Full)
23.36
0.7748
0.8934
0.2850
NAF+CSR w/o Li
21.91
0.7556
0.8748
0.3121
NAF+CSR
22.56
0.7648
0.8869
0.3037
Table 3: Ablation study of the proposed two-stage framework. Intermediate supervision improves performance by leveraging multi-frame information, while the last two rows show that the design also benefits other architectures.
Figure 7: Visual comparison of ablation results. The proposed intermediate supervision helps preserve fine details and improve color fidelity, enhancing overall image quality.
Figure 8: Visualization of diverse scenes from our LC-ISP dataset. The first four rows depict daytime scenes, while the last three rows illustrate nighttime scenes.
Figure 9: Visual comparison of competing methods. Please zoom in for more details.
Figure 10: Visual comparison of competing methods. Please zoom in for more details.
Computational imaging combines optical and sensor design with algorithms to recover or infer scene information from indirect, compressed, or noisy measurements.
Cameras capture scene-referred linear raw images, which are processed by onboard image signal processors (ISPs) into display-referred 8-bit sRGB outputs. Although raw data is more faithful for low-level vision tasks, collecting large-scale raw datasets remains a major bottleneck, as existing datasets are limited and tied to specific camera hardware. Generative models offer a promising way to address this scarcity -- however, existing diffusion frameworks are designed to synthesize photo-finished sRGB images rather than physically consistent linear representations. This paper presents RawGen, to our knowledge the first diffusion-based framework enabling text-to-raw generation for arbitrary target cameras, alongside sRGB-to-raw inversion. RawGen leverages the generative priors of large-scale sRGB diffusion models to synthesize physically meaningful linear outputs, such as CIE XYZ or camera-specific raw representations, via specialized processing in latent and pixel spaces. To handle unknown and diverse ISP pipelines and photo-finishing effects in diffusion-model training data, we build a many-to-one inverse-ISP dataset where multiple sRGB renditions of the same scene generated using diverse ISP parameters are anchored to a common scene-referred target. Fine-tuning a conditional denoiser and specialized decoder on this dataset allows RawGen to obtain camera-centric linear reconstructions that effectively invert the rendering pipeline. We demonstrate RawGen's superior performance over traditional inverse-ISP methods that assume a fixed ISP. Furthermore, we show that augmenting training pipelines with RawGen's scalable, text-driven synthetic data can benefit downstream low-level vision tasks.
Dongyoung Kim, Junyong Lee, Abhijith Punnappurath +4
AI Center - Toronto, Samsung Electronics · Yonsei University
Evaluating camera image signal processing (ISP) pipelines requires measuring low-level artifacts introduced by operations such as denoising, demosaicing, tone mapping, and compression. Blind image quality assessment (IQA) techniques can grade visual quality without a reference, but they typically focus on semantic and high-level visual cues or human perceptual scores rather than the low-level image-processing artifacts introduced by camera pipelines. In contrast, full-reference metrics such as PSNR and SSIM measure pixel-level differences and structural similarity, while LPIPS measures perceptual similarity in deep feature space. However, these metrics require perfectly aligned image pairs, which are difficult to collect in practical settings. We propose a reference-free learning framework that estimates full-reference image quality metrics from a processed sRGB image and its ISO metadata. Our method predicts a proxy sRGB reference, which is then compared with the processed image to compute PSNR, SSIM, and LPIPS in their standard full-reference form. Our experiments show that the proxy-reference model can be learned from synthetic data and applied to real camera data. We further show that lightweight LoRA fine-tuning enables efficient adaptation when ISP components or pipeline configurations are changed. The proposed method outperforms direct metric regression in estimating metric values and achieves higher agreement with full-reference rankings than conventional blind IQA methods. These results demonstrate the feasibility of reference-free estimation of full-reference metrics for practical camera-pipeline evaluation.
Yujin Cho, Sira Ferradans, Jean-Michel Morel +2
ENS Paris-Saclay, France · DXOMARK, France · Lingnan University, Hong Kong +2
Image reconstruction from corrupted images is crucial across many domains. Most reconstruction networks are trained on post-ISP sRGB images, even though the image-signal-processing pipeline irreversibly mixes colors, clips dynamic range, and blurs fine detail. This paper uses the rain degradation problem as a use case to show that these losses are avoidable, and demonstrates that learning directly on raw Bayer mosaics yields superior reconstructions. To substantiate the claim, we (i) evaluate post-ISP and Bayer reconstruction pipelines, (ii) curate Raw-Rain, the first public benchmark of real rainy scenes captured in both 12-bit Bayer and bit-depth-matched sRGB, and (iii) introduce Information Conservation Score (ICS), a color-invariant metric that aligns more closely with human opinion than PSNR or SSIM. On the test split, our raw-domain model improves sRGB results by up to +0.99 dB PSNR and +1.2% ICS, while running faster with half of the GFLOPs. The results advocate an ISP-last paradigm for low-level vision and open the door to end-to-end learnable camera pipelines.
Nate Rothschild, Moshe Kimhi, Avi Mendelson +1
Department of ECE, Technion, Israel · Department of CS, Technion, Israel · School of ECE, Ben-Gurion University