AerialRAW2RGB: Paired High-Resolution Aerial RAW-to-RGB Images for Neural Image Signal Processing
AerialRAW2RGB is the first paired RAW-to-RGB benchmark dataset for airborne imagery. It contains 130 image pairs captured with a Phase One iXM-RS150F medium-format camera on board a manned aircraft at flight altitudes of 1,200–3,500 m, giving a ground sampling distance of 5–15 cm/pixel. Each image measures 14,204 × 10,652 pixels (150 MP) and is provided as 16-bit IIQ RAW with an RGGB Bayer color filter array, together with an expert-curated 8-bit RGB reference image.
The scenes cover urban, agricultural, coastal, and forested areas across Germany at ISO 400–6400 under direct sunlight, overcast skies, and haze of varying density. All images are released at full native resolution, split into 100 training, 10 validation, and 20 test images. Unlike existing benchmarks, which are limited to ground-level scenes below 60 MP and less than 16-bit depth, AerialRAW2RGB captures the atmospheric scattering, wide dynamic range, and illumination gradients typical of aerial acquisition, supporting research on neural image signal processing (ISP), tile-consistent inference on very large images, and efficient on-board processing for aerial platforms and High-Altitude Platform Systems (HAPS). The examples show a 16-bit RAW image and the corresponding expert-curated 8-bit RGB reference image.
Please cite the following paper if you use AerialRAW2RGB in your work:
Jangir, S. K., Gstaiger, V., and Bahmanyar, R., "EdgeISPNet: Ultra-Lightweight Neural ISP for High-Resolution RAW-to-RGB High Altitude Imaging Systems," German Conference on Pattern Recognition (GCPR), 2026.
Downloads:
Dataset AerialRAW2RGB.zip (61,4 GB) (gigamove.rwth-aachen.de)