Reference UnderWater Dataset (RUWD)

The RUWD (Reference UnderWater Dataset) for RGB images and Depth provides paired in-air and underwater RGB-D images captured under a structured experimental setup to analyze underwater image degradation. It is designed to support research in underwater image enhancement, object detection, stereo depth estimation, and machine perception. The dataset consists of images collected using a stereo RGB-D camera (ZED 2i) in both in-air and underwater environments, across varying lighting conditions (L1 ~500 lm, L2 ~1000 lm, L3 ~1500 lm, and no_light), object distances (0.55 m, 0.85 m, 1.15 m), and scene complexities including single-object, double-object, triple-object, and no-object scenarios. Each image pair is accompanied by corresponding depth maps and calibration files. Images are organized into folders based on: Domain: in_air / underwater Lighting: L1, L2, L3, no_light Distance: 0.55m, 0.85m, 1.15m (not present in no_object category) Object Class: single_object / double_object / triple_object / no_object Camera Side: left / right Each subfolder includes: RGB images: left_0.png, right_0.png Depth maps: aligned disparity or depth images The dataset enables channel-wise color analysis, inter-channel color ratio analysis, contrast and sharpness analysis, and performance benchmarking under realistic underwater image degradation scenarios. This dataset is approximately 60 GB in size and is best suited for researchers developing and validating underwater computer vision algorithms with a focus on realism, reproducibility, and task fidelity.