Descriptor: Underwater Stereo Vision Dataset for Transparent Object Detection & Ranging (UW-TransStereo)
Abstract
Robust detection and ranging of transparent debris in water is challenging due to refraction, low contrast, backscatter, and turbidity. We present a paired, multi-environment stereo dataset acquired with a stereo camera, comprising 25 recordings totaling over 9000 stereo frame pairs of three bottle targets (two transparent plastic bottles with filled/unfilled variants and a brown glass medicine bottle) across air and four underwater conditions (freshwater, freshwater with enhancement, saltwater, and saltwater with suspended pellets), along with analysis-ready detector spreadsheets and scripts. To promote reproducible use, we provide a standard evaluation protocol with well-defined accuracy measures—mean absolute error (MAE), root mean square error, coefficient of determination (<inline-formula><tex-math notation="LaTeX">$R^{2}$</tex-math></inline-formula>), and mean signed error—and report the proportion of frames with valid estimates (“valid %”), including vertical region-of-interest sampling (upper/centre/lower). We include a lightweight vision enhancement variant, “<italic>basicEnhance,</italic>” to illustrate how pre-processing can be benchmarked against the same protocol without retraining models. Users can load the paired dataset files to benchmark alternative detectors or compare ranging strategies across the five environmental conditions. The release is intended as a neutral testbed for benchmarking underwater transparent-object perception and for stress-testing stereo ranging under refractive media. File naming, storage structure, and missing-value conventions (including explicit use of not-a-number cases) are documented to ease integration. We release all raw stereo streams, detector outputs, qualitative screenshots, and analysis scripts. We note that the release is deliberately scoped to transparent plastic bottles in a controlled laboratory tank; extension to open-water sequences and to a broader transparent-target taxonomy is left as priority future work.</p> <p><bold>IEEE SOCIETY/COUNCIL</bold> Robotics and Automation Society (RAS)</p> <p><bold>DATA TYPE/LOCATION</bold> Stereo video and detector spreadsheets; London, U.K. (lab tank)</p> <p><bold>DATA DOI/PID</bold> <ext-link ext-link-type="uri" xlink:href="https://doi.org/10.5281/zenodo.16753748">10.5281/zenodo.16753748</ext-link>