Raw seabed images (Bayer mosaic), survey 20260708_104341 (aborted early because of the conditions), SmartBay test site, Galway Bay (53.2283° N, 9.2670° W, 24 m), Graal Tech AUV with DIGI4ECO stereo camera, 2026-07-08, DIGI4ECO D4.2 DS2-S09 v1.0

Raw seabed images from survey 20260708_104341 (aborted early because of the conditions) of the DIGI4ECO SmartBay campaign: 4 072 images from the left camera of the DIGI4ECO stereo camera, carried by a Graal Tech AUV over the SmartBay test site in Galway Bay, Ireland (53.2283° N, 9.2670° W; about 24 m depth), on 2026-07-08 between 08:43:41 and 09:06:22 UTC. The record also includes the camera calibration and the AUV navigation of the survey, so the images can be corrected for refraction and placed in time and space. This is record DS2-S09 of DIGI4ECO Deliverable D4.2 "Seabed mapping: optic, laser, and acoustic 2D maps and turbidity maps". The source dataset DS2 is published as one record per survey (DS2-S09 is survey 9 of 11), with a campaign index record (DS2) holding the calibration, the full navigation and the list of all records. The enhanced images of this survey are in DS2P1-S09. This survey is not part of the 2D photomosaic DS2P2. Survey Survey Date Images (UTC) Images Planned / median altitude Median depth Pattern 20260708_104341 2026-07-08 08:43:41 – 09:06:22 4 072 2.0 / 2.01 m 23.7 m Aborted early because of the conditions Campaign and acquisition The SmartBay campaign of DIGI4ECO took place on 7 and 8 July 2026 at the SmartBay test site in Galway Bay, Ireland, centred near 53.2283° N, 9.2670° W, at about 24 m depth. A Graal Tech AUV carried the DIGI4ECO stereo camera (model PDH050S-C, serial number 261300517): two 2448 × 2048 px colour sensors (Bayer BG, 8 bit), a baseline of about 100 mm and one flat glass port, 5.75 mm thick, shared by both cameras. The camera looks down, tilted 16° forward from the vertical. Only the left camera's images are published, since they are all that is needed to build the photomosaic. Most surveys were lawnmower patterns at 2 m altitude, 3 frames per second and 0.3–0.5 m/s. On day 2, more particles in the water and less natural light led to a lower altitude (1.5 m) and higher exposure and gain. Artificial lights were not used, because they made the suspended particles visible. In total, the campaign has 11 surveys and 40 707 images. Files File Content Size DIGI4ECO_SMARTBAY_20260708_DS2-S09_raw-images_20260708_104341_v1.0.zip images/raw/20260708_104341/: 4 072 images, manifest.csv, SHA256SUMS 5.46 GB DIGI4ECO_SMARTBAY_20260708_DS2-S09_calibration-navigation_20260708_104341_v1.0.zip Calibration, navigation of the survey and image documentation (see below) ≈ 0.1 GB Extracting both archives into the same folder gives the documented tree (images/, calibration/, navigation/). Images Each raw image stores the left sensor's data as recorded: the 8-bit Bayer mosaic (BG pattern) as a single-channel, 2448 × 2048 px PNG (lossless), with no debayering, no rectification and no colour correction. One OpenCV call turns it into a colour image, and the seawater maps of the calibration then give an ideal pinhole image, which is exactly the geometric step used for the processed images: import cv2 bayer = cv2.imread('images/raw/20260708_104341/left_<stamp>.png', cv2.IMREAD_UNCHANGED) # 2048 x 2448, uint8 bgr = cv2.cvtColor(bayer, cv2.COLOR_BayerBG2BGR) fs = cv2.FileStorage('calibration/SN261300517_20260521/pinax_seawater/left_pinax_maps.yml.gz', cv2.FILE_STORAGE_READ) rect = cv2.remap(bgr, fs.getNode('mapx').mat(), fs.getNode('mapy').mat(), cv2.INTER_LINEAR) File names are left_<YYYYMMDD>_<HHMMSS>_<mmm>_<frame>.png: capture date, time and milliseconds, and the frame number within the survey. The time is the camera computer's local time, UTC+2: subtract 2 hours to match the navigation, which is in UTC. Each survey folder also has a manifest.csv (file, source file, frame index, date, time, size, content type) and a SHA256SUMS file (sha256sum -c SHA256SUMS). Calibration and navigation DIGI4ECO_SMARTBAY_20260708_DS2-S09_calibration-navigation_20260708_104341_v1.0.zip makes this record usable on its own: images/README.md description of the raw and processed images calibration/SN261300517_20260521/ camera calibration (README, SHA256SUMS) in_air/ in-air calibration of each camera and of the pair (OpenCV YAML) pinax_seawater/ refraction-correction maps (.yml.gz) and the virtual pinhole cameras navigation/README.md navigation format, time reference and camera mounting navigation/auv_to_left_camera.yaml pose of the left camera in the AUV body frame navigation/per_survey/20260708_104341_navigation.csv Calibration: in air on 21 May 2026 with a ChArUco board (89 stereo pairs), pinhole model with radial terms k1, k2. RMS reprojection error 0.39 px (left), 0.36 px (right), 0.52 px (stereo). Refraction at the flat port (glass index 1.492, seawater index 1.34) is handled with the Pinax model (Łuczyński et al., 2017): per-pixel maps turn a debayered raw image into an image from an ideal pinhole camera with one cv2.remap call. The in-air calibration alone is not valid underwater. Navigation: the AUV's navigation-filter estimate at about 5 Hz, from 10 s before the first image of the survey to 10 s after its last one. Columns: timestamp (Unix epoch, UTC), x latitude and y longitude (WGS84), z = −depth (m), roll, pitch, yaw (degrees) and the DVL altitude above the seabed (m). Every image has a navigation sample less than 1 s away. Usage notes Only the left camera's images are published, so stereo depth cannot be computed. Underwater, the position is the estimate of the AUV navigation filter, not a GPS fix, and its absolute accuracy has not been characterised. On day 1 it was 4–10 m off, as found when building the photomosaic. The camera and AUV clocks were not synchronised by hardware. On day 2, more suspended particles and less natural light required a lower altitude (1.5 m) and higher exposure and gain, so the images are noisier. This survey was aborted early because of the conditions. It never registered into one connected scene and is not part of the photomosaic. Related datasets DS2: SmartBay campaign index: calibration, full navigation and list of all records, DOI: 10.5281/zenodo.23572251 DS2P1-S09: Enhanced and rectified images of this survey, DOI: 10.5281/zenodo.23572299 References Łuczyński, T., Pfingsthorn, M., Birk, A. (2017). The Pinax-model for accurate and efficient refraction correction of underwater cameras in flat-pane housings. Ocean Engineering, 133, 9–22. https://doi.org/10.1016/j.oceaneng.2017.01.029 Responsible partners CORONIS, Graal Tech, Marine Institute. Licence Creative Commons Attribution 4.0 International (CC BY 4.0). Acknowledgement This dataset was produced within the DIGI4ECO project (Digital Twin-sustained 4D ecological monitoring of restoration in fishery depleted areas). Funded by the European Union under grant agreement No 101112883 (DIGI4ECO). Views and opinions expressed are however those of the author(s) only and do not necessarily reflect those of the European Union or the European Climate, Infrastructure and Environment Executive Agency (CINEA). Neither the European Union nor the granting authority can be held responsible for them.