Fish_YOLO_Dataset: A Synthetic and Real Underwater Image Dataset for Fish Species Detection and Abundance Estimation

The Fish_YOLO_Dataset.zip comprises a collection of fish images captured with the Deep Vision camera system during two research surveys in the North Atlantic (2017 and 2018) conducted by Allken and Rosen (2020), complemented with synthetic images generated from fish crops and real seabed backgrounds. The dataset is specifically designed for marine species detection and abundance measurement using YOLO architectures. Annotation errors in the original Deep Vision dataset were corrected. In the “test_annotations.csv” file, lines 36 to 47 corresponding to image “ST032‑713‑20170520023956431.jpg” were incorrectly labelled as “mackerel” instead of “mixed”. Additionally, in the “source‑train2017‑annotation.csv” file, lines 41 and 42 were identified as duplicates of lines 20 and 21 and were removed. These corrections were applied before data splitting. Since the original dataset does not define a training partition with real images, the “val_annotations.csv” file was used as the source of real images. A stratified split (50% training, 50% validation) by year (2017, 2018) and sampling station was applied, and both subsets were adjusted to 309 images each. Key Features: Real Images: 1,879 images from the 2017 campaign (863 images, resolution 1392×1040 px) and 2018 campaign (1,016 images, resolution 1228×1027 px) Synthetic Images: 5,000 images generated using an underwater scene simulator, combining fish crops (blue whiting, herring, mackerel, mesopelagic) with real seabed backgrounds Annotations: YOLO format (.txt files with normalized coordinates and class labels) Distribution: Training (5,652 images), Validation (309 images), Test (918 images) Annotated Species: Blue whiting (Micromesistius poutassou), Herring (Clupea harengus), Mackerel (Scomber scombrus), Mesopelagic (group of mid-water species) Total Size: 5.4 GB Format: ZIP with folder structure ready for YOLO Applications: Fish detection, abundance measurement, automated counting, fish stock assessment, machine learning for underwater computer vision. File Structure: Fish_YOLO_Dataset.zip│├── data.yaml # YOLO configuration file│├── train/│ ├── images/ # 5,652 images (5,000 synthetic + 652 real)│ └── labels/ # Corresponding YOLO annotations│├── val/│ ├── images/ # 309 validation images│ └── labels/ # Corresponding YOLO annotations│└── test/ ├── images/ # 918 test images └── labels/ # Corresponding YOLO annotations