TilapiaVisionDataset: Annotated Image Dataset for Oreochromis niloticus

The Tilapia VisionDataset is a high-quality repository of 8,512 annotated images designed for the development and benchmarking of Artificial Intelligence models applied to precision aquaculture. The dataset captures Nile tilapia (Oreochromis niloticus) specimens within an experimental flume with a controlled water flow of 2700 L/h, simulating real-world conditions for automated handling and counting systems.
The images were extracted from video footage recorded in Full HD resolution (1920×1080 px) and manually annotated with bounding boxes in the YOLO format. The dataset contains a total of 63,004 instances of the "Tilapia" class, presenting typical computer vision challenges such as:

Occlusion and overlapping of multiple individuals.
Scale variations, with a predominance of small and medium objects (45% of instances are below 32x32 pixels, following the COCO Small standard).
High-speed movement of fish captured at 30 fps.
Peripheral spatial distribution, reflecting the natural swimming behavior of fish along the edges of the tank.

This dataset is highly recommended for training and evaluating deep learning architectures, specifically the YOLO family (v8 through v11) and Faster R-CNN.

Folder structure:

Dataset/
├── data.yaml
├── train/
│ ├── images/ (5.958 jpg files)
│ └── labels/ (5.958 txt files)
├── valid/
│ ├── images/ (1.276 jpg files)
│ └── labels/ (1.276 txt files)
└── test/
├── images/ (1.278 jpg files)
└── labels/ (1.278 txt files)