Oreochromis niloticus Fingerlings WhiteTray Dataset (40 Fish per Image Polygon annotation)

Description

The Oreochromis niloticus Fingerlings WhiteTray Dataset is a curated repository of annotated images designed for the development and benchmarking of Artificial Intelligence models in precision aquaculture, particularly for fish detection and counting tasks. The dataset contains images of Nile tilapia (Oreochromis niloticus) fingerlings captured in a controlled environment using white trays (white tray setup), which enhances visual contrast and facilitates object detection and segmentation processes. Each image contains exactly 40 fish, ensuring a fixed and controlled density across the entire dataset, annotation method Polygon (Segments). This characteristic makes the dataset especially suitable for evaluating counting algorithms, density estimation methods, and object detection models under standardized conditions. The images were obtained under consistent lighting and background conditions, minimizing environmental variability while still preserving common computer vision challenges such as: Variations in orientation and positioning Subtle scale differences among fingerlings Visual similarity between instances (low inter-class variance) All images are annotated with bounding boxes in YOLO format, supporting direct use in state-of-the-art object detection frameworks. This dataset is highly recommended for training and evaluating deep learning architectures, including the YOLO family (v8 to v11) and Faster R-CNN, particularly in scenarios requiring precise counting and detection under controlled densities. Folder structure: ON_Fingerlings_WhiteTray_40_polygon_fish.zip/ ├── data.yaml ├── train/ │ ├── images/ (160 jpg files) │ └── labels/ (160 txt files) ├── valid/ │ ├── images/ (20 jpg files) │ └── labels/ (20 txt files) └── test/ ├── images/ (20 jpg files) └── labels/ (20 txt files)

Steps to reproduce

Download and unzip the dataset. Install Python 3.10 and Ultralytics YOLOv11. Run: pip install ultralytics Use the provided data.yaml file. Train the model: yolo detect train data=data.yaml model=yolov11n.pt epochs=200 imgsz=640 Evaluate performance: yolo detect val model=runs/detect/train/weights/best.pt

Institutions

• Instituto Federal Goiano

  Goiás, Goiás

Categories

Funders

• Coordenação de Aperfeicoamento de Pessoal de Nível Superior

  Ministry of Education

  Brasília

Licence