BengalDeltaFish: A local dataset for fish detection in Bangladeshi markets

The BengalDeltaFish [1] dataset resolves the common challenge of recognizing fish species in real-world fish market environments by providing a diverse, large-scale image collection of different fish species captured under uncontrolled and realistic conditions. Unlike traditional fish datasets that are captured in controlled conditions and backgrounds, this dataset's images are collected directly from local fish markets, maintaining natural lighting variations and uncontrolled backgrounds, where fish are often placed on ice, in baskets, trays, or alongside other fish. The dataset contains 33 different fish species commonly found in local markets, including rare species that are not widely available in existing datasets. It includes 4560 annotated images that have been preprocessed and randomly split into training, validation, and testing sets for optimal use in deep learning model training, validation, and evaluation. The 98.23% mAP@50 achieved in YOLOv11s [2] training verifies the dataset's potential for creating applications or tools that work reliably in the field. This dataset has significant reuse potential across multiple domains, including: preventing fish mislabeling to ensure consumers receive the correct species, ensuring fair pricing, and monitoring illegal sales. By reducing the gap between controlled datasets and real-world applications, BengalDeltaFish [1] serves as a valuable resource for AI-driven fish detection and classification. The dataset loading sample code for this dataset is available on GitHub with proper documentation. https://github.com/281096alif/BengalDeltaFish.