PakhiderChobi: A Comprehensive Dataset for Real-time Detection of Bangladeshi Birds

Urbanization and rapid development have led to a decline in natural habitats for birds, posing a threat to avian biodiversity worldwide. This study addresses the crucial task of accurately detecting and classifying bird species in urban environments, a vital component of bird conservation efforts, with a specific focus on Bangladeshi bird species. To overcome this challenge, we introduce the PakhiderChobi dataset, a collection of 8,670 annotated images featuring 33 commonly found Bangladeshi bird species. This dataset is designed to support real-time bird recognition and classification, encompassing a variety of scenarios, including instances where multiple birds of the same class occur, as well as cases where multiple bird species coexist within a single frame. To evaluate the quality of our dataset, we employ the YOLOv8 object detection model, achieving a mean average precision (mAP) of 95.3%, with efficient inference times of 6.6 milliseconds per image.