DSLR and Smartphone Camera-Based Plastic Waste Detection Dataset from Cox’s Bazar Sea Beach, Bangladesh

The Cox’s Bazar Beach Plastic Waste Detection Dataset (CoxPWD 2025) is an object detection dataset of 4,195 RGB images collected from Cox’s Bazar, Bangladesh. Images were captured at Kolatoli Beach, Sugandha Beach and Laboni Beach using an iPhone 15 Pro Max, Google Pixel 6 and Canon 650D under sunny, cloudy and rainy conditions during morning, noon and afternoon, with camera heights of approximately 1 m and 6 m and viewpoints of approximately 15° and 45° from the left and right to capture realistic variation in scale, occlusion, lighting and background. All images are stored as 640×640 JPEG files and annotated for bounding box object detection in COCO format with 14 plastic waste classes which are packet, polythene, cup, spoon, straw, rope, bag, bottle, bottle_cap, net, sunglass, toy, fishing_item and others. The dataset is split into 80 percent training (3,367 images), 10 percent validation (414 images) and 10 percent test (414 images) with a separate _annotations.coco.json file for each subset, making it directly usable for YOLOv8 and other deep learning models for coastal plastic waste monitoring and benchmarking.