WasteSegCluster: dataset of annotated waste images from urban water bodies in Bangladesh

Description

WasteSegCluster is a dataset developed to support automated waste monitoring in aquatic environments of Bangladesh where poor waste management in canals, lakes and ponds poses serious environmental and public health challenges.Due to the lack of real-world datasets capturing waste in complex and unconstrained conditions, this dataset was created using images collected from various water bodies in Dhaka.It contains 1,947 images with approximately 3,900 annotated waste instances across four categories such as i. Composite Rubber Textile, ii.Organic, iii.Paper and iv.Plastic.All instances are annotated using polygon-based masks and provided in COCO format and it make the dataset suitable for object detection,instance segmentation and classification tasks.

Steps to reproduce

WasteSegCluster was developed using real-world images collected from canals, lakes, ponds, and other water bodies in Dhaka, Bangladesh. Images were captured using a mobile camera under natural and unconstrained conditions to represent variations in illumination, background, object size, occlusion, and multiple waste items. After quality screening, visible waste instances were manually categorized into four classes: Composite Rubber Textile, Organic, Paper, and Plastic. Each instance was manually annotated using polygon-based masks, and the annotations were organized in COCO format. The final dataset contains 1,947 images with approximately 3,900 annotated waste instances and can support object detection, instance segmentation, and classification tasks.

Institutions

• United International University

  Dhaka Division, Dhaka

Categories

Licence