1,500 images of litter across 60 waste categories captured in diverse outdoor environments. The collection includes manual annotations for trash found in 'the wild' such as beaches, streets, and parks to facilitate environmental computer vision research.
Use Cases
- Train instance segmentation algorithms using the segmentation masks to delineate trash boundaries
- Build object detection systems for mobile apps using the category_id and bbox fields
- Analyze litter distribution patterns by correlating category labels with image metadata
Strengths
- 60 hierarchical categories of waste including plastics, paper, and glass
- COCO-style JSON format containing segmentation polygons and bbox coordinates
- Annotations for small-scale litter such as cigarette butts and bottle caps