Plastic waste poses both an environmental crisis and a data scarcity challenge for machine learning (ML)-based sorting systems, which require diverse, annotated datasets to differentiate across materials, forms, and regional waste contexts. This paper presents Scan Your Trash , a mobile Citizen Science platform for the collection of multi-modal data to support ML-based recognition of post-consumer plastic waste. The app guides users through capturing short RGB videos, plastic polymer types, and metadata, while optionally leveraging LiDAR-equipped iPhones for depth data. We detail the app’s user flow, technical architecture, and data pipeline, as well as results from a workshop-based study involving 25 students who contributed both single images and videos of plastic items. The YOLOv11 models trained on these datasets show higher performance (mAP@50 = 98.3%) with video-derived frames compared to single images (85.9%). The broader contribution lies in demonstrating how citizen-driven, multi-modal data collection can enhance plastic waste classification pipelines.
Basedow et al. (Thu,) studied this question.