A Review of Computer Vision for Industrial-Grade Waste Classification

The automation of municipal solid waste sorting is a rapidly advancing field, driven by the need to improve efficiency, accuracy, and recycling outcomes in material recovery facilities. A key component of this automation is accurate waste classification, where computer vision holds great potential, but most prior work has been limited to controlled and simplified laboratory setups or in-the-wild scenarios. The recently published industrial-grade waste datasets, recorded in real-world sorting facilities showing waste streams on conveyor belts with high level of clutter and occlusion, present a severe challenge for state-of-the-art computer vision models. Despite their significance, these datasets have not yet been comprehensively reviewed. This paper addresses that gap by systematically analyzing and comparing existing industrial-grade waste datasets. We further summarize well-reported model performances and investigate model architectures and methods evaluated on the presented datasets to support future model development. We identify both dataset-specific and general challenges in this domain and conclude by outlining future research directions to advance CV-based waste sorting in industrial environments.