Small Object Detection in Industrial Recycling: A New Dataset and YOLO Performance Evaluation

Researchers released a new industrial recycling dataset with 10,000+ images and 120,000+ instances.

The paper benchmarks deep-learning detectors, especially YOLO variants, on small, dense, and overlapping objects.

It also evaluates object detection, length measurement, and anomaly detection tasks.

Data augmentation and synthetic images are analyzed for their effect on performance.

The dataset and comparisons aim to improve recycling automation and other small-object vision systems.