Open-Source Two-Stage Vision Transformer Pipeline for Fine-Grained Vehicle Classification

arXiv 2606.05149·low signal

Padmanaban and Feng release an open-source two-stage computer vision pipeline using Vision Transformers to classify vehicle body types, motivated by cyclist injury-severity analysis in overtaking crashes. Code availability and a clear applied use case make it reproducible. Narrow domain, but a clean reference for fine-grained ViT classification.