Supplemental Material for ROADWork: A Dataset and Benchmark for Learning to Recognize, Observe, Analyze and Drive Through Work Zones

Author/Presenter: Ghosh, Anurag; Zheng, Shen; Tamburo, Robert; Vuong, Khiem; Alvarez-Padilla, Juan; Zhu, Hailiang; Cardei, Michael; Dunn, Nicholas; Mertz, Christoph; Narasimhan, Srinivasa G.
Abstract:

The supplemental material for ROADWork: A Dataset and Benchmark for Learning to Recognize, Observe, Analyze and Drive Through Work Zones provides extended details on dataset construction, annotation protocols, and additional experimental results supporting the main ICCV 2025 paper. It includes expanded statistics on work zone clusters identified via GPS, class distributions across annotated categories, and detailed quantitative performance comparisons for object detection and vision-language models on work zone tasks. Additional analyses further illustrate the benefits of supervised training on the ROADWork dataset compared to open-vocabulary methods, including improvements in instance segmentation and sign reading. The supplement also describes technical procedures for georeferencing, 3D reconstruction of driving trajectories, and conversion of video sequences to standardized waypoints for navigational path prediction. These expanded results and methodological specifics give comprehensive support to the claims and benchmarks reported in the main manuscript.