Waterloo Urban Scene Dataset: An Annotation-Efficient Dataset for Urban Scene Classification with Minimal Supervision

High-definition (HD) urban scene mapping is crucial for urban applications and autonomous driving.However, achieving high performance in HD mapping requires large amounts of high-quality annotated data.While the SkyScape dataset is valuable, it is limited by its focus on lane markings in Germany.In this paper, we present the Waterloo Urban Scene Dataset, built upon the Waterloo Building Dataset, designed for minimal supervision deep learning.The dataset includes 907 well-annotated, 775 roughly annotated, and 23,172 intact patches (512 512 pixels, 0.12m/pixel resolution in RGB).Combined with the SkyScape dataset, it supports HD urban scene mapping with minimal supervision.Future work will focus on benchmarking and method development.