A City-Scale Individual-Tree Dataset within Beijing Derived from Street View Imagery via Weakly Supervised Segmentation and Multi-View Geometric Fusion

Accurate, scalable, and up-to-date information on urban street trees is fundamental for urban ecological assessment, climate adaptation, and sustainable city planning, yet existing inventories are often incomplete, labor-intensive to maintain, and difficult to update at metropolitan scales. Here, we present a city-scale, individual-tree dataset of roadside vegetation derived from street view imagery using an end-to-end artificial intelligence framework that integrates weakly supervised instance segmentation, multi-view geometric fusion, and triangulation-based structural estimation. The framework simultaneously identifies individual trees, estimates structural attributes, and derives ecological indicators while substantially reducing annotation requirements through a weakly supervised segmentation model trained using only bounding-box annotations. To improve the reliability of geometric measurements from street-view imagery, a depth-independent estimation strategy combining multi-view triangulation with camera-pitch and radial-distortion correction is adopted. Validation against a newly released benchmark dataset containing 6,395 field-measured trees demonstrates high measurement accuracy, reducing crown-width estimation error from 178% to 14.5% and trunk-diameter estimation error from 270% to 30% compared with uncorrected baselines. The framework was applied across the area within Beijing (668 km²), where 351,225 individual roadside trees were mapped together with their locations, species composition, structural attributes, species diversity, and carbon storage capacity. The resulting open dataset provides a high-resolution, individual-tree representation of the urban roadside forest and reveals pronounced spatial disparities in ecological characteristics between historical urban cores and peripheral districts. The dataset is openly available through the Zenodo repository (DOI: 10.5281/zenodo.21770715) and provides a standardized and reusable data resource for urban ecological monitoring, green infrastructure management, biodiversity assessment, carbon accounting, and future studies of urban environmental change.