virDepth

It is designed to support the training and evaluation of target center depth prediction for urban road users, with a particular focus on vehicles and pedestrians under long-range perception conditions. The dataset is generated through a reproducible simulation pipeline and provides synchronized RGB images, depth-related annotations, and object-level labels required for object-centric metric depth estimation.

Unlike conventional dense-depth datasets that primarily emphasize pixel-wise scene reconstruction, virDepth is constructed to facilitate controlled analysis of object-level distance perception, especially for small and distant targets in complex urban environments. virDepth supports research on target center depth estimation, object-centric perception, and long-range monocular depth analysis for autonomous driving. It is intended for academic research and algorithm evaluation.