This paper introduces VEnvision3D, a large synthetic dataset for multi-task learning in 3D computer vision, spanning depth completion, segmentation, upsampling, place recognition and 3D reconstruction. Since data for each task was collected in shared scenarios, tasks have inherent alignment without needing separate training regimes. This enables exploring relationships ...