Recovering and Simulating Pedestrians in the Wild
Authors
Ze Yang (Uber ATG, University of Toronto)*; Sivabalan Manivasagam (Uber ATG, University of Toronto); Ming Liang (Uber); Bin Yang (Uber ATG & University of Toronto); Wei-Chiu Ma (MIT); Raquel Urtasun (Uber ATG)
Interactive Session
2020-11-16, 12:30 - 13:00 PST | PheedLoop Session
Abstract
Sensor simulation is a key component for testing the performance of self-driving vehicles and for data augmentation to better train perception systems. Typical approaches rely on artists to create both 3D assets and their animations to generate a new scenario. This, however, does not scale. In contrast, we propose to recover the shape and motion of pedestrians from sensor readings captured in the wild by a self-driving car driving around. Towards this goal, we formulate the problem as energy minimization in a deep structured model that exploits human shape priors, reprojection consistency with 2D poses extracted from images, and a ray-caster that encourages the reconstructed mesh to agree with the LiDAR readings. Importantly, we do not require any ground-truth 3D scans or 3D pose annotations. We then incorporate the reconstructed pedestrian assets bank in a realistic LiDAR simulation system by performing motion retargeting, and show that the simulated LiDAR data can be used to significantly reduce the amount of annotated real-world data required for visual perception tasks.