LiDAR Sensor modeling and Data augmentation with GANs for Autonomous driving
Ahmad El Sallab, Ibrahim Sobh, Mohamed Zahran, Nader Essam
TL;DR
This work tackles the scarcity of paired LiDAR data for autonomous driving by framing sensor modeling as an unsupervised, unpaired image-to-image translation using CycleGANs. It translates LiDAR representations between simulated (CARLA) and real (KITTI) domains, in both Bird-eye View and Polar Grid Map formats, and also demonstrates real-to-real channel-density translation. The authors propose a loss framework that couples adversarial and cycle-consistency terms with task-specific and extrinsic evaluations, enabling data augmentation without ground-truth pairs. The approach shows promise for improving perception systems by generating realistic LiDAR data and preserving critical scene content, with practical implications for safer, cost-effective autonomous driving development.
Abstract
In the autonomous driving domain, data collection and annotation from real vehicles are expensive and sometimes unsafe. Simulators are often used for data augmentation, which requires realistic sensor models that are hard to formulate and model in closed forms. Instead, sensors models can be learned from real data. The main challenge is the absence of paired data set, which makes traditional supervised learning techniques not suitable. In this work, we formulate the problem as image translation from unpaired data and employ CycleGANs to solve the sensor modeling problem for LiDAR, to produce realistic LiDAR from simulated LiDAR (sim2real). Further, we generate high-resolution, realistic LiDAR from lower resolution one (real2real). The LiDAR 3D point cloud is processed in Bird-eye View and Polar 2D representations. The experimental results show a high potential of the proposed approach.