Quantifying Data Augmentation for LiDAR based 3D Object Detection
This work provides insights into optimizing data augmentation for 3D object detection, which is incremental but useful for researchers and practitioners in autonomous driving.
The paper investigates various data augmentation techniques for LiDAR-based 3D object detection, finding that both global and local augmentations can improve performance, with gains of up to 1.5% on KITTI and 1.7% on STF datasets in 3D mAP for the moderate car class, though some techniques like object translation can be harmful.
In this work, we shed light on different data augmentation techniques commonly used in Light Detection and Ranging (LiDAR) based 3D Object Detection. For the bulk of our experiments, we utilize the well known PointPillars pipeline and the well established KITTI dataset. We investigate a variety of global and local augmentation techniques, where global augmentation techniques are applied to the entire point cloud of a scene and local augmentation techniques are only applied to points belonging to individual objects in the scene. Our findings show that both types of data augmentation can lead to performance increases, but it also turns out, that some augmentation techniques, such as individual object translation, for example, can be counterproductive and can hurt the overall performance. We show that these findings transfer and generalize well to other state of the art 3D Object Detection methods and the challenging STF dataset. On the KITTI dataset we can gain up to 1.5% and on the STF dataset up to 1.7% in 3D mAP on the moderate car class.