Pattern-Aware Data Augmentation for LiDAR 3D Object Detection

Abstract

For 3D object detection, labeling lidar point cloud is difficult, so data augmentation is an important module to make full use of precious annotated data. As a widely used data augmentation method, GT-sample effectively improves detection performance by inserting groundtruths into the lidar frame during training. However, these samples are often placed in unreasonable areas, which misleads model to learn the wrong context information between targets and backgrounds. To address this problem, in this paper, we propose a context-aware data augmentation method (CA-aug) , which ensures the reasonable placement of inserted objects by calculating the "Validspace" of the lidar point cloud. CA-aug is lightweight and compatible with other augmentation methods. Compared with the GT-sample and the similar method in Lidar-aug(SOTA), it brings higher accuracy to the existing detectors. We also present an indepth study of augmentation methods for the range-view-based(RVbased) models and find that CA-aug can fully exploit the potential of RV-based networks. The experiment on KITTI val split shows that CA-aug can improve the mAP of the test model by 8%.

Adaptive Scale and Correlative Attention PointPillars: An Efficient Real-Time 3D Point Cloud Object Detection Algorithm

“…Addressing this issue, Ref. [24] proposed Context-Aware augmentation (CA-aug), which initially segments training samples into ground points and obstacle points. These are then projected onto a range view (RV) to identify a "ValidSpace" (an area suitable for placement).…”

“…The widely used GT-Aug [9] overlooks the semantic information of the scene during random sampling and augmentation. Consequently, it might introduce ground truths into unreasonable areas, resulting in unreasonable scenes [24]. These unreasonable augmenta tions could mislead networks to learn incorrect information.…”

“…However, this method may insert ground truths into inappropriate areas. This may create incorrect contextual information for the training of the model [24]. But semantic information of the current scene could help to resolve this problem.…”

MetaLiDAR: Automated metamorphic testing of LiDAR‐based autonomous driving systems

“…MetaLiDAR performs the sparse operation to make the new point cloud similar in sparsity to the target point cloud. Specifically, transform the coordinates of the new object into spherical coordinates 38 and down‐sample the point cloud. Point clouds consist of a series of points shaped like , which are based on the plane rectangular coordinate system.…”