Improving 3D Object Detection Through Progressive Population Based Augmentation

Abstract

Data augmentation has been widely adopted for object detection in 3D point clouds. All previous efforts have focused on manually designing specific data augmentation methods for individual architectures, however no work has attempted to automate the design of data augmentation in 3D detection problems -as is common in 2D imagebased computer vision. In this work, we present the first attempt to automate the design of data augmentation policies for 3D object detection. We present an algorithm, termed Progressive Population Based Augmentation (PPBA). PPBA learns to optimize augmentation strategies by narrowing down the search space and adopting the best parameters discovered in previous iterations. On the KITTI test set, PPBA improves the StarNet detector by substantial margins on the moderate difficulty category of cars, pedestrians, and cyclists, outperforming all current state-of-the-art single-stage detection models. Additional experiments on the Waymo Open Dataset indicate that PPBA continues to effectively improve 3D object detection on a 20x larger dataset compared to KITTI. The magnitude of the improvements may be comparable to advances in 3D perception architectures and the gains come without an incurred cost at inference time. In subsequent experiments, we find that PPBA may be up to 10x more data efficient than baseline 3D detection models without augmentation, highlighting that 3D detection models may achieve competitive accuracy with far fewer labeled examples.

3D Point Cloud Target Detection Method Based on R-PointGNN

“…Jing Bai [26] et al proposed a lightweight real-time point cloud network, LightPointNet, with less than 0.07M parameters and good generalizability, to address the problem of more parameters in existing models. In addition, the literature [27] introduced the Progressive Population Augmentation (PPBA) algorithm to significantly optimize the StarNet detector for point clouded sparsity. The literature [28] conducted a study on the robustness of point cloud 3D target detection.…”

3D-FCT: Simultaneous 3D Object Detection and Tracking Using Feature Correlation

“…For all evaluations, we begin with a model pretrained on the KITTI [9] 3D object detection dataset before training it on the tracking dataset. Also, we do not follow the data augmentation technique of oversampling ground truth objects as described in [5,38] and common practice with most 3D object detection neural network implementations. 1.…”

“…Sequences 0,1,3,4,5,9,11,12,15,17,19,20 were used for training, while the remaining were chosen for validation.…”

Data Augmentation of Automotive LIDAR Point Clouds under Adverse Weather Situations

“…Regarding data augmentation, there have been multiple approaches going from transformations of a real training set [ 11 , 12 , 13 , 14 , 15 , 16 ], to combinations of real and synthetic data [ 17 , 18 , 19 , 20 ], to purely synthetic data [ 21 , 22 , 23 ] and domain adaptation techniques [ 24 , 25 , 26 ]. In [ 12 ], the point clouds of previously labeled objects are added by concatenation at different positions into the training data in order to improve the training of the network.…”

“…When trained simultaneously, the classifier learns features that are independent of the possible modifications caused by the augmentor. As there is a broad range of operations that can be applied for augmentation, in [ 16 ] a search method is proposed in order to find the optimal augmentation policy, understood as a series of augmentation operations with their respective parameters.…”