Structure Aware Single-Stage 3D Object Detection From Point Cloud

By Chenhang He, Huiyu Zeng, Jianqiang Huang, Xiansheng Hua, Lei Zhang. Computer Vision and Pattern Recognition, 2020

613 citations recorded by the source corpus.

3D object detection from point cloud data plays an essential role in autonomous driving. Current single-stage detectors are efficient by progressively downscaling the 3D point clouds in a fully convolutional manner. However, the downscaled features inevitably lose spatial information and cannot make full use of the structure information of 3D point cloud, degrading their localization precision. In this work, we propose to improve the localization precision of single-stage detectors by explicitly leveraging the structure information of 3D point cloud. Specifically, we design an auxiliary network which converts the convolutional features in the backbone network back to point-level representations. The auxiliary network is jointly optimized, by two point-level supervisions, to guide the convolutional features in the backbone network to be aware of the object structure. The auxiliary network can be detached after training and therefore introduces no extra computation in the inference stage. Besides, considering that single-stage detectors suffer from the discordance between the predicted bounding boxes and corresponding classification confidences, we develop an efficient part-sensitive warping operation to align the confidences to the predicted bounding boxes. Our proposed detector ranks at the top of KITTI 3D/BEV detection leaderboards and runs at 25 FPS for inference.

Subfields: Aerospace Engineering, Computational Mechanics, Computer Vision and Pattern Recognition

View Paper: https://www.semanticscholar.org/paper/f6ca0f76ef6b69a3c1ef63251c78116af34626b4

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Aerospace Engineering

In Computational Mechanics

In Computer Vision and Pattern Recognition

Citations in the corpus

ST3D: Self-training for Unsupervised Domain Adaptation on 3D Object Detection (2021, background): “Another line of research [46, 53, 37, 15, 34] adopts 3D convolutional networks to learn 3D features from voxelized point clouds, and the extracted 3D feature volumes are also further compressed to bird-view feature maps as the above.”
3D Object Detection for Autonomous Driving: A Comprehensive Survey (2022, background): “PVCNN [165] Point-Voxel Convolution SPVNAS [273] Sparse Point-Voxel Convolution SA-SSD [94] Auxiliary Point Network PVGNet [195] Point-Voxel-Grid Fusion”
Focal Sparse Convolutional Networks for 3D Object Detection (2022, background): “, the R-CNN family [11, 28, 36, 37] and the SSD family [17, 51, 60, 61].”
PV-RCNN++: Point-Voxel Feature Set Abstraction With Local Vector Representation for 3D Object Detection (2021, background): “Voxel CNN with 3D sparse convolution [61], [81] is a popular choice of stateof-the-art 3D detectors [16], [24], [83] for its efficiency of converting irregular point clouds into 3D sparse feature volumes.”
Center-based 3D Object Detection and Tracking (2020, methodology): “Modern 3D object detectors [19, 27, 54, 64] uses a 3D encoder that quantizes the point-cloud into regular bins.”
CIA-SSD: Confident IoU-Aware Single-Stage Object Detector From Point Cloud (2020, methodology): “Compared with the commonlyused 2D feature extraction module (Yan, Mao, and Li 2018; He et al. 2020), our SSFA module boosts the performance effectively with a moderate increase in the model complexity.”
Improving 3D Object Detection with Channel-wise Transformer (2021, methodology): “Later studies [43, 24, 4, 10] further apply the point-voxel mixed strategy to capture multi-scale features while retaining fine-grained localization but are strongly tied to the specific RPN architectures.”
Voxel R-CNN: Towards High Performance Voxel-based 3D Object Detection (2020, methodology): “We follow the similar design of (Yan, Mao, and Li 2018; He et al. 2020; Shi et al. 2020a) to build our backbone networks.”
Not All Points Are Equal: Learning Highly Efficient Point-based Detectors for 3D LiDAR Point Clouds (2022, neutral): “53 - - - - - - 20 SA-SSD [11] CVPR 2020 1-stage 88.”
SE-SSD: Self-Ensembling Single-Stage Object Detector From Point Cloud (2021, methodology): “Recent trend (see Figure 1 and Table 1) reveals that the precisions of single-stage detectors [8, 31] gradually approach those of the two-stage detectors [23, 25, 32].”
PointAugmenting: Cross-Modal Augmentation for 3D Object Detection (2021, methodology): “The grid-based methods divide point clouds into regular 3D voxels [22, 34, 27, 6] or BEV maps [9, 24, 23].”
Point2Seq: Detecting 3D Objects as Sequences (2022, methodology): “SASSD [11] applies the part-sensitive warping scheme to enhance the spatial features.”
ARKitScenes: A Diverse Real-World Dataset For 3D Indoor Scene Understanding Using Mobile RGB-D Data (2021, background): “3D object detection is a computer vision task that has gained a lot of popularity in recent years [27, 28, 29, 30, 31, 32, 33, 4, 5, 34, 7].”