STD: Sparse-to-Dense 3D Object Detector for Point Cloud

By Zetong Yang, Yanan Sun, Shu Liu, Xiaoyong Shen, Jiaya Jia. 2019

854 citations recorded by the source corpus.

We propose a two-stage 3D object detection framework, named sparse-to-dense 3D Object Detector (STD). The first stage is a bottom-up proposal generation network that uses raw point clouds as input to generate accurate proposals by seeding each point with a new spherical anchor. It achieves a higher recall with less computation compared with prior works. Then, PointsPool is applied for proposal feature generation by transforming interior point features from sparse expression to compact representation, which saves even more computation. In box prediction, which is the second stage, we implement a parallel intersection-over-union (IoU) branch to increase awareness of localization accuracy, resulting in further improved performance. We conduct experiments on KITTI dataset, and evaluate our method on 3D object and Bird’s Eye View (BEV) detection. Our method outperforms other methods by a large margin, especially on the hard set, with 10+ FPS inference speed.

Subfields: Aerospace Engineering, Computer Vision and Pattern Recognition

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 Computer Vision and Pattern Recognition

Citations in the corpus

Static-Dynamic Co-Teaching for Class-Incremental 3D Object Detection (2021, background): “…2018; Yang, Luo, and Urtasun 2018; Zeng et al. 2018; Zhou and Tuzel 2018; Chen et al. 2019; Lang et al. 2019; Qi et al. 2019; Shi, Wang, and Li 2019; Yang et al. 2019; Zhou et al. 2019; Yang et al. 2020; Zheng et al. 2021) are proposed and have shown impressive performance in localizing and…”
Structure Aware Single-Stage 3D Object Detection From Point Cloud (2020, methodology): “In addition, our method is 2.5 times faster than the second top method STD [28].”
CIA-SSD: Confident IoU-Aware Single-Stage Object Detector From Point Cloud (2020, methodology): “STD (Yang et al. 2019) exploits spherical anchors to generate proposals, together with a segmentation branch, to reduce the number of positive anchors.”
PointAugmenting: Cross-Modal Augmentation for 3D Object Detection (2021, methodology): “For point-based approaches, PointRCNN [15] and STD [26] apply PointNet [14] to segment foreground points and generate proposals for each point.”
V2X-ViT: Vehicle-to-Everything Cooperative Perception with Vision Transformer (2022, background): “To satisfy both computational and flexible receptive field requirements, [31, 50, 46] combine voxelbased and point-based approaches to detect 3D objects.”
EPNet: Enhancing Point Features with Image Semantics for 3D Object Detection (2020, methodology): “ry. However, methods based on the camera image have diculty in generating accurate 3D bounding boxes due to the lack of depth information. 3D object detection based on LiDAR.Many LiDAR-based methods [39,24,40] are proposed in recent years. VoxelNet [43] divides a point cloud into voxels and employs ”
Class-balanced Grouping and Sampling for Point Cloud 3D Object Detection (2019, methodology): “Methods taking advantage of both voxel-based and point-based methods like [25, 30, 24] ﬁrst use pointnet fashions to acquire high-quality proposals, then voxel-based methods is applied to obtain ﬁnal predictions.”
Unifying Voxel-based Representation with Transformer for 3D Object Detection (2022, background): “Point-based detectors usually aggregate features from raw point clouds with set abstraction [13] and then predict box proposals [14, 15, 16].”
Towards Robust LiDAR-based Perception in Autonomous Driving: General Black-box Adversarial Sensor Attack and Countermeasures (2020, background): “Instead of transforming point clouds to regular 2D structures or voxels for feature extraction, recent studies propose to directly operate on point clouds for 3D object detection [25, 53, 54, 71] and achieve the state-of-the-art performance.”
Improving 3D Object Detection with Channel-wise Transformer (2021, background): “STD [37] further extended the proposal refinement by transferring sparse point features into dense voxel representation.”
3DSSD: Point-Based 3D Single Stage Object Detector (2020, methodology): “Another stream is with point-based methods [31, 32, 23].”
CLOCs: Camera-LiDAR Object Candidates Fusion for 3D Object Detection (2020, methodology): “PointPillars [9] uses PointNets [7] in an encoder that represents point clouds organized in vertical columns (pillars) followed with a 2D CNN detection head to perform 3D object detection; it enables inference at 62 Hz; Compared with one-stage methods discussed above, PointRCNN [8], Fast PointRCNN [”
Geometry Uncertainty Projection Network for Monocular 3D Object Detection (2021, neutral): “Compared with the LiDAR/stereo-based methods [32, 35, 37, 40, 41, 49, 57], monocular 3D object detection is still a challenging task due to the lack of depth cues, which makes monocular object-level depth estimation naturally ill-posed.”
Voxel R-CNN: Towards High Performance Voxel-based 3D Object Detection (2020, methodology): “STD (Yang et al. 2019) exploits the PointsPool operation to extract points features of each proposal and transfer the sparse points into dense voxel representation.”
Multimodal Virtual Point 3D Detection (2021, methodology): “Two-stage 3D detectors [8, 45– 47 , 63] use a PointNet-based PointNet-based Camera-based 3D Object Detection Camera-based 3D object detection predicts 3D bounding boxes from camera images.”
SE-SSD: Self-Ensembling Single-Stage Object Detector From Point Cloud (2021, background): “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].”
PV-RCNN++: Point-Voxel Feature Set Abstraction With Local Vector Representation for 3D Object Detection (2021, methodology): “STD (Yang et al., 2019) transforms point-wise features to dense voxels for reﬁning the proposals.”