3DSSD: Point-Based 3D Single Stage Object Detector

By Zetong Yang, Yanan Sun, Shu Liu, Jiaya Jia. 2020

1,185 citations recorded by the source corpus.

Prevalence of voxel-based 3D single-stage detectors contrast with underexplored point-based methods. In this paper, we present a lightweight point-based 3D single stage object detector 3DSSD to achieve decent balance of accuracy and efficiency. In this paradigm, all upsampling layers and the refinement stage, which are indispensable in all existing point-based methods, are abandoned. We instead propose a fusion sampling strategy in downsampling process to make detection on less representative points feasible. A delicate box prediction network, including a candidate generation layer and an anchor-free regression head with a 3D center-ness assignment strategy, is developed to meet the demand of high accuracy and speed. Our 3DSSD paradigm is an elegant single-stage anchor-free one. We evaluate it on widely used KITTI dataset and more challenging nuScenes dataset. Our method outperforms all state-of-the-art voxel-based single-stage methods by a large margin, and even yields comparable performance with two-stage point-based methods, with amazing inference speed of 25+ FPS, 2x faster than former state-of-the-art point-based methods.

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): “…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 categorizing objects of interest in the…”
Center-based 3D Object Detection and Tracking (2020, background): “3D object detection aims to predict three dimensional rotated bounding boxes [11,15,27,30,37,54,58,59].”
Unifying Voxel-based Representation with Transformer for 3D Object Detection (2022, neutral): “Method Backbone NDS(%) mAP(%) mATE↓ mASE↓ mAOE↓ mAVE↓ mAAE↓ LiDAR-based 3DSSD [45] Point-based 56.”
BEVFusion: Multi-Task Multi-Sensor Fusion with Unified Bird's-Eye View Representation (2022, methodology): “Researchers have designed single-stage 3D object detectors [9], [10], [11], [12], [13], [14] that extract flattened point cloud features using PointNets [15] or SparseConvNet [16] and perform detection in the BEV space.”
Focal Sparse Convolutional Networks for 3D Object Detection (2022, background): “The first is to process point clouds [37,51] directly, based on PointNet++ [33] networks.”
DAIR-V2X: A Large-Scale Dataset for Vehicle-Infrastructure Cooperative 3D Object Detection (2022, background): “PointPillars [15], SECOND [27], and 3DSSD [28] are such approaches that achieve convincing detection results from point clouds.”
BEVFusion: A Simple and Robust LiDAR-Camera Fusion Framework (2022, background): “…lie in two categories based on their feature modality: i) point-based methods that directly operate on the raw LiDAR point clouds [38, 37, 36, 42, 57, 22]; and ii) transforming the original point clouds into a Euclidean feature space, such as 3D voxels [65] and feature pillar [20, 52, 59].”
SE-SSD: Self-Ensembling Single-Stage Object Detector From Point Cloud (2021, background): “3DSSD [31] combines feature- and point-based sampling to improve the classification.”
Improving 3D Object Detection with Channel-wise Transformer (2021, background): “Moreover, 3DSSD [36] improved the point-based approach with a new sampling strategy based on feature distance.”
TransFusion: Robust LiDAR-Camera Fusion for 3D Object Detection with Transformers (2022, background): “Other works [36,37,55,56] directly operate on raw point clouds without quantization.”
Voxel R-CNN: Towards High Performance Voxel-based 3D Object Detection (2020, methodology): “(Lang et al. 2019; Shi et al. 2020a; Yang et al. 2020; Ye, Xu, and Cao 2020).”
TransFuser: Imitation With Transformer-Based Sensor Fusion for Autonomous Driving (2022, methodology): “While the orientation is a single scalar value, directly regressing this value is challenging, as observed in existing 3D detectors [105], [106], [107].”
BEVDet: High-performance Multi-camera 3D Object Detection in Bird-Eye-View (2021, methodology): “Following the up-to-date LiDAR-based methods [34, 55, 56, 63], common data augmentation operations in 2D space are adopted including flipping, scaling, and rotating.”
Voxel Transformer for 3D Object Detection (2021, background): “Point-based approaches [26, 19, 34, 35] directly operate and generate 3D bounding boxes on point clouds.”
PV-RCNN++: Point-Voxel Feature Set Abstraction With Local Vector Representation for 3D Object Detection (2021, neutral): “3DSSD (Yang et al., 2020) Fig.”
CIA-SSD: Confident IoU-Aware Single-Stage Object Detector From Point Cloud (2020, background): “Also, the detection precisions of recent single-stage detectors (He et al. 2020; Yang et al. 2020; Shi and Rajkumar 2020) gradually approach that of the state-of-the-art two-stage detectors.”
The YOLO Framework: A Comprehensive Review of Evolution, Applications, and Benchmarks in Object Detection (2024, neutral): “The architecture of RFB Net is shown in Figure 5.”