mlsanityData to 14 Oct 2026

Structure Information is the Key: Self-Attention RoI Feature Extractor in 3D Object Detection

arXiv 2111.00931 ↗First posted 1 Nov 2021cs.CVcited by 0

Not on the map. Why?

Abstract (arXiv)

Unlike 2D object detection where all RoI features come from grid pixels, the RoI feature extraction of 3D point cloud object detection is more diverse. In this paper, we first compare and analyze the differences in structure and performance between the two state-of-the-art models PV-RCNN and Voxel-RCNN. Then, we find that the performance gap between the two models does not come from point information, but structural information. The voxel features contain more structural information because they do quantization instead of downsampling to point cloud so that they can contain basically the complete information of the whole point cloud. The stronger structural information in voxel features makes the detector have higher performance in our experiments even if the voxel features don't have accurate location information. Then, we propose that structural information is the key to 3D object detection. Based on the above conclusion, we propose a Self-Attention RoI Feature Extractor (SARFE) to enhance structural information of the feature extracted from 3D proposals. SARFE is a plug-and-play module that can be easily used on existing 3D detectors. Our SARFE is evaluated on both KITTI dataset and Waymo Open dataset. With the newly introduced SARFE, we improve the performance of the state-of-the-art 3D detectors by a large margin in cyclist on KITTI dataset while keeping real-time capability.

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Builds on 38 of its 47 references that are in the map. First posted on arXiv 1 Nov 2021.

Builds on

38 of 47 references.

References · 38

  1. 2017Attention Is All You Needcited by 60,361
  2. 2015Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networkscited by 9,639
  3. 2016PointNet: Deep Learning on Point Sets for 3D Classification and Segmentationcited by 5,248
  4. 2017PointNet++: Deep Hierarchical Feature Learning on Point Sets in a Metric Spacecited by 4,401
  5. 2015SSD: Single Shot MultiBox Detectorcited by 3,130
  6. 2019Scalability in Perception for Autonomous Driving: Waymo Open Datasetcited by 2,451
  7. 2018Dynamic Graph CNN for Learning on Point Cloudscited by 2,223
  8. 2018PointPillars: Fast Encoders for Object Detection from Point Cloudscited by 1,715
  9. 2017VoxelNet: End-to-End Learning for Point Cloud Based 3D Object Detectioncited by 1,323
  10. 2017Cascade R-CNN: Delving into High Quality Object Detectioncited by 1,276
  11. 2018PointRCNN: 3D Object Proposal Generation and Detection from Point Cloudcited by 941
  12. 2019PV-RCNN: Point-Voxel Feature Set Abstraction for 3D Object Detectioncited by 891
26 more
  1. 20173D Semantic Segmentation with Submanifold Sparse Convolutional Networkscited by 748
  2. 2020PCT: Point cloud transformercited by 732
  3. 2016Multi-View 3D Object Detection Network for Autonomous Drivingcited by 724
  4. 2017Frustum PointNets for 3D Object Detection from RGB-D Datacited by 640
  5. 2019Bridging the Gap Between Anchor-based and Anchor-free Detection via Adaptive Training Sample Selectioncited by 540
  6. 20203DSSD: Point-based 3D Single Stage Object Detectorcited by 440
  7. 2020Voxel R-CNN: Towards High Performance Voxel-based 3D Object Detectioncited by 410
  8. 2019A Survey of Deep Learning Techniques for Autonomous Drivingcited by 403
  9. 2018ChauffeurNet: Learning to Drive by Imitating the Best and Synthesizing the Worstcited by 397
  10. 2019From Points to Parts: 3D Object Detection from Point Cloud with Part-aware and Part-aggregation Networkcited by 368
  11. 2020Point-GNN: Graph Neural Network for 3D Object Detection in a Point Cloudcited by 335
  12. 2019STD: Sparse-to-Dense 3D Object Detector for Point Cloudcited by 282
  13. 2021PV-RCNN++: Point-Voxel Feature Set Abstraction With Local Vector Representation for 3D Object Detectioncited by 232
  14. 2017Submanifold Sparse Convolutional Networkscited by 223
  15. 2021PAConv: Position Adaptive Convolution with Dynamic Kernel Assembling on Point Cloudscited by 211
  16. 2021RangeDet:In Defense of Range View for LiDAR-based 3D Object Detectioncited by 140
  17. 2017Single-Shot Refinement Neural Network for Object Detectioncited by 129
  18. 2021LiDAR R-CNN: An Efficient and Universal 3D Object Detectorcited by 129
  19. 2021SE-SSD: Self-Ensembling Single-Stage Object Detector From Point Cloudcited by 119
  20. 2020CIA-SSD: Confident IoU-Aware Single-Stage Object Detector From Point Cloudcited by 107
  21. 2019Object as Hotspots: An Anchor-Free 3D Object Detection Approach via Firing of Hotspotscited by 84
  22. 2021HVPR: Hybrid Voxel-Point Representation for Single-stage 3D Object Detectioncited by 47
  23. 2019StarNet: Targeted Computation for Object Detection in Point Cloudscited by 46
  24. 2021From Voxel to Point: IoU-guided 3D Object Detection for Point Cloud with Voxel-to-Point Decodercited by 23
  25. 2021Anchor-free 3D Single Stage Detector with Mask-Guided Attention for Point Cloudcited by 13
  26. 2020Deformable PV-RCNN: Improving 3D Object Detection with Learned Deformationscited by 10

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2021DV-Det: Efficient 3D Point Cloud Object Detection with Dynamic Voxelization
  2. 2023PVT-SSD: Single-Stage 3D Object Detector with Point-Voxel Transformer
  3. 2022Boosting Single-Frame 3D Object Detection by Simulating Multi-Frame Point Clouds
  4. 2024LION: Linear Group RNN for 3D Object Detection in Point Clouds
  5. 2025Voxel Densification for Serialized 3D Object Detection: Mitigating Sparsity via Pre-serialization Expansion
  6. 20263DTMDet: A Dual-Path Synergy Network of Transformer and SSM for 3D Object Detection in Point Clouds
  7. 2026PointLAM: Local Attentive Mamba for Efficient Point-based 3D Object Detection
  8. 2024Point-DETR3D: Leveraging Imagery Data with Spatial Point Prior for Weakly Semi-supervised 3D Object Detection