Aligning Bird-Eye View Representation of Point Cloud Sequences using Scene Flow
Not on the map. Why?
Abstract (arXiv)
Low-resolution point clouds are challenging for object detection methods due to their sparsity. Densifying the present point cloud by concatenating it with its predecessors is a popular solution to this challenge. Such concatenation is possible thanks to the removal of ego vehicle motion using its odometry. This method is called Ego Motion Compensation (EMC). Thanks to the added points, EMC significantly improves the performance of single-frame detectors. However, it suffers from the shadow effect that manifests in dynamic objects' points scattering along their trajectories. This effect results in a misalignment between feature maps and objects' locations, thus limiting performance improvement to stationary and slow-moving objects only. Scene flow allows aligning point clouds in 3D space, thus naturally resolving the misalignment in feature spaces. By observing that scene flow computation shares several components with 3D object detection pipelines, we develop a plug-in module that enables single-frame detectors to compute scene flow to rectify their Bird-Eye View representation. Experiments on the NuScenes dataset show that our module leads to a significant increase (up to 16%) in the Average Precision of large vehicles, which interestingly demonstrates the most severe shadow effect. The code is published at https://github.com/quan-dao/pc-corrector.
Key facts. Cited by 1 paper in the map (data through 14 Oct 2026). Builds on 27 of its 32 references that are in the map. First posted on arXiv 4 May 2023.
Builds on
27 of 32 references.
References · 27
- 2017Attention Is All You Needcited by 60,361
- 2019nuScenes: A multimodal dataset for autonomous drivingcited by 4,523
- 2017Decoupled Weight Decay Regularizationcited by 1,831
- 2018PointPillars: Fast Encoders for Object Detection from Point Cloudscited by 1,715
- 2019Objects as Pointscited by 1,175
- 2020Center-based 3D Object Detection and Trackingcited by 1,144
- 2022TransFusion: Robust LiDAR-Camera Fusion for 3D Object Detection with Transformerscited by 466
- 20203DSSD: Point-based 3D Single Stage Object Detectorcited by 440
- 2019Class-balanced Grouping and Sampling for Point Cloud 3D Object Detectioncited by 350
- 2018FlowNet3D: Learning Scene Flow in 3D Point Cloudscited by 242
- 2020Fast and Furious: Real Time End-to-End 3D Detection, Tracking and Motion Forecasting with a Single Convolutional Netcited by 228
- 2018A disciplined approach to neural network hyper-parameters: Part 1 -- learning rate, batch size, momentum, and weight decaycited by 226
15 more
- 2021IntentNet: Learning to Predict Intention from Raw Sensor Datacited by 158
- 2020PnPNet: End-to-End Perception and Prediction with Tracking in the Loopcited by 144
- 2020FLOT: Scene Flow on Point Clouds Guided by Optimal Transportcited by 129
- 2021Neural Scene Flow Priorcited by 91
- 20213D-MAN: 3D Multi-frame Attention Network for Object Detectioncited by 73
- 2017MegDet: A Large Mini-Batch Object Detectorcited by 66
- 2021Moving Object Segmentation in 3D LiDAR Data: A Learning-based Approach Exploiting Sequential Datacited by 64
- 2021Weakly Supervised Learning of Rigid 3D Scene Flowcited by 60
- 2022Dynamic 3D Scene Analysis by Point Cloud Accumulationcited by 47
- 2020An LSTM Approach to Temporal 3D Object Detection in LiDAR Point Cloudscited by 45
- 2020MultiXNet: Multiclass Multistage Multimodal Motion Predictioncited by 31
- 2022Automatic Labeling to Generate Training Data for Online LiDAR-based Moving Object Segmentationcited by 27
- 2021MVFuseNet: Improving End-to-End Object Detection and Motion Forecasting through Multi-View Fusion of LiDAR Datacited by 18
- 2022INT: Towards Infinite-frames 3D Detection with An Efficient Frameworkcited by 14
- 2021Sparse PointPillars: Maintaining and Exploiting Input Sparsity to Improve Runtime on Embedded Systemscited by 11
Built on by
Citing papers per year: 2023: 1.
1 paper.