What You See is What You Get: Exploiting Visibility for 3D Object Detection
Preprint
Abstract
Recent advances in 3D sensing have created unique challenges for computer vision. One fundamental challenge is finding a good representation for 3D sensor data. Most popular representations (such as PointNet) are proposed in the context of processing truly 3D data (e.g. points sampled from mesh models), ignoring the fact that 3D sensored data such as a LiDAR sweep is in fact 2.5D. We argue that representing 2.5D data as collections of (x, y, z) points fundamentally destroys hidden information about freespace. In this paper, we demonstrate such knowledge can be efficiently recovered through 3D raycasting and readily incorporated into batch-based gradient learning. We describe a simple approach to augmenting voxel-based networks with visibility: we add a voxelized visibility map as an additional input stream. In addition, we show that visibility can be combined with two crucial modifications common to state-of-the-art 3D detectors: synthetic data augmentation of virtual objects and temporal aggregation of LiDAR sweeps over multiple time frames. On the NuScenes 3D detection benchmark, we show that, by adding an additional stream for visibility input, we can significantly improve the overall detection accuracy of a state-of-the-art 3D detector.Keywords
All Related Versions
This publication has 18 references indexed in Scilit:
- nuScenes: A Multimodal Dataset for Autonomous Driving2020
- 4D Spatio-Temporal ConvNets: Minkowski Convolutional Neural Networks2019
- PointConv: Deep Convolutional Networks on 3D Point Clouds2019
- PointRCNN: 3D Object Proposal Generation and Detection From Point Cloud2019
- PointPillars: Fast Encoders for Object Detection From Point Clouds2019
- Complex-YOLO: An Euler-Region-Proposal for Real-Time 3D Object Detection on Point Clouds2019
- VoxelNet: End-to-End Learning for Point Cloud Based 3D Object Detection2018
- Fast and Furious: Real Time End-to-End 3D Detection, Tracking and Motion Forecasting with a Single Convolutional Net2018
- OctoMap: an efficient probabilistic 3D mapping framework based on octreesAutonomous Robots, 2013
- Using spin images for efficient object recognition in cluttered 3D scenesIEEE Transactions on Pattern Analysis and Machine Intelligence, 1999