Skip to Main content Skip to Navigation
Master thesis

Attention Network for 3D Object Detection in Point Clouds

Anshul Paigwar 1, 2, 3
Abstract : Accurate detection of objects in 3D point clouds is a central problem for autonomous navigation. Most existing methods use techniques of handcrafted features representation or multi-modal approaches prone to sensor failure. Approaches like PointNet that directly operate on sparse point data have shown good accuracy in the classification of single 3D objects. However, LiDAR sensors on Autonomous vehicles generate a large scale pointcloud. Real-time object detection in such a cluttered environment still remains a challenge. In this thesis, we propose Attentional PointNet, a novel end-toend trainable deep architecture for object detection in point clouds. We extend the theory of visual attention mechanism to 3D point clouds and introduce a new recurrent 3D Spatial Transformer Network module. Rather than processing whole point cloud, the network learns "where to look" (find regions of interest), thus significantly reducing the number of points and hence, inference time. Evaluation on KITTI car detection benchmark shows that our Attentional PointNet is notably faster and achieves comparable results with state-of-the-art LiDAR-based 3D detection methods.
Complete list of metadata

Cited literature [60 references]  Display  Hide  Download
Contributor : Anshul Paigwar Connect in order to contact the contributor
Submitted on : Friday, December 6, 2019 - 11:57:14 AM
Last modification on : Tuesday, May 11, 2021 - 11:36:28 AM
Long-term archiving on: : Saturday, March 7, 2020 - 2:33:54 PM


Files produced by the author(s)


  • HAL Id : hal-02396962, version 1



Anshul Paigwar. Attention Network for 3D Object Detection in Point Clouds. Artificial Intelligence [cs.AI]. 2018. ⟨hal-02396962⟩



Les métriques sont temporairement indisponibles