English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Detecting objects from LiDAR point clouds is an important component ofself-driving car technology as LiDAR provides high resolution spatialinformation. Previous work on point-cloud 3D object detection has re-purposedconvolutional approaches from traditional camera imagery. In this work, wepresent an object detection system called StarNet designed specifically to takeadvantage of the sparse and 3D nature of point cloud data. StarNet is entirelypoint-based, uses no global information, has data dependent anchors, and usessampling instead of learned region proposals. We demonstrate how this designleads to competitive or superior performance on the large Waymo Open Datasetand the KITTI detection dataset, as compared to convolutional baselines. Inparticular, we show how our detector can outperform a competitive baseline onPedestrian detection on the Waymo Open Dataset by more than 7 absolute mAPwhile being more computationally efficient. We show how our redesign---namelyusing only local information and using sampling instead of learnedproposals---leads to a significantly more flexible and adaptable system: wedemonstrate how we can vary the computational cost of a single trained StarNetwithout retraining, and how we can target proposals towards areas of interestwith priors and heuristics. Finally, we show how our design allows forincorporating temporal context by using detections from previous frames totarget computation of the detector, which leads to further improvements inperformance without additional computational cost.

StarNet: Targeted Computation for Object Detection in Point Clouds