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 Tools annual

Global: Zendy Free Plan

Global: Zendy Tools monthly

Global: Zendy Plus monthly

Accurate and fast 3D object detection from point clouds is a key task inautonomous driving. Existing one-stage 3D object detection methods can achievereal-time performance, however, they are dominated by anchor-based detectorswhich are inefficient and require additional post-processing. In this paper, weeliminate anchors and model an object as a single point--the center point ofits bounding box. Based on the center point, we propose an anchor-freeCenterNet3D network that performs 3D object detection without anchors. OurCenterNet3D uses keypoint estimation to find center points and directlyregresses 3D bounding boxes. However, because inherent sparsity of pointclouds, 3D object center points are likely to be in empty space which makes itdifficult to estimate accurate boundaries. To solve this issue, we propose anextra corner attention module to enforce the CNN backbone to pay more attentionto object boundaries. Besides, considering that one-stage detectors suffer fromthe discordance between the predicted bounding boxes and correspondingclassification confidences, we develop an efficient keypoint-sensitive warpingoperation to align the confidences to the predicted bounding boxes. Ourproposed CenterNet3D is non-maximum suppression free which makes it moreefficient and simpler. We evaluate CenterNet3D on the widely used KITTI datasetand more challenging nuScenes dataset. Our method outperforms allstate-of-the-art anchor-based one-stage methods and has comparable performanceto two-stage methods as well. It has an inference speed of 20 FPS and achievesthe best speed and accuracy trade-off. Our source code will be released athttps://github.com/wangguojun2018/CenterNet3d.

CenterNet3D: An Anchor Free Object Detector for Point Cloud