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

3D object detection using point clouds has attracted increasing attention dueto its wide applications in autonomous driving and robotics. However, mostexisting studies focus on single point cloud frames without harnessing thetemporal information in point cloud sequences. In this paper, we designTransPillars, a novel transformer-based feature aggregation technique thatexploits temporal features of consecutive point cloud frames for multi-frame 3Dobject detection. TransPillars aggregates spatial-temporal point cloud featuresfrom two perspectives. First, it fuses voxel-level features directly frommulti-frame feature maps instead of pooled instance features to preserveinstance details with contextual information that are essential to accurateobject localization. Second, it introduces a hierarchical coarse-to-finestrategy to fuse multi-scale features progressively to effectively capture themotion of moving objects and guide the aggregation of fine features. Besides, avariant of deformable transformer is introduced to improve the effectiveness ofcross-frame feature matching. Extensive experiments show that our proposedTransPillars achieves state-of-art performance as compared to existingmulti-frame detection approaches. Code will be released.

TransPillars: Coarse-to-Fine Aggregation for Multi-Frame 3D Object Detection