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

This paper considers a network referred to as SoftGroup for accurate andscalable 3D instance segmentation. Existing state-of-the-art methods producehard semantic predictions followed by grouping instance segmentation results.Unfortunately, errors stemming from hard decisions propagate into the grouping,resulting in poor overlap between predicted instances and ground truth andsubstantial false positives. To address the abovementioned problems, SoftGroupallows each point to be associated with multiple classes to mitigate theuncertainty stemming from semantic prediction. It also suppresses falsepositive instances by learning to categorize them as background. Regardingscalability, the existing fast methods require computational time on the orderof tens of seconds on large-scale scenes, which is unsatisfactory and far fromapplicable for real-time. Our finding is that the $k$-Nearest Neighbor ($k$-NN)module, which serves as the prerequisite of grouping, introduces acomputational bottleneck. SoftGroup is extended to resolve this computationalbottleneck, referred to as SoftGroup++. The proposed SoftGroup++ reduces timecomplexity with octree $k$-NN and reduces search space with class-aware pyramidscaling and late devoxelization. Experimental results on various indoor andoutdoor datasets demonstrate the efficacy and generality of the proposedSoftGroup and SoftGroup++. Their performances surpass the best-performingbaseline by a large margin (6\% $\sim$ 16\%) in terms of AP$_{50}$. On datasetswith large-scale scenes, SoftGroup++ achieves a 6$\times$ speed boost onaverage compared to SoftGroup. Furthermore, SoftGroup can be extended toperform object detection and panoptic segmentation with nontrivial improvementsover existing methods. The source code and trained models are available at\url{https://github.com/thangvubk/SoftGroup}.

Scalable SoftGroup for 3D Instance Segmentation on Point Clouds