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

Accurate 3D object detection in large-scale outdoor scenes, characterized byconsiderable variations in object scales, necessitates features rich in bothlong-range and fine-grained information. While recent detectors have utilizedwindow-based transformers to model long-range dependencies, they tend tooverlook fine-grained details. To bridge this gap, we propose MsSVT++, aninnovative Mixed-scale Sparse Voxel Transformer that simultaneously capturesboth types of information through a divide-and-conquer approach. This approachinvolves explicitly dividing attention heads into multiple groups, eachresponsible for attending to information within a specific range. The outputsof these groups are subsequently merged to obtain final mixed-scale features.To mitigate the computational complexity associated with applying awindow-based transformer in 3D voxel space, we introduce a novel ChessboardSampling strategy and implement voxel sampling and gathering operationssparsely using a hash map. Moreover, an important challenge stems from theobservation that non-empty voxels are primarily located on the surface ofobjects, which impedes the accurate estimation of bounding boxes. To overcomethis challenge, we introduce a Center Voting module that integrates newly votedvoxels enriched with mixed-scale contextual information towards the centers ofthe objects, thereby improving precise object localization. Extensiveexperiments demonstrate that our single-stage detector, built upon thefoundation of MsSVT++, consistently delivers exceptional performance acrossdiverse datasets.

MsSVT++: Mixed-scale Sparse Voxel Transformer with Center Voting for 3D Object Detection