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

Whilst the availability of 3D LiDAR point cloud data has significantly grownin recent years, annotation remains expensive and time-consuming, leading to ademand for semi-supervised semantic segmentation methods with applicationdomains such as autonomous driving. Existing work very often employs relativelylarge segmentation backbone networks to improve segmentation accuracy, at theexpense of computational costs. In addition, many use uniform sampling toreduce ground truth data requirements for learning needed, often resulting insub-optimal performance. To address these issues, we propose a new pipelinethat employs a smaller architecture, requiring fewer ground-truth annotationsto achieve superior segmentation accuracy compared to contemporary approaches.This is facilitated via a novel Sparse Depthwise Separable Convolution modulethat significantly reduces the network parameter count while retaining overalltask performance. To effectively sub-sample our training data, we propose a newSpatio-Temporal Redundant Frame Downsampling (ST-RFD) method that leveragesknowledge of sensor motion within the environment to extract a more diversesubset of training data frame samples. To leverage the use of limited annotateddata samples, we further propose a soft pseudo-label method informed by LiDARreflectivity. Our method outperforms contemporary semi-supervised work in termsof mIoU, using less labeled data, on the SemanticKITTI (59.5@5%) andScribbleKITTI (58.1@5%) benchmark datasets, based on a 2.3x reduction in modelparameters and 641x fewer multiply-add operations whilst also demonstratingsignificant performance improvement on limited training data (i.e., Less isMore).

Less is More: Reducing Task and Model Complexity for 3D Point Cloud Semantic Segmentation