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

The scarcity of annotated data in outdoor point cloud segmentation poses asignificant obstacle in harnessing the modeling capabilities of advancednetworks like transformers. Consequently, scholars have been activelyinvestigating efficacious self-supervised pre-training strategies, e.g.contrasting learning and reconstruction-based pretext tasks. Nevertheless,temporal information, which is inherent in the LiDAR point cloud sequence, isconsistently disregarded. To better utilize this property, we propose aneffective pre-training strategy, namely Temporal Masked AutoEncoders (T-MAE),which takes as input temporally adjacent frames and learns temporal dependency.A SiamWCA backbone, containing a Siamese encoder and a window-basedcross-attention (WCA) module, is established for the two-frame input. Takinginto account that the motion of an ego-vehicle alters the illumination anglesof the same instance, temporal modeling also serves as a robust and naturaldata augmentation, enhancing the comprehension of target objects. Moreover,instead of utilizing consecutive frames, it is more cost-effective and powerfulby using distant historical frames. SiamWCA is a powerful architecture butheavily relies on annotated data. With our T-MAE pre-training strategy, weachieve the best performance on the Waymo dataset among self-supervisedlearning methods. Comprehensive experiments are conducted to validate allcomponents of our proposal. Upon acceptance, the source code will be madeaccessible.

T-MAE: Temporal Masked Autoencoders for Point Cloud Representation Learning