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/

Zendy Plus

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

Zendy Tools

Zendy Open

Airborne laser scanning (ALS) point cloud segmentation is a fundamental taskfor large-scale 3D scene understanding. In real-world applications, models aretypically fixed after training. However, domain shifts caused by changes in theenvironment, sensor types, or sensor degradation often lead to a decline inmodel performance. Continuous Test-Time Adaptation (CTTA) offers a solution byadapting a source-pretrained model to evolving, unlabeled target domains.Despite its potential, research on ALS point clouds remains limited, facingchallenges such as the absence of standardized datasets and the risk ofcatastrophic forgetting and error accumulation during prolonged adaptation. Totackle these challenges, we propose APCoTTA, the first CTTA method tailored forALS point cloud semantic segmentation. We propose a dynamic trainable layerselection module. This module utilizes gradient information to selectlow-confidence layers for training, and the remaining layers are kept frozen,mitigating catastrophic forgetting. To further reduce error accumulation, wepropose an entropy-based consistency loss. By losing such samples based onentropy, we apply consistency loss only to the reliable samples, enhancingmodel stability. In addition, we propose a random parameter interpolationmechanism, which randomly blends parameters from the selected trainable layerswith those of the source model. This approach helps balance target adaptationand source knowledge retention, further alleviating forgetting. Finally, weconstruct two benchmarks, ISPRSC and H3DC, to address the lack of CTTAbenchmarks for ALS point cloud segmentation. Experimental results demonstratethat APCoTTA achieves the best performance on two benchmarks, with mIoUimprovements of approximately 9% and 14% over direct inference. The newbenchmarks and code are available at https://github.com/Gaoyuan2/APCoTTA.

APCoTTA: Continual Test-Time Adaptation for Semantic Segmentation of Airborne LiDAR Point Clouds