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 Tools annual

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 Tools monthly

Global: Zendy Plus monthly

Presents the access of premium content as premium feature

Premium Content

Global: Zendy Plus annual

Global: Zendy Free Plan

Despite recent success of self-supervised based contrastive learning modelfor 3D point clouds representation, the adversarial robustness of suchpre-trained models raised concerns. Adversarial contrastive learning (ACL) isconsidered an effective way to improve the robustness of pre-trained models. Incontrastive learning, the projector is considered an effective component forremoving unnecessary feature information during contrastive pretraining andmost ACL works also use contrastive loss with projected feature representationsto generate adversarial examples in pretraining, while "unprojected " featurerepresentations are used in generating adversarial inputs duringinference.Because of the distribution gap between projected and "unprojected"features, their models are constrained of obtaining robust featurerepresentations for downstream tasks. We introduce a new method to generatehigh-quality 3D adversarial examples for adversarial training by utilizingvirtual adversarial loss with "unprojected" feature representations incontrastive learning framework. We present our robust aware loss function totrain self-supervised contrastive learning framework adversarially.Furthermore, we find selecting high difference points with the Difference ofNormal (DoN) operator as additional input for adversarial self-supervisedcontrastive learning can significantly improve the adversarial robustness ofthe pre-trained model. We validate our method, PointACL on downstream tasks,including 3D classification and 3D segmentation with multiple datasets. Itobtains comparable robust accuracy over state-of-the-art contrastiveadversarial learning methods.

PointACL:Adversarial Contrastive Learning for Robust Point Clouds Representation under Adversarial Attack