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

As Segment Anything Model (SAM) becomes a popular foundation model incomputer vision, its adversarial robustness has become a concern that cannot beignored. This works investigates whether it is possible to attack SAM withimage-agnostic Universal Adversarial Perturbation (UAP). In other words, weseek a single perturbation that can fool the SAM to predict invalid masks formost (if not all) images. We demonstrate convetional image-centric attackframework is effective for image-independent attacks but fails for universaladversarial attack. To this end, we propose a novel perturbation-centricframework that results in a UAP generation method based on self-supervisedcontrastive learning (CL), where the UAP is set to the anchor sample and thepositive sample is augmented from the UAP. The representations of negativesamples are obtained from the image encoder in advance and saved in a memorybank. The effectiveness of our proposed CL-based UAP generation method isvalidated by both quantitative and qualitative results. On top of the ablationstudy to understand various components in our proposed method, we shed light onthe roles of positive and negative samples in making the generated UAPeffective for attacking SAM.

SAM Meets UAP: Attacking Segment Anything Model With Universal Adversarial Perturbation