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

State-of-the-art object detectors are vulnerable to localized patch hidingattacks, where an adversary introduces a small adversarial patch to makedetectors miss the detection of salient objects. The patch attacker can carryout a physical-world attack by printing and attaching an adversarial patch tothe victim object. In this paper, we propose DetectorGuard as the first generalframework for building provably robust object detectors against localized patchhiding attacks. DetectorGuard is inspired by recent advancements in robustimage classification research; we ask: can we adapt robust image classifiersfor robust object detection? Unfortunately, due to their task difference, anobject detector naively adapted from a robust image classifier 1) may notnecessarily be robust in the adversarial setting or 2) even maintain decentperformance in the clean setting. To build a high-performance robust objectdetector, we propose an objectness explaining strategy: we adapt a robust imageclassifier to predict objectness for every image location and then explain eachobjectness using the bounding boxes predicted by a conventional objectdetector. If all objectness is well explained, we output the predictions madeby the conventional object detector; otherwise, we issue an attack alert.Notably, 1) in the adversarial setting, we formally prove the end-to-endrobustness of DetectorGuard on certified objects, i.e., it either detects theobject or triggers an alert, against any patch hiding attacker within ourthreat model; 2) in the clean setting, we have almost the same performance asstate-of-the-art object detectors. Our evaluation on the PASCAL VOC, MS COCO,and KITTI datasets further demonstrates that DetectorGuard achieves the firstprovable robustness against localized patch hiding attacks at a negligible cost(<1%) of clean performance.

DetectorGuard: Provably Securing Object Detectors against Localized Patch Hiding Attacks