AI Paper Atlas

Research map / Adversarial robustness and security / Attacks on text and graphs

Exploiting Multi-Object Relationships for Detecting Adversarial Attacks in Complex Scenes

Mingjun Yin, Shasha Li, Zikui Cai, Chengyu Song, M. Salman Asif, Amit K. Roy-Chowdhury, Srikanth V. Krishnamurthy

ICCV 2021 · Adversarial robustness and security › Attacks on text and graphs · approach: Adversarial perturbations beyond images · 26 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
check coarse-grained location as well as category in the consistency score (SCENE-BERT Strict vs Relax) (ablation)SCENE-BERT · PASCAL VOC (F-RCNN)ROC-AUCImproved
use context-aware SCENE-BERT instead of context-agnostic Feature Squeeze, misclassification attacksFaster R-CNN detector · PASCAL VOCROC-AUC: 0.53 -> 0.88Improved
use SCENE-BERT instead of Feature Squeeze, hiding attacks on COCOFaster R-CNN detector · MS COCOROC-AUC: 0.74 -> 0.55Hurt
use SCENE-BERT instead of Feature Squeeze, hiding attacksYOLO detector · PASCAL VOCROC-AUC: 0.75 -> 0.74No clear effect
use SCENE-BERT instead of Feature Squeeze, appearing attacksYOLO detector · MS COCOROC-AUC: 0.67 -> 0.88Improved
use region-level SCEME (Faster R-CNN specific) instead of model-agnostic SCENE-BERT, hiding attacksFaster R-CNN detector · PASCAL VOCROC-AUC: 0.74 -> 0.95Improved
use SCENE-BERT instead of SCEME for appearing attacksFaster R-CNN detector · PASCAL VOCROC-AUC: 0.87 -> 0.88No clear effect

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy