STROKE uses sparse rectangular occlusions for black-box image attacks
Original title: STROKE: Simple Sparse Structural Occlusions for Black-Box Adversarial Attacks
The score-based method reports high success at 1–2% area with low query counts and shorter wall-clock time than matched patch attacks.
The paper introduces STROKE, a query-efficient score-based black-box adversarial attack that builds perturbations from simple oriented rectangular occlusions. A greedy add-and-refine procedure optimizes those occlusions. Across image classifiers, the authors report high attack success with low query counts at nominal 1–2% area settings, and favorable results versus existing patch-based attacks in several configurations. Matched comparisons show large cuts in wall-clock attack time. Ablations cover design choices and other structural primitives.
Unconfirmed: Success rates and timing gains are the authors' results; victim models are not named in the abstract.
- Perturbations are oriented rectangles rather than arbitrary patches.
- Optimization is a lightweight greedy add-and-refine loop.
- Ablations compare effectiveness across structural primitives.
- The attack is score-based and does not need model gradients.
From the original article
1 min read at arxiv.orgWe introduce STROKE, a query-efficient score-based black-box adversarial attack that constructs perturbations from simple oriented rectangular occlusions. STROKE optimizes these perturbations through a lightweight greedy add-and-refine procedure.
By Ankit Gupta · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.