AI Paper Atlas

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

Read or Ignore? A Unified Benchmark for Typographic-Attack Robustness and Text Recognition in Vision-Language Models

Futa Waseda, Shojiro Yamabe, Daiki Shiono, Kento Sasaki, Tsubasa Takahashi

ECCV 2026 · Adversarial robustness and security › Attacks on text and graphs · approach: Vision-language model robustness · 5 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
add text-attack data to training (RIO-RT instead of IT-RT) (ablation)LLaVA-1.5-7B · Text-VQA attack avgVQA accuracy: 23.3 -> 52.8Improved
add text-attack data to training (RIO-RT instead of IT-RT) (ablation)LLaVA-1.5-13B · Text-VQA attack avgVQA accuracy: 12.3 -> 50.5Improved
train on object-attack data only (IT-RT)LLaVA-1.5-7B · Text-VQA attack avgVQA accuracy: 45.1 -> 23.3Hurt
train on object-attack data only (IT-RT)Qwen3-VL-8B · Text-VQA attack avgVQA accuracy: 78.5 -> 60.8Hurt
train on object-attack data only (IT-RT)LLaVA-1.5-7B · Obj-VQA MC attack avgaccuracy: 71.6 -> 98.4Improved
balanced read-or-ignore training (RIO-RT)Qwen2.5-VL-7B · Obj-VQA MC attack avgaccuracy: 65.9 -> 98.6Improved
balanced read-or-ignore training (RIO-RT)Llama-3.2-Vision-11B · Text-VQA attack avgVQA accuracy: 58.2 -> 80.8Improved
balanced read-or-ignore training (RIO-RT)SmolVLM-2B · Text-VQA attack avgVQA accuracy: 61.7 -> 62.9No clear effect
apply CoT defense promptingQwen2.5-VL-7B · Obj-VQA MC attack avgaccuracy: 65.9 -> 77.2Improved
apply CoT defense promptingQwen2.5-VL-7B · Text-VQA attack avgVQA accuracy: 77.7 -> 68.0Hurt
apply CoT defense promptingQwen3-VL-8B · Obj-VQA MC attack avgaccuracy: 73.2 -> 67.0Hurt
add language-model LoRA to vision-only tuning (V to V+L) (ablation)Qwen2.5-VL-7B · Obj-VQA MC attackaccuracy: 62.3 -> 98.5Improved

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