Research map / Adversarial robustness and security / Attacks on text and graphs
ViewFool: Evaluating the Robustness of Visual Recognition to Adversarial Viewpoints
Yinpeng Dong, Shouwei Ruan, Hang Su, Caixin Kang, Xingxing Wei, Jun Zhu
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 changed | Where | Measured | Effect |
|---|---|---|---|
| add entropic regularizer (λ=0 to 0.01) (ablation) | ViewFool · ResNet-50 translation+rotation, optimal viewpoint | attack success rate (%): 96 -> 98 | Improved |
| add entropic regularizer (λ=0 to 0.01) (ablation) | ViewFool · ResNet-50 translation+rotation, viewpoint distribution | attack success rate (%): 92.01 -> 88.79 | Hurt |
| add entropic regularizer (ablation) | ViewFool · ViT-B/16 translation+rotation, real image | attack success rate (%): 77 -> 88 | Improved |
| add entropic regularizer (ablation) | ViewFool · ResNet-50 transfer to Swin-B | attack success rate (%): 55.81 -> 59.62 | Improved |
| add entropic regularizer (ablation) | ViewFool · ViT-B/16 transfer to Mixer-B | attack success rate (%): 87.80 -> 85.72 | Hurt |
| optimize viewpoint instead of random search | ViewFool · ResNet-50 translation+rotation | attack success rate (%): 53.98 -> 88.79 | Improved |
| optimize translation and rotation jointly instead of alone (ablation) | ViewFool · ResNet-50 | attack success rate (%): 55.24 -> 88.79 | Improved |
| increase λ from 0.01 to 1.0 (ablation) | ViewFool · ResNet-50 | attack success rate (%): 88.79 -> 77.00 | Hurt |
| increase λ from 0 to 1.0 (ablation) | ViewFool · ResNet-50 | std of psi: 4.554 -> 23.937 | Improved |
| use 3D viewpoints instead of 2D transformations (ablation) | ViewFool · ResNet-50 | attack success rate (%): 76.93 -> 88.79 | Improved |
| widen rotation ranges to full (ablation) | ViewFool · ResNet-50 rotation only | attack success rate (%): 68.57 -> 84.25 | Improved |
| optimize only theta rather than all three angles (ablation) | ViewFool · ResNet-50 | attack success rate (%): 84.25 -> 59.64 | Hurt |
Closest papers by meaning
- Revisiting Adversarial Training at Scale (CVPR 2024 · similarity 0.86)
- MVI-Bench: A Comprehensive Benchmark for Evaluating Robustness to Misleading Visual Inputs in LVLMs (ICML 2026 · similarity 0.85)
- A Black-Box Evaluation Framework for Semantic Robustness in Bird’s Eye View Detection (AAAI 2025 · similarity 0.85)
- VLM-RobustBench: A Comprehensive Benchmark for Robustness of Vision-Language Models (ICML 2026 · similarity 0.85)
- NeRFail: Neural Radiance Fields-Based Multiview Adversarial Attack (AAAI 2024 · similarity 0.85)
- LORE: Lagrangian-Optimized Robust Embeddings for Visual Encoders (NeurIPS 2025 · similarity 0.84)
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.
