Research map / Adversarial robustness and security / Black-box and transfer attacks
Improving the Transferability of Adversarial Examples with Resized-Diverse-Inputs, Diversity-Ensemble and Region Fitting
Junhua Zou, Zhisong Pan, Junyang Qiu, Xin Liu, Ting Rui, Wei Li
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 |
|---|---|---|---|
| replace DIM with RDIM (no probability, larger diversity size, resize back) | TI-DIM, Inc-v3 source · Inc-v3ens3 black-box, single model | attack success rate (%): 46.6 -> 59.1 | Improved |
| replace DIM with RDIM | NI-TI-DIM, Inc-v3 source · Inc-v3ens3 black-box, single model | attack success rate (%): 50.0 -> 53.4 | Improved |
| replace DIM with RDIM | TI-DIM, Res-v2-152 source · HGD black-box, single model | attack success rate (%): 52.6 -> 53.4 | Improved |
| replace SIM with DEM multi-scale diversity ensemble | SI-TIM, Inc-v3 source · Inc-v3ens3 black-box, single model | attack success rate (%): 48.4 -> 70.1 | Improved |
| replace SIM with DEM | SI-NI-TIM, Inc-v3 source · Inc-v3ens3 black-box, single model | attack success rate (%): 52.1 -> 66.4 | Improved |
| replace SIM with DEM | SI-TIM, IncRes-v2 source · HGD black-box, single model | attack success rate (%): 60.6 -> 80.0 | Improved |
| replace DIM with RDIM in ensemble attack | TI-DIM, ensemble of four models · IncRes-v2ens black-box, ensemble | attack success rate (%): 78.5 -> 79.1 | Improved |
| replace DIM with RDIM in ensemble attack | TI-DIM, ensemble of four models · HGD black-box, ensemble | attack success rate (%): 83.0 -> 82.1 | Hurt |
| add region fitting (step size epsilon instead of alpha) (ablation) | TI-RDIM, ensemble of four models · Inc-v3ens3 black-box, ensemble | attack success rate (%): 85.0 -> 91.3 | Improved |
| add region fitting (ablation) | DE-TIM, ensemble of four models · Inc-v3ens3 black-box, ensemble | attack success rate (%): 91.2 -> 94.7 | Improved |
| add region fitting (ablation) | DE-TIM, ensemble of four models · IncRes-v2ens black-box, ensemble | attack success rate (%): 88.2 -> 89.1 | Improved |
| add Nesterov momentum (NI) to DE-TIM | DE-TIM, ensemble of four models · IncRes-v2ens black-box, ensemble | attack success rate (%): 88.2 -> 69.8 | Hurt |
Closest papers by meaning
- Understanding Model Ensemble in Transferable Adversarial Attack (ICML 2025 · similarity 0.88)
- Ensemble Diversity Facilitates Adversarial Transferability (CVPR 2024 · similarity 0.88)
- Enhancing Adversarial Transferability with Checkpoints of a Single Model's Training (CVPR 2025 · similarity 0.88)
- Rethinking Model Ensemble in Transfer-based Adversarial Attacks (ICLR 2024 · similarity 0.88)
- Strong Transferable Adversarial Attacks via Ensembled Asymptotically Normal Distribution Learning (CVPR 2024 · similarity 0.88)
- Enhancing Adversarial Transferability with Adversarial Weight Tuning (AAAI 2025 · similarity 0.88)
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.
