Research map / Adversarial robustness and security / Attacks on text and graphs
CIIDefence: Defeating Adversarial Attacks by Fusing Class-Specific Image Inpainting and Image Denoising
Puneet Gupta, Esa Rahtu
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 class-specific inpainting to wavelet denoising (ablation) | Wavelet denoiser · ImageNet 500-image train set, FGSM | top-1 accuracy: 39.2% -> 87.6% | Improved |
| add class-specific inpainting to wavelet denoising (ablation) | Wavelet denoiser · ImageNet 500-image train set, IGSM | top-1 accuracy: 30.4% -> 93.8% | Improved |
| add wavelet denoising to class-specific inpainting (ablation) | Inpainting only · ImageNet 500-image train set, FGSM | top-1 accuracy: 78.2% -> 87.6% | Improved |
| add wavelet denoising to class-specific inpainting (ablation) | Inpainting only · ImageNet 500-image train set, clean images | top-1 accuracy: 98.2% -> 99.2% | Improved |
| replace image quilting with class-specific inpainting (ablation) | Quilting · ImageNet 500-image train set, clean images | top-1 accuracy: 74.2% -> 98.2% | Improved |
| replace image quilting with class-specific inpainting (ablation) | Quilting · ImageNet 500-image train set, FGSM | top-1 accuracy: 68.4% -> 78.2% | Improved |
| replace JPEG compression with wavelet denoising when combined with inpainting (ablation) | Inpainting + JPEG · ImageNet 500-image train set, FGSM | top-1 accuracy: 81.6% -> 87.6% | Improved |
| add inpainting to JPEG compression (ablation) | JPEG · ImageNet 500-image train set, FGSM | top-1 accuracy: 35.8% -> 81.6% | Improved |
| apply CIIDefence to clean images | ResNet-101 · ImageNet val | top-1 accuracy: 100% -> 98.6% | Hurt |
| add CIIDefence under FGSM | ResNet-101 · ImageNet val FGSM | top-1 accuracy: 18.9% -> 85.7% | Improved |
| add CIIDefence under PGD | ResNet-101 · ImageNet val PGD | top-1 accuracy: 0% -> 97.8% | Improved |
| use CIIDefence instead of PD with wavelet denoising | PD + wavelet [23] · ImageNet val FGSM | destruction rate: 0.638 -> 0.822 | Improved |
Closest papers by meaning
- PixelDefend: Leveraging Generative Models to Understand and Defend against Adversarial Examples (ICLR 2018 · similarity 0.88)
- ComDefend: An Efficient Image Compression Model to Defend Adversarial Examples (CVPR 2019 · similarity 0.86)
- Adversarial Attacks Are Reversible With Natural Supervision (ICCV 2021 · similarity 0.86)
- Feature Denoising for Improving Adversarial Robustness (CVPR 2019 · similarity 0.86)
- Sparse and Imperceivable Adversarial Attacks (ICCV 2019 · similarity 0.86)
- Class-Disentanglement and Applications in Adversarial Detection and Defense (NeurIPS 2021 · similarity 0.86)
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.
