Research map / Adversarial robustness and security / Attacks on text and graphs
Towards Large Yet Imperceptible Adversarial Image Perturbations With Perceptual Color Distance
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Optimize perturbations using CIEDE2000 perceptual color distance instead of RGB L₂ | C&W · ImageNet-Compatible, targeted, budget 9×1000 | C̄₂: 114.36 -> 67.79 | Helped |
| Optimize perturbations using CIEDE2000 perceptual color distance instead of RGB L₂ | C&W · ImageNet-Compatible, targeted, budget 9×1000 | RGB L₂: 0.92 -> 1.22 | Hurt |
| Use alternating classification-loss and perceptual-distance updates instead of DDN's RGB L₂ optimization | DDN · ImageNet-Compatible, targeted, budget 1000 | C̄₂: 111.65 -> 57.10 | Helped |
| Use alternating classification-loss and perceptual-distance updates instead of DDN's RGB L₂ optimization | DDN · ImageNet-Compatible, targeted, budget 1000 | RGB L₂: 0.82 -> 1.13 | Hurt |
| Increase the confidence margin from κ=20 to κ=40 | PerC-AL · ImageNet-Compatible, targeted | C̄₂: 75.43 -> 115.17 | Hurt |
| Add the texture-complexity weighting map to PerC-AL (ablation) | PerC-AL · ImageNet-Compatible images with smooth and visually varied regions | imperceptibility | Helped |
Closest papers by meaning
- Explaining Classifiers Using Adversarial Perturbations on the Perceptual Ball (CVPR 2021 · similarity 0.87)
- Unrestricted Adversarial Examples via Semantic Manipulation (ICLR 2020 · similarity 0.87)
- ColorFool: Semantic Adversarial Colorization (CVPR 2020 · similarity 0.87)
- Learning perturbation sets for robust machine learning (ICLR 2021 · similarity 0.86)
- Training on Foveated Images Improves Robustness to Adversarial Attacks (NeurIPS 2023 · similarity 0.86)
- The Unreasonable Effectiveness of Linear Prediction as a Perceptual Metric (ICLR 2024 · 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.