Research map / Adversarial robustness and security / Attacks on text and graphs
Adversarial Imaging Pipelines
Buu Phan, Fahim Mannan, Felix Heide
Check 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 |
|---|---|---|---|
| Use a targeted attack that optimizes against one ISP while preserving predictions on other ISPs | Camera pipeline attack · Physical ISP attack, Movidius Myriad 2 | success rate: 92.2% | Improved |
| Use a targeted attack that optimizes against one ISP while preserving predictions on other ISPs | Camera pipeline attack · Physical ISP attack, ARM Mali C71 | success rate: 93.8% | Improved |
| Use a targeted attack that optimizes against one ISP while preserving predictions on other ISPs | Camera pipeline attack · Physical ISP attack, Bilateral Filter ISP | success rate: 97.3% | Improved |
| Use a targeted attack that optimizes against one ISP while preserving predictions on other ISPs | Camera pipeline attack · Physical ISP attack, Demosaicing ISP | success rate: 98.2% | Improved |
| Target one optical system while preserving predictions through the other | Camera optics attack · Fujinon CF12.5HA-1 and Cooke Triplet | targeted success rate: 90.7%; 91.5% | Improved |
| Use targeted rather than untargeted camera-specific optimization | Camera pipeline attack · Physical ISP attack, other ISPs | transfer rate: 28.4% -> 4.3%; 34.3% -> 4.8%; 31.3% -> 0.0%; 39.7% -> 0.0% | Improved |
| Use a local proxy fine-tuned with target images and Jacobian augmentation instead of the general proxy (ablation) | Camera attack gradient estimation · Camera pipeline attacks | success rate | Improved |
| Use the general proxy instead of a local proxy for gradient estimation (ablation) | Camera attack gradient estimation · Targeted camera attacks | success rate | Hurt |
| Add bilinear demosaicing as the first layer of the proxy model (ablation) | Differentiable proxy ISP · Proxy training | training stability and accuracy | Improved |
Closest papers by meaning
- All You Need Is RAW: Defending against Adversarial Attacks with Camera Image Pipelines (ECCV 2022 · similarity 0.89)
- Attacking Optical Flow (ICCV 2019 · similarity 0.88)
- Adversarial camera stickers: A physical camera-based attack on deep learning systems (ICML 2019 · similarity 0.86)
- Revisiting Adversarial Patches for Designing Camera-Agnostic Attacks against Person Detection (NeurIPS 2024 · similarity 0.86)
- Deflecting Adversarial Attacks With Pixel Deflection (CVPR 2018 · similarity 0.85)
- NeRFail: Neural Radiance Fields-Based Multiview Adversarial Attack (AAAI 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.
