Research map / Adversarial robustness and security / Backdoor and poisoning
Robust and Transferable Backdoor Attacks Against Deep Image Compression With Selective Frequency Prior
Yi Yu, Yufei Wang, Wenhan Yang, Lanqing Guo, Shijian Lu, Ling‐Yu Duan, Yap‐Peng Tan, Alex Chichung Kot
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 dynamic loss balancing (ablation) | BAvAFT · Cheng-Anchor q3 BPP attack | poisoned BPP: 8.52 -> 9.45 | Improved |
| add topK selection in trigger generation (ablation) | BAvAFT · Cheng-Anchor q3 BPP attack | poisoned BPP: 9.27 -> 9.45 | Improved |
| add patch-wise weight (ablation) | BAvAFT · Cheng-Anchor q3 BPP attack | poisoned BPP: 9.08 -> 9.45 | Improved |
| use K=16, N=64 instead of K=4, N=16 (ablation) | BAvAFT · Cheng-Anchor q3 BPP attack | poisoned BPP: 9.08 -> 9.45 | Improved |
| use K=16, N=64 instead of K=64, N=256 (ablation) | BAvAFT · Cheng-Anchor q3 BPP attack | poisoned BPP: 9.14 -> 9.45 | Improved |
| backdoor injection does not hurt clean compression (ablation) | BAvAFT · Kodak clean input | PSNR: 30.81 vs 30.80 | No clear effect |
| use BAvAFT instead of LIRA | LIRA · Cheng-Anchor BPP attack | poisoned BPP: 10.56 -> 11.18 | Improved |
| use BAvAFT instead of Blended | Blended · Cheng-Anchor PSNR attack | poisoned PSNR: 29.84 -> 6.18 | Improved |
| add robust trigger generator and encoder (BAvAFT++) vs BAvAFT | BAvAFT · Cheng-Anchor PSNR attack | poisoned PSNR: 6.18 -> 4.23 | Improved |
| add transfer training (BAvAFT+Trans) | BAvAFT · Cityscapes CarToRoad | pixel-wise ASR: 94.6 -> 99.0 | Improved |
| add transfer training (BAvAFT+Trans) | BAvAFT · CelebA face recognition | attacked accuracy: 22.3 -> 2.8 | Improved |
| use BAvAFT++ instead of BAvAFT | BAvAFT · BPP attack, AE-Hyperprior | poisoned BPP: 6.21 -> 6.07 | Hurt |
Closest papers by meaning
- Backdoor Attacks Against Deep Image Compression via Adaptive Frequency Trigger (CVPR 2023 · similarity 0.98)
- Conditional Backdoor Attack via JPEG Compression (AAAI 2024 · similarity 0.91)
- A Dual Stealthy Backdoor: From Both Spatial and Frequency Perspectives (AAAI 2024 · similarity 0.88)
- Rethinking Backdoor Attacks on Dataset Distillation: A Kernel Method Perspective (ICLR 2024 · similarity 0.86)
- Attention-Imperceptible Backdoor Attacks on Vision Transformers (AAAI 2025 · similarity 0.86)
- Temporal-Distributed Backdoor Attack against Video Based Action Recognition (AAAI 2024 · 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.
