Research map / Adversarial robustness and security / Backdoor and poisoning
Backdoor Attacks Against Deep Image Compression via Adaptive Frequency Trigger
Read the paperCheck 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 |
|---|---|---|---|
| Use the adaptive frequency trigger instead of LIRA's trigger | Cheng-Anchor compression model · Cityscapes validation, Car to Road | pixel-wise ASR (%): 56.9 -> 71.5 | Helped |
| Use the adaptive frequency trigger instead of LIRA's trigger | Cheng-Anchor compression model · CelebA paired face images | face recognition accuracy (%): 35.3 -> 28.3 | Helped |
| Use the adaptive frequency trigger instead of LIRA's trigger | AE-Hyperprior and Cheng-Anchor · Kodak, BPP attack | bpp | Helped |
| Use the adaptive frequency trigger instead of LIRA's trigger | AE-Hyperprior and Cheng-Anchor · Kodak, PSNR attack | PSNR | Helped |
| Add dynamic balance adjustment to the training loss (ablation) | Adaptive frequency trigger attack · Kodak, BPP attack | bpp | Helped |
| Add topK selection to general trigger generation (ablation) | Adaptive frequency trigger attack · Kodak, BPP attack | poisoned-image bpp: 9.27 -> 9.45 | Helped |
| Add patch-wise weighting to the trigger (ablation) | Adaptive frequency trigger attack · Kodak, BPP attack | poisoned-image bpp: 9.08 -> 9.45 | Helped |
| Use K=4 and N=16 instead of K=16 and N=64 (ablation) | Adaptive frequency trigger attack · Kodak, BPP attack | poisoned-image bpp: 9.45 -> 9.08 | Hurt |
| Use K=64 and N=256 instead of K=16 and N=64 (ablation) | Adaptive frequency trigger attack · Kodak, BPP attack | poisoned-image bpp: 9.45 -> 9.14 | Hurt |
Closest papers by meaning
- Conditional Backdoor Attack via JPEG Compression (AAAI 2024 · similarity 0.89)
- A Dual Stealthy Backdoor: From Both Spatial and Frequency Perspectives (AAAI 2024 · similarity 0.87)
- 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 0.85)
- WaveAttack: Asymmetric Frequency Obfuscation-based Backdoor Attacks Against Deep Neural Networks (NeurIPS 2024 · similarity 0.84)
- Temporal-Distributed Backdoor Attack against Video Based Action Recognition (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.