Research map / Image generation / Fast diffusion sampling
ASIR: Steganography for Diffusion Models via Antipodal Sampling and Iterative Recovery
Yaofei Wang, Yufeng Zheng, Han Fang, Wenzhao Cao, Donghui Hu
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 Antipodal Sampling in place of random sampling (direct comparison decoding) (ablation) | ASIR · CelebaHQ | extraction accuracy (%): 68.7% to 76.7% | Improved |
| add Iterative Recovery to random sampling (ablation) | Random Sampling · CelebaHQ | extraction accuracy (%): 68.7% to 94.6% | Improved |
| add Iterative Recovery to Antipodal Sampling (ablation) | ASIR · CelebaHQ | extraction accuracy (%): 76.7% to 99.0% | Improved |
| add Antipodal Sampling to Iterative Recovery decoding (ablation) | ASIR · CelebaHQ | extraction accuracy (%): 94.6 -> 99.0 | Improved |
| ASIR instead of Pulsar in pixel space | ASIR · Pixel-based diffusion | capacity (bits): 3610 -> 65536 | Improved |
| ASIR instead of Diffusion-Stego in latent space | ASIR · Latent-based diffusion | capacity (bits): 4096 -> 16384 | Improved |
| ASIR instead of LDStega in latent space | ASIR · Latent-based diffusion | detection error PE: 0.32 -> 0.50 | Improved |
| ASIR instead of Diffusion-Stego in pixel space | ASIR · Pixel-based diffusion | extraction accuracy (%): 93.2 -> 99.0 | Improved |
| expand candidate pool from {2} to {2,3,4} at tau=0.0 (ablation) | ASIR · Church | capacity (bits): 65536 -> 131072 capacity, 99.0 -> 98.0 accuracy | Mixed |
| expand candidate pool from {2} to {2,3,4} at tau=0.0 (ablation) | ASIR · Bedroom | extraction accuracy (%): 97.0 -> 95.1 | Hurt |
| raise threshold tau from 0.0 to 0.6 with CP={2} (ablation) | ASIR · Bedroom | extraction accuracy (%): 97.0 -> 99.7 accuracy, 65536 -> 14620 capacity | Mixed |
| ASIR instead of Pulsar for decoding time | ASIR · Computational efficiency | decoding time: 11.8s -> 3.9s | Improved |
Closest papers by meaning
- Hiding Images in Diffusion Models by Editing Learned Score Functions (CVPR 2025 · similarity 0.88)
- SSHR: More Secure Generative Steganography with High-Quality Revealed Secret Images (ICML 2025 · similarity 0.87)
- TERD: A Unified Framework for Safeguarding Diffusion Models Against Backdoors (ICML 2024 · similarity 0.86)
- Variance as a Catalyst: Efficient and Transferable Semantic Erasure Adversarial Attack for Customized Diffusion Models (ICML 2025 · similarity 0.85)
- T2SMark: Balancing Robustness and Diversity in Noise-as-Watermark for Diffusion Models (NeurIPS 2025 · similarity 0.85)
- LoRID: Low-Rank Iterative Diffusion for Adversarial Purification (AAAI 2025 · similarity 0.85)
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.
