AI Paper Atlas

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

ICML 2026 · Image generation › Fast diffusion sampling · approach: Denoising diffusion probabilistic models

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 changedWhereMeasuredEffect
add Antipodal Sampling in place of random sampling (direct comparison decoding) (ablation)ASIR · CelebaHQextraction accuracy (%): 68.7% to 76.7%Improved
add Iterative Recovery to random sampling (ablation)Random Sampling · CelebaHQextraction accuracy (%): 68.7% to 94.6%Improved
add Iterative Recovery to Antipodal Sampling (ablation)ASIR · CelebaHQextraction accuracy (%): 76.7% to 99.0%Improved
add Antipodal Sampling to Iterative Recovery decoding (ablation)ASIR · CelebaHQextraction accuracy (%): 94.6 -> 99.0Improved
ASIR instead of Pulsar in pixel spaceASIR · Pixel-based diffusioncapacity (bits): 3610 -> 65536Improved
ASIR instead of Diffusion-Stego in latent spaceASIR · Latent-based diffusioncapacity (bits): 4096 -> 16384Improved
ASIR instead of LDStega in latent spaceASIR · Latent-based diffusiondetection error PE: 0.32 -> 0.50Improved
ASIR instead of Diffusion-Stego in pixel spaceASIR · Pixel-based diffusionextraction accuracy (%): 93.2 -> 99.0Improved
expand candidate pool from {2} to {2,3,4} at tau=0.0 (ablation)ASIR · Churchcapacity (bits): 65536 -> 131072 capacity, 99.0 -> 98.0 accuracyMixed
expand candidate pool from {2} to {2,3,4} at tau=0.0 (ablation)ASIR · Bedroomextraction accuracy (%): 97.0 -> 95.1Hurt
raise threshold tau from 0.0 to 0.6 with CP={2} (ablation)ASIR · Bedroomextraction accuracy (%): 97.0 -> 99.7 accuracy, 65536 -> 14620 capacityMixed
ASIR instead of Pulsar for decoding timeASIR · Computational efficiencydecoding time: 11.8s -> 3.9sImproved

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy