DiffStega: Towards Universal Training-Free Coverless Image Steganography with Diffusion Models

Authors: Yiwei Yang, Zheyuan Liu, Jun Jia, Zhongpai Gao, Yunhao Li, Wei Sun, Xiaohong Liu, Guangtao Zhai

IJCAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments indicate substantial improvements in our method over existing ones, particularly in aspects of versatility, password sensitivity, and recovery quality. Codes are available at https://github.com/evtricks/Diff Stega.
Researcher Affiliation Collaboration Yiwei Yang1 , Zheyuan Liu1 , Jun Jia1 , Zhongpai Gao2 , Yunhao Li1 , Wei Sun1 , Xiaohong Liu1 and Guangtao Zhai1 1Shanghai Jiao Tong University 2United Imaging Intelligence
Pseudocode No The paper describes the steps of its pipeline but does not include any formally labeled pseudocode or algorithm blocks.
Open Source Code Yes Codes are available at https://github.com/evtricks/Diff Stega.
Open Datasets Yes All images are from public dataset COCO [Lin et al., 2014], AFHQ [Choi et al., 2020], FFHQ [Karras et al., 2019], Celeb A-HQ [Karras et al., 2018] and Internet, center cropped and resized to 512 512.
Dataset Splits No The paper describes the Uni Stega dataset and its subsets but does not specify training, validation, and test splits (e.g., percentages or counts) for model reproduction.
Hardware Specification Yes All experiments are conducted on single Nvidia RTX 3090 GPU, requiring no additional training or fine-tuning.
Software Dependencies No The paper mentions using "pre-trained SD v1.5", "Pic X real", and "IP-Adapter-plus" but does not specify the versions of underlying software dependencies like Python, PyTorch, etc.
Experiment Setup Yes We set T = 50, and the mixing coefficient of EDICT is 0.93. We use IP-Adapter-plus [Ye et al., 2023] in Guidance Injection, and its weight factor is 1. The guidance scale of diffusion models is 1. η = 0.05 in Noise Flip. The diffusion process for ours is executed over steps [0, ξT]. We set ξ = 0.7 for experiments on style prompts and ξ = 0.6 for other prompts.