CRoSS: Diffusion Model Makes Controllable, Robust and Secure Image Steganography
Authors: Jiwen Yu, Xuanyu Zhang, Youmin Xu, Jian Zhang
NeurIPS 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | In the experimental section, we conducted detailed experiments to demonstrate the advantages of our proposed CRo SS framework in controllability, robustness, and security. Code is available at https://github.com/vvictoryuki/CRo SS. |
| Researcher Affiliation | Collaboration | Jiwen Yu1 Xuanyu Zhang1 Youmin Xu1,2 Jian Zhang1 1 Peking University Shenzhen Graduate School 2 Peng Cheng Laboratory |
| Pseudocode | Yes | Algorithm 1 The Hide Process of CRo SS. ... Algorithm 2 The Reveal Process of CRo SS. |
| Open Source Code | Yes | Code is available at https://github.com/vvictoryuki/CRo SS. |
| Open Datasets | Yes | To perform a quantitative and qualitative analysis of our method, we collect a benchmark with a total of 260 images and generate corresponding prompt keys specifically tailored for the coverless image steganography, dubbed Stego260. We categorize the dataset into three classes, namely humans, animals, and general objects (such as architecture, plants, food, furniture, etc.). The images in the dataset are sourced from publicly available datasets [1, 2] and Google search engines. For generating prompt keys, we utilize BLIP [32] to generate private keys and employ Chat GPT or artificial adjustment to perform semantic modifications and produce public keys in batches. More details about the dataset can be found in the supplementary material. [1] https://github.com/aisegmentcn/matting_human_datasets, 2019. [2] https://www.kaggle.com/datasets/iamsouravbanerjee/ animal-image-dataset-90-different-animals, 2022. |
| Dataset Splits | No | The paper mentions using a 'benchmark with a total of 260 images' and refers to training/testing implicitly (e.g., 'retrained with leaked samples from testing results'). However, it does not provide explicit numerical splits for training, validation, and testing sets (e.g., percentages or exact counts) or describe a cross-validation setup. |
| Hardware Specification | Yes | All experiments were conducted on a Ge Force RTX 3090 GPU card, and our method did not require any additional training or fine-tuning for the diffusion model. |
| Software Dependencies | No | The paper mentions 'Stable Diffusion [49] v1.5' and 'deterministic DDIM [54]', providing version numbers for Stable Diffusion. However, it does not list other common software dependencies like operating systems, programming languages (e.g., Python), or deep learning frameworks (e.g., PyTorch, TensorFlow) with specific version numbers, which are typically needed for full reproducibility. |
| Experiment Setup | Yes | In our experiment, we chose Stable Diffusion [49] v1.5 as the conditional diffusion model, and we used the deterministic DDIM [54] sampling algorithm. Both the forward and backward processes consisted of 50 steps. To achieve invertible image translation, we set the guidance scale of Stable Diffusion to 1. For the given conditions, which serve as the private and public keys, we had three options: prompts, conditions for Control Nets [74] (depth maps, scribbles, segmentation maps), and Lo RAs [23]. |