Model weights can be downloaded at https://drive.google.com/open?id=1wOhXdC9jWjYH60-qwTwegv1W-rWH5dVk The model is trained on the S-Uniward 0.4bpp in the same setting as reported in the paper: "Deep Residual Network for Steganalysis of Digital Images" The model can be tested using the file test.py The tensorflow code of the same can be found at: http://dde.binghamton.edu/download/feature_extractors/
The test accuracy reported in the paper is 89.77%. My implementation achieved 89.43% on S-Uniward 0.4bpp.
The model is trained and tested on Tesla V-100-DGX with 32GB GPU.
| SRNet architecture | Train/Validation/Test Accuracies |
 |
 |