This is the PyTorch implementation of the manuscript "Universal Deep Network for Steganalysis of Color Image based on Channel Representation".
CUDA (10.2) cuDNN (7.4.1) python (3.6.9)
"UCNet_Spatial.py" and "UCNet_JPEG.py" are the main program in spatial and JPEG domain, respectively.
"High-pass filters" contains the 30 SRM filters.
"J-UNIWARD-pretrain-parameters.pt" is pretrain parameter under a small ImageNet (training/validation:190,000/8,820 images) with J-UNIWARD steganography.
Example:
If you want to detect CMDC-HILL steganography method at 0.4 bpc (on GPU #1), you can enter following command:
"python3 UCNet_Spatial.py -alg CMDC-HILL -rate 0.4 -g 1"
If you have any question, please contact me. (revere.wei@outlook.com)