Skip to content

Latest commit

 

History

History

config_file

Folders and files

NameName
Last commit message
Last commit date

parent directory

..
 
 
 
 
 
 
 
 

Usage

If you use json mode, it's recommended to open the json file and modify the corresponding parameters. And, if you use command line mode, modify the corresponding parameters, copy it and paste it into command line.

Json mode

Example(train):
(sudo) python main.py (train)

Example(test):
(sudo) python main.py test

Example(steganalysis):
(sudo) python main.py steganalysis

Command line mode

Example(train and validation):
sudo python main.py             \
--task_name xxx                 \
--mode train                    \
--network wasdn                 \
--carrier qmdct                 \
--gpu_selection manu            \
--gpu 0                         \
--height 200                    \
--width 380                     \
--channel 1                     \
--batch_size 64                 \
--cover_train_path xxx          \
--cover_valid_path xxx          \
--stego_train_path xxx          \
--stego_valid_path xxx          \
--tfrecords_path xxx            \
--models_path xxx               \
--logs_path xxx                 \
--task_name xxx                 \

Example(test): 
sudo python main.py             \
--mode test                     \
--network wasdn                 \
--carrier qmdct                 \
--gpu_selection manu            \
--gpu 0                         \
--height 200                    \
--width 380                     \
--channel 1                     \
--batch_size 64                 \
--cover_test_path xxx           \
--stego_test_path xxx           \
--models_path xxx               \

Example(steganalysis for **one**):
sudo python main.py             \
--mode steganalysis             \
--submode one                   \
--network wasdn                 \
--carrier qmdct                 \
--gpu_selection manu            \
--gpu 0                         \
--height 200                    \
--width 380                     \
--channel 1                     \
--steganalysis_file_path xxx    \
--models_path xxx               \

Example(steganalysis for **batch**):
sudo python main.py             \
--mode steganalysis             \
--submode batch                 \
--network wasdn                 \
--carrier qmdct                 \
--gpu_selection manu            \
--gpu 0                         \
--height 200                    \
--width 380                     \
--channel 1                     \
--steganalysis_files_path xxx   \
--models_path xxx               \

Path Mode

There are three modes for dataset path setting: full, simple and semi. The three modes are independent each other, and only one setting is needed at a time.

full

cover_train_path -- path of cover train files
cover_valid_path -- path of cover validation files
stego_train_path -- path of stego train files
stego_valid_path -- path of stego validation files

If you use "full" mode, the cover and stego files for train and validation need placing in different folders.

simple

cover_files_root -- root of cover files
stego_files_root -- root of stego files

If you use "simple" mode, the cover and stego files for train and validation need placing in different folders. And, you should name the two folders as "train" and "validation".

semi

cover_files_path -- path of cover files
stego_files_path -- path of stego files

If you use "semi" mode, the cover and stego files for train and validation are placed in the same folder.

Parameters

All parameters match the mode of argparse, and commonly used parameters are bold.

training and validation

Parameters Function Option
path_mode the mode of path simple(default) and full
task_name the name of task task name is the same with the steg folder name if you select simple path mode
cover_train_path the path of cover train files for full path mode -
cover_valid_path the path of cover validation files for full path mode -
stego_train_path the path of stego train files for full path mode -
stego_valid_path the path of stego validation files for full path mode -
cover_files_root the directory of cover files for simple path mode -
stego_files_root the directory of stego files for simple path mode -
cover_files_path the path of cover files for semi path mode -
stego_files_path the path of stego files for semi path mode -
tfrecords_path the path of tfrecords -
models_path the path of models including checkpoint, data, index and meta files -
logs_path the path of log files including train and validation -
start_index_train the start index of files for train in "semi" mode -
end_index_train the end index of files for validation in "semi" mode -
start_index_valid the start index of files for train in "semi" mode -
end_index_valid the end index of files for validation in "semi" mode -
gpu_selection the mode of gpu selection auto (default), manu
gpu the index of gpu -
mode the mode of running train (default), test and steganalysis
carrier the carrier for steganalysis qmdct (default), audio and image
network the name of designed network -
siamese whether to use siamese network or not true, false (default)
checkpoint whether to continue training on a trained model true, false (default)
fine_tune_model_file_path the path of a trained model
batch_size batch size for training 16 (default)
learning_rate initialized learning rate 1e-3 (default)
epoch epoch of training 500 (default)
seed seed of random generation 1 (default)
is_regulation whether add regulation or not True (default), False
coeff_regulation coefficient of regulation 1e-3 (default)
loss_method method of loss function sigmoid_cross_entropy, softmax_cross_entropy, sparse_softmax_cross_entropy(default)
class_num number of classifier 2 (default)
height height of input data matrix -
width width of input data matrix -
channel channel of input data matrix 1 (default), 3
decay_method method of learning rate decay fixed, step, exponential(defaut), inverse_time, natural_exp, polynomial
decay_step step for learning rate decay 5000 (default)
decay_rate rate of learning rate decay 0.9 (default)
staircase whether the decay the learning rate at discrete intervals or not True and False (default)
max_to_keep number of models to be saved 3 (default)

test

Parameters Function Option
cover_test_path path of cover train files -
stego_test_path path of stego test files -
models_path path of models including checkpoint, data, index and meta files -
gpu_selection mode of gpu selection auto (default), manu
mode mode of running train, test (default) and steganalysis
carrier carrier for steganalysis qmdct (default), audio and image
network name of designed network -
class_num number of classifier 2 (default)
height height of input data matrix -
width width of input data matrix -
channel channel of input data matrix 1 (default), 3

steganalysis

Parameters Function Option
steganalysis_file_path path of file for steganalysis -
steganalysis_files_path path of files for steganalysis -
models_path path of models including checkpoint, data, index and meta files -
gpu_selection mode of gpu selection auto (default), manual
mode mode of running train, test and steganalysis(default)
submode submode of steganalysis one and batch (default, this parameter is just for the mode of steganalysis)
carrier carrier for steganalysis qmdct (default), audio and image
network name of designed network -
class_num number of classifier 2 (default)
height height of input data matrix -
width width of input data matrix -
channel channel of input data matrix 1 (default), 3