This is the code I used in my bachelor thesis, which can be found here:
No functionality changes were made since, only a small code clean-up.
What you need:
- the dataset
/Dataset_visual_change/from https://zenodo.org/record/3908124 - extract and integrate the included view masks from
view_masks.zipin the/Dataset_visual_change/data structure
Install following python modules:
- opencv-python
- torch
- colorama
After that run train_cnn.py help in your terminal for further instructions.
Run the script with:
python3 train.py PATH/TO/DATA/ model_name <WEBSITE_NAME> <USER> modi**(optional)
<WEBSITE_NAME> may be any combination of:
amazon, cnn, gm, guardian, kia, mayo, nih, nissan, reddit, steam, walmart, webmd
<USER> may be any combination of:
p1, p2, p3, p4
Several users and websites are seperated by space, do not use any comma.
Include help as parameter to show all available modi.
Run the script with:
python3 test.py PATH/TO/DATA/ model_name <WEBSITE_NAME> <USER> modi**(optional)
Parameter work the same as with training.
WARNING: Both training and testing may take a very long time!