Implementation of "On Function-Coupled Watermarks for Deep Neural Networks"

Datasets:

CIFAR10
CIFAR100
MNIST
Tiny-ImageNet

Network Structures:

LeNet-5
VGG-16
ResNet-18

Models:

LeNet-5 on MNIST
VGG-16 on CIFAR-100
ResNet-18 on CIFAR-10 and Tiny-ImageNet

Script Structure and Description:

Main folder
    |--checkpoint
        |--checkpoint list
    |--data
        |--data list
    |--models
        |--VGG16
        |--ResNet
        |--LeNet
        |--...
    |--pytorch_grad_cam
        |--going to delete
    --data_loader.py (load data)
    --finetune-with-same-dataset.py (script of finetuning a model)
    --generate-finetune-data-same-dataset.py (script of generating data for finetuning)
    --prune_model.py (prune a model)
    --pruning.py (dependent library for pruning)
    --select_combine_img.py (generate wm images for embedding watermarks)
    --select_wm_images.py (select wm images for validation)
    --show_tiny_imagenet.ipynb (show something)
    --test_acc.py (test the benign accuracy of a model)
    --test_injection.py (test the wm performance)
    --tinyimagenet-wm.py (core script to train a model for wm embedding)
    --tinyimagenet.py (train a clean model)
    --test_injection_noise.py (test the robustness under noise preprocessing)
    --test_injection_flip.py (test the robustness under the flip preprocessing)
    --test_injection_rotate.py (test the robustness under the rotation preprocessing)

Run:

python tinyimagenet-wm.py