This is the codes for GR5242 Final Project.
We built a neural transfer system and compared the performance of this system under different hyper-parameter settings as well as different neural networks architecture. We compared the results between different neural networks architectures including VGG-19, ResNet50, Xception in our neural transfer system.
The hyper-parameters we studied in this projects are weights of loss & style functions(α and β), noise ratio for the content pictures(θ), learning rate for the gradient descent optimizer(η, we used Stochastic gradient descent as optimizer) We also observed the results under different iterations and random seeds.
./
├── README.md
├── contents
│ └── zelda.jpg
├── main.ipynb
├── outputs
│ ├── 0.jpg
│ ├── 10.jpg
│ ├── 100.jpg
│ ├── 1000.jpg
│ ├── 110.jpg
│ ├── 120.jpg
│ ├── 1200.jpg
│ ├── 130.jpg
│ ├── 140.jpg
│ ├── 1400.jpg
│ ├── 150.jpg
│ ├── 160.jpg
│ ├── 1600.jpg
│ ├── 170.jpg
│ ├── 180.jpg
│ ├── 1800.jpg
│ ├── 190.jpg
│ ├── 1999.jpg
│ ├── 20.jpg
│ ├── 200.jpg
│ ├── 30.jpg
│ ├── 40.jpg
│ ├── 400.jpg
│ ├── 50.jpg
│ ├── 60.jpg
│ ├── 600.jpg
│ ├── 70.jpg
│ ├── 80.jpg
│ ├── 800.jpg
│ └── 90.jpg
├── styles
│ └── kanagawa.jpeg
└── utils
└── model.py
4 directories, 35 files