Chinese-Landscape-Painting-style-transfer

Chinese Landscape Painting style transfer by adversarial network.

Result

Installation

The code was tested with Anaconda and Python 3.7. After installing the Anaconda environment:

0. Clone the repo:

   git clone https://github.com/Robin-WZQ/Chinese-Landscape-Painting-Style-Transfer.git
   cd Chinese-Landscape-Painting-Style-Transfer

1. Install dependencies:

   For PyTorch dependency, see pytorch.org for more details.

   For custom dependencies:

   pip install -r requirements.txt

2. Download pretrained model from BaiduYun extract code: qls1

   I trained the model for 100 epoches.

3. Pre-process the dataset

   python canny.py
   python picture2texture.py
   python process_all.py

4. Configure your dataset and pretrained model path in opts.py.

   To train the model, please do:

   python pix2pix.py

   To do a style trasfer, please do:

   python test.py

5. In the folder UI, I make a simple interactive interface by tkinter.

   This is a separate folder. Do not use it in the same directory as other folders！！

   To run the DEMO, please do:

   python UI/UI.py

Datasets:

I used the dataset from here

Dataset directory tree is shown below

• Alice Make sure to put the files as the following structure:

  Alice
  ├── Harvard
  │   ├── Harvard_0.jpg
  │   └── ...
  ├── met-1
  │   ├── met_0.jpg
  │   └── ...
  └── met-2
  │   ├── met-221.jpg
  │   └── ...
  

Also, I graped some pictures(nearly 800) from website and I added these pictures into the Alice dataset. U can run tools/Web_Spider.py to generate you own dataset. (And modify names in tools/name.txt)

After preprocessing, it will generate 3 dataset:

• processed by canny (I use Histogram Equalization for image enhancement first), like:

- processed by HED, like:

- canny + HED, like:

Experiments

These models were trained in machine with NVIDIA TITAN X 11gb GPU. I trained it in 100 epoches and bachsize is 1. More details please see in my code.