matthias-wright/siamese_architecture

PyTorch implementation of the siamese architecture for style transfer I developed for my master's thesis.

A Siamese Architecture for Neural Style Transfer

The implementation of the siamese architecture that I developed for my master's thesis.

Dependencies

• Python 3.5.2+
• NumPy 1.11.0+
• PyTorch 0.4.1+
• PIL 1.1.7+
• Matplotlib 3.0.1+

Example

Instructions

Training

import siamese

data_photo = '...' # file path to the photograph images (RGB) folder
data_oil = '...' # file path to the oil images (RGB) folder
siam_model = siamese.Siamese(in_path_photo=data_photo,
                             in_path_oil=data_oil, 
                             autoencoder_path='saves/autoencoders', 
                             num_epochs=100, batch_size=50,
                             learning_rate=0.0002, 
                             recon_loss_weight=10, 
                             penalty_coef=10, verbose=True)
                          
siam_model.train()

Load Pretrained Model

import siamese

data_photo = '...' # file path to the photograph images (RGB) folder
data_oil = '...' # file path to the oil images (RGB) folder
siam_model = siamese.Siamese(in_path_photo=data_photo,
                             in_path_oil=data_oil, 
                             autoencoder_path='saves/autoencoders', 
                             num_epochs=100, batch_size=50,
                             learning_rate=0.0002, 
                             recon_loss_weight=10, 
                             penalty_coef=10, verbose=True)

siam_model.load(epoch=18, path='saves')                                 

Style Transfer

import siamese
import util

siam_model = siamese.Siamese(autoencoder_path='saves/autoencoders')
siam_model.load(epoch=18, path='saves')     

x_photo = util.image_to_tensor('image.jpg')
x_oil = siam_model.translate_photo_to_oil(x_photo)
x_oil = util.tensor_to_numpy(x_oil)
x_oil = util.numpy_to_image(x_oil)
x_oil.save('stylized_image.png')                                    

License

This project is licensed under the MIT Licence