Bh4r4t/Neural-Style-Transfer

PyTorch Implementation of Image Stylization method using vgg19 net.

Neural-Style-Transfer

A PyTorch implementation of the paper Image Style Transfer Using Convolutional Neural Network by LA Gatys et al. - CVPR 2016.

How to run

1. Install all dependencies

   pip install -r requirements.txt
   

2. To Run

   python main.py --content_img <content_image_path> --style_img <style_image_path>
   

3. To check for other arguments, run

   python main.py -h
   

Model Description

1. Style features are extracted using the conv1_1, conv2_1, conv3_1, conv4_1, conv5_1 layers and content features from conv4_2 layer of vgg19 net. Here I have used pretrained vgg19 net.

2. Algorithm of style transfer as proposed by the authors.

3. α (alpha) is content weight and β (beta) is style weight. Often the β is kept much larger than α, but sometimes it depends on the style image, and how much the style is to be superimposed with content.

Artistic results generated using Above model