Neural Style Transfer

This is a Tensorflow implementation for several techniques in the following papers:

"A Neural Algorithm of Artistic Style" by Leon A. Gatys, Alexander S. Ecker, Matthias Bethge

"Image Style Transfer Using Convolutional Neural Networks" by Leon A. Gatys, Alexander S. Ecker, Matthias Bethge

"Preserving Color in Neural Artistic Style Transfer" by Leon A. Gatys, Matthias Bethge, Aaron Hertzmann, Eli Shechtman

"Laplacian-Steered Neural Style Transfer" by Shaohua Li, Xinxing Xu, Liqiang Nie, Tat-Seng Chua

Setup

This code is based on Tensorflow. It has been tested on Windows 10 and Ubuntu 16.04 LTS.
Dependencies:

• Tensowflow
• nmupy, matplotlib, scipy, Pillow  
• opencv Recommended but optional:
• CUDA
• cudnn

Download VGG19 via following URL:

http://www.vlfeat.org/matconvnet/models/imagenet-vgg-verydeep-19.mat

And please save the file imagenet-vgg-verydeep-19.mat under pre_trained_model

Commands

Quick start

python run_main.py --content images/01.jpg --style images/starry-night.jpg --output result.jpg

Multi-Style transfer

python run_main.py --content images/01.jpg --style images/starry-night.jpg --output result_multi-style.jpg --multi_style True --style_ratio 0.5 

Color Preserving

python run_main.py --content images/01.jpg --style images/starry-night.jpg --output result_color-preserve_alg-2.jpg --color_preseving True --color_preserve_algo 2

Add Laplacian Loss

python run_main.py --content images/01.jpg --style images/starry-night.jpg --output result_lap_4.jpg --laplace True --pooling_size 4 

Arguments

Required:

• --content: Filename of the content image. Default: images/tubingen.jpg
• --style: Filename of the style image. Default: images/starry-night.jpg
• --output: Filename of the output image. Default: result.jpg

Optional:

• --model_path: Path to the pre-trained VGG model. Default: pre_trained_model
• --loss_ratio: Weight of content-loss relative to style-loss, the alpha over beta in the paper. Default: 0.5
• --content_layers: Space-separated VGG-19 layer names used for content loss computation. Default: conv4_2
• --style_layers: Space-separated VGG-19 layer names used for style loss computation. Default: relu1_1, relu2_1, relu3_1, relu4_1, relu5_1
• --content_layer_weights: Space-separated weights of each content layer to the content loss. Default: 1.0
• --style_layer_weights: Space-separated weights of each style layer to loss. Default: 0.2 0.2 0.2 0.2 0.2
• --initial_type: The initial image for optimization. (notation in the paper : x) Choices: content, style, random. Default: content
• --max_size: Maximum width or height of the input images. Default: 512
• --content_loss_norm_type: Different types of normalization for content loss. Choices: 1, 2, 3. Default: 3
• --num_iter: The number of iterations to run. Default: 1000
• --multi_style: If to use multiple style transfer. Choices: True/False Default: False
• --style2: Filename to use multiple style images. Default: images/kandinsky.jpg
• --style_ratio: The ratio between two styles. Default: 0.5
• --color_preseving: If to use color preserving. Choices: True/False Default: False
• --color_convert_type: Color spaces (YUV, YCrCb, CIE L*u*v*, CIE L*a*b*) for luminance-matching conversion to original colors. Choices: yuv, ycrcb, luv, lab. Default: yuv
• --color_preserve_algo: Color preserving algorithm. Choices: 1, 2 The Approach #2, Default: 1
• --tv: If to add the total variational loss function. Choices: True/False Default: False
• --laplace If to add the Laplacian loss. Choices: True/False Default: False
• --lap_lambda Weight for the Laplacian loss. Default: 100
• --pooling_size The filter size for pooling layer before using Laplacian operator. choices: 4, 16, 20 (choosing 20 means to use the multiple Laplacians with 4x4 and 16x16 pooling layer) Default:* 4

Acknowledgements

The implementation is based on the project:

• Tensorflow implementation 'neural-style-tf' by cysmith
• Tensorflow implementation 'tensorflow-style-transfer' by hwalsuklee