gsurma / style_transfer

CNN image style transfer 🎨.

Style Transfer

Style Transfer is a process in which we strive to modify the style of an image while preserving its content. Given an input image and a style image, we can compute an output image with the original content but a new style.

Kaggle kernel

Check out corresponding Medium article:

Style Transfer - Styling Images with Convolutional Neural Networks

How does it work?

1. We take input image and style images and resize them to equal shapes.
2. We load a pre-trained CNN (VGG16).
3. Knowing that we can distinguish layers that are responsible for the style (basic shapes, colors etc.) and the ones responsible for the content (image-specific features), we can separate the layers to independently work on the content and style.
4. Then we set our task as an optimization problem where we are going to minimize:
   • content loss (distance between the input and output images - we strive to preserve the content)
   • style loss (distance between the style and output images - we strive to apply a new style)
   • total variation loss (regularization - spatial smoothness to denoise the output image)
5. Finally, we set our gradients and optimize with the L-BFGS algorithm.

Results

Input

Style

Output

1 iteration

2 iterations

5 iterations

10 iterations

15 iterations

Other examples

Author

Greg (Grzegorz) Surma

PORTFOLIO

GITHUB

BLOG

style_transfer's Issues

save model

How do i save that model, I don't want to train it every time with different input image and same style image

def content_loss

Variable += value not supported. Use variable.assign_add(value) to modify the variable value and variable = variable + value to get a new Tensor object.

How to use it?

I wanna use it for generating images with styles