Neural Style Transfer is one of the creative applications of deep learning. Here we transfer one artist's style of painting onto another image.

Neural Style Transfer includes 3 images - Style Image, Content Image, and Generated Image. We take the style of style image, apply it to the content of the content image, and generate a generated image. The Generated Image will have content of the content image but the style of the style image.

Content: Objects and their arrangement

Style: Style, Colors, Textures

The Scream + Bouddha Stupa

Inside Neural Style Transfer

To generate a generated image, we will start with an image full of noise. We will pass the noisy image through the network and find out the activation after each convolution layer. Both content image and style images are also passed and their activation maps are also calculated. Then using those activation maps we find the content error and style error. The total error will be the linear combination of content and style error. Next, we will change the pixel of the generated image such that the total error decreases.

Images from the paper:

A is the content image. B, C, D consist of style image in a smaller size and generated image in a larger size.

Paper: https://www.cv-foundation.org/openaccess/content_cvpr_2016/papers/Gatys_Image_Style_Transfer_CVPR_2016_paper.pdf

Detailed Explanation: https://www.fritz.ai/style-transfer/

Pytorch Implementation: https://github.com/theonlyNischal/Deep-Learning-Projects-and-Paper-Implementation/tree/master/4-Neural-Style-Transfer

Try it Online: https://deepart.io/ https://demos.algorithmia.com/deep-style