Style Transfer
technique that implements the artistic style of another image while maintaining the contents of a image.
Style transfer (style transfer) is a deep learning technique that implements the artistic style (color palette, brush bumps, texture) of another image by maintaining the content of a image. The main idea of this technique comes from noticeing that a pre-trained evaporous network (usually a trained network for image classification like VGG) captures different information of different layers: deep layers represent the "inner" of the image, feature correlations in shallow layers, the "fox" of the image. Algorithm optimizes a new image in terms of content image, style image in style to benzene.
This technique has gained popularity in the first research of Gatys and colleagues, then developed accelerated versions that can work in real-time. In practice, a holiday photo is used in creative applications such as Van Gogh's "Star Night" table to style or transform into a Picasso style; mobile applications such as Prisma have delivered this technology to millions of users. The style transfer has now placed the idea of the development of more general producers models (such as GAN and diffusion models) as well as the conceptual ground prepared, the content and style in the image production can be separated.
This technique has gained popularity in the first research of Gatys and colleagues, then developed accelerated versions that can work in real-time. In practice, a holiday photo is used in creative applications such as Van Gogh's "Star Night" table to style or transform into a Picasso style; mobile applications such as Prisma have delivered this technology to millions of users. The style transfer has now placed the idea of the development of more general producers models (such as GAN and diffusion models) as well as the conceptual ground prepared, the content and style in the image production can be separated.