fast-style-transfer

Why we included this project

The idea here is simple: train a neural network to imitate the brushwork of a particular painting, then feed new photos through it. The result is fast enough to be practical. Styling a 1024×680 image takes about 100 milliseconds on a 2015 Titan X, which is why the same approach can also run across every frame of a video. The TensorFlow code draws on the influential papers by Gatys, Johnson, and Ulyanov, and it is organized clearly enough to read as a genuine study of how fast style transfer works. Researchers who want a readable implementation, and developers who need a runnable baseline to adapt for their own projects, will both find something useful here.