This repository implements the ECCV 2016 paper Perceptual Losses for Real-Time Style Transfer and Super-Resolution by Johnson et al., using Torch/Lua. It trains feedforward convolutional networks to apply artistic styles to images hundreds of times faster than optimization-based approaches. The project also includes an implementation of instance normalization to improve stylization quality, and provides pre-trained models, webcam demo, and training code for new style transfer models.