1603.08155

Total: 1

#1 Perceptual Losses for Real-Time Style Transfer and Super-Resolution [PDF10] [Copy] [Kimi6] [REL]

Authors: Justin Johnson ; Alexandre Alahi ; Li Fei-Fei

We consider image transformation problems, where an input image is transformed into an output image. Recent methods for such problems typically train feed-forward convolutional neural networks using a \emph{per-pixel} loss between the output and ground-truth images. Parallel work has shown that high-quality images can be generated by defining and optimizing \emph{perceptual} loss functions based on high-level features extracted from pretrained networks. We combine the benefits of both approaches, and propose the use of perceptual loss functions for training feed-forward networks for image transformation tasks. We show results on image style transfer, where a feed-forward network is trained to solve the optimization problem proposed by Gatys et al in real-time. Compared to the optimization-based method, our network gives similar qualitative results but is three orders of magnitude faster. We also experiment with single-image super-resolution, where replacing a per-pixel loss with a perceptual loss gives visually pleasing results.

Subjects: Computer Vision and Pattern Recognition ; Machine Learning

Publish: 2016-03-27 01:04:27 UTC