Meta Networks for Neural Style Transfer
This addresses the computational bottleneck for artists and developers needing fast, lightweight style transfer, though it is an incremental improvement over existing methods.
The paper tackles the inefficiency of training separate networks for each style in neural style transfer by introducing a meta network that generates style-specific transformation networks in one feed-forward pass, achieving generation in 19ms per style and a model size of 449KB for real-time mobile use.
In this paper we propose a new method to get the specified network parameters through one time feed-forward propagation of the meta networks and explore the application to neural style transfer. Recent works on style transfer typically need to train image transformation networks for every new style, and the style is encoded in the network parameters by enormous iterations of stochastic gradient descent. To tackle these issues, we build a meta network which takes in the style image and produces a corresponding image transformations network directly. Compared with optimization-based methods for every style, our meta networks can handle an arbitrary new style within $19ms$ seconds on one modern GPU card. The fast image transformation network generated by our meta network is only 449KB, which is capable of real-time executing on a mobile device. We also investigate the manifold of the style transfer networks by operating the hidden features from meta networks. Experiments have well validated the effectiveness of our method. Code and trained models has been released https://github.com/FalongShen/styletransfer.