eduzhai > Applied Sciences > Engineering >

Real-time Universal Style Transfer on High-resolution Images via Zero-channel Pruning

  • Save

... pages left unread,continue reading

Document pages: 12 pages

Abstract: Extracting effective deep features to represent content and style informationis the key to universal style transfer. Most existing algorithms use VGG19 asthe feature extractor, which incurs a high computational cost and impedesreal-time style transfer on high-resolution images. In this work, we propose alightweight alternative architecture - ArtNet, which is based on GoogLeNet, andlater pruned by a novel channel pruning method named Zero-channel Pruningspecially designed for style transfer approaches. Besides, we propose atheoretically sound sandwich swap transform (S2) module to transfer deepfeatures, which can create a pleasing holistic appearance and good localtextures with an improved content preservation ability. By using ArtNet and S2,our method is 2.3 to 107.4 times faster than state-of-the-art approaches. Thecomprehensive experiments demonstrate that ArtNet can achieve universal,real-time, and high-quality style transfer on high-resolution imagessimultaneously, (68.03 FPS on 512 times 512 images).

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×