Exploring Fast Artistic Style Merging on Heterogenous Architecture
Peian Wang, Jinyun Gu, Zichen Xu · 2017
We propose a novel approach to realize fast style merging, which provides unique understanding for art pieces from eastern and western world onto a cache-memory hier- archy in a CPU-mGPU architecture. We successfully generate style merged images with a stream of source images and produce 34% faster than the original framework, which converge iterations in a limited number(e.g., 300).