Multi-Stage Optimization for Photorealistic Neural Style Transfer

Richard R. Yang; Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2019, pp. 0-0

Abstract


This work introduces a new approach toward photorealistic style transfer. When applying current style transfer techniques on real world photographs, the generated results often contain distortions and artifacts that diminish the real-world quality of the photograph. To address these issues, we propose a two-stage optimization process that transfers style globally and regionally and applies a sharpening filter after each step. As evaluated by a user study, our method is qualitatively comparable to existing state-of-the-art methods, but successfully handles previous failure cases. Our method also quantitatively outperform previous methods as evaluated by natural scene statistic metrics.

Related Material


[pdf]
[bibtex]
@InProceedings{Yang_2019_CVPR_Workshops,
author = {Yang, Richard R.},
title = {Multi-Stage Optimization for Photorealistic Neural Style Transfer},
booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops},
month = {June},
year = {2019}
}