SPADE 阅读笔记
@inproceedings{RN55,
author = {Park, Taesung and Liu, Ming-Yu and Wang, Ting-Chun and Zhu, Jun-Yan},
title = {Semantic image synthesis with spatially-adaptive normalization},
booktitle = {Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition},
pages = {2337-2346},
type = {Conference Proceedings}
}
Contribution
作者提出了 spatially-adaptive normalization(SPADE),用于在 semantic image synthesis 中更好地提供语义信息,而且 SPADE 也支持 multi-modal 和 style-guided image synthesis。
Important Points
- 作者认为以前论文中的normalization不是最优的,会 wash away 语义信息,所以提出了SPADE,用semantic map来更好地指导图片合成;