Github项目推荐 | GAN评估指标的Tensorflow简单实现

GAN_Metrics-Tensorflow

简单的GAN评估指标的Tensorflow实现

• Inception score

• Frechet-Inception distance

• Kernel-Inception distance

Github项目地址：

https://github.com/taki0112/GAN_Metrics-Tensorflow

摘要

Name - 名称	Description - 描述	Performance   score - 评分
Inception score	KL-Divergence between conditional and marginal label   distributions over generated data KL-散度上条件和边缘标签分布在生成数据上的差异。	Higher is better 评分越高越好
Frechet-Inception distance	Wasserstein-2 distance between multi-variate   Gaussians fitted to data embedded into a feature space 多变量高斯函数之间的Wasserstein-2距离，适用于嵌入到特征空间中的数据。	Lower is better. 评分越低越好
Kernel-Inception distance	Measures the dissimilarity between two probability   distributions Pr and Pg using samples drawn   independently from each distribution 使用独立于每个分布绘制的样本测量两个概率分布Pr和Pg之间的差异。	Lower is better. 评分越低越好

使用

├── real_source     ├── aaa.png    ├── bbb.jpg├── real_target     ├── ccc.png    ├── ddd.jpg├── fake     ├── ccc_fake.png    ├── ddd_fake.jpg├── main.py├── inception_score.py└── frechet_kernel_Inception_distance.py

> pyth main.py

参考

• Pros and Cons of GAN Evaluation Measures - GAN评价方法的优缺点

• Inception score

• Frechet-Inception score

作者

Junho Kim

点击阅读原文，查看更多内容