Feature-wise distance metrics in VAE-GAN

2017-fall-DL-training-program / VAE-GAN-and-VAE-GAN

An assignment to learn how to implement three differnent kinds of generative models

2 stars 0 forks source link

Feature-wise distance metrics in VAE-GAN #18

Closed jessejchuang closed 6 years ago

jessejchuang commented 6 years ago

Hi TA,

In the TF code below, it seems the author uses gamma=1e-6 but we use gamma=5. His metrics is put below as well. I'm wondering if your L2 norm definition in the handout is same as his or just use MSE without multiply with 1/2 and constant item. Could you share the formula? Thanks.

https://github.com/zhangqianhui/vae-gan-tensorflow/blob/master/vaegan.py#L84 Feature-wise distance metrics in TF = -(1/2)log(2pi) - (1/2)MSE(pred,target)

hui-po-wang commented 6 years ago

Hi @jessejchuang ,

I just used pure MSE as the metrics of feature matching without multiplying 1/2 or adding any constants.

However, you can use it if you think it's easier to implement hw.

Thanks