search

SSRheart / TSFlow

MIT License
12 stars 0 forks source link

Where does expert_c_latent.npy come from？ #2

Closed freedomsb closed 2 years ago

freedomsb commented 2 years ago

I appreciate your work, and would like to ask a question is： How the expert_c_latent.npy used in the testing phase is calculated? I looked at the training code and didn't find the code to save expert_c_latent.npy

freedomsb commented 2 years ago

According to my understanding, the input of this z should be a random Gaussian distribution?

SSRheart commented 2 years ago

The input of this z is sampled from random Gaussian distribution. Thus, different z can generate different retouching styles. In the testing phase, we first obtain the latent z by reversing TSFlow and the tone style representation in the training set. The expert_c_latent can be calculated by formula (13) in the paper.

freedomsb commented 2 years ago

Woooow，thank you for such a quick reply. I'll check it out. By the way, if it's convenient，which function in this code is used to calculate this latent.npy?

freedomsb commented 2 years ago

I seem to understand, thank you for your answer, this Z is the average of the trained model running the training set again, it should be easy to implement~