ZZUTK / Face-Aging-CAAE

Age Progression/Regression by Conditional Adversarial Autoencoder
636 stars 233 forks source link

test or train the model use specified ethnic face #6

Open tianyu06030020 opened 7 years ago

tianyu06030020 commented 7 years ago

thx for your shared work. i have a question: I trained the model follwed your instruction, when i test the asian faces, the result looks like europens or americans . is there any solutions or chaned some params to generate asian faces, thx?

susanqq commented 7 years ago

Which dataset did you use to train your model? What is the percentage of Asian faces? You can increase the Asian face percentage to solve this issue.

tianyu06030020 commented 7 years ago

This is a method what you said. I wanna konw if i don not change the utkface dataset and only modify your sample code whether can meet my demands. because your sample output has male and female.png used self.gender parameter, like this , I wanna add self.races and hope the output looks like the specified ethnic . Is it workable ?? Thx.

okmmsky888 commented 7 years ago

@tianyu06030020 You can not modify the way the code, or the need for data can be. When you test, the image preprocessing is what to deal with, do the face alignment?

ZZUTK commented 7 years ago

@tianyu06030020 I think it worth to try. Empirically, the generated images present aging and gender effects corresponding to the age and gender labels, respectively. It should also show race difference when you control the race label.