search

LeeDoYup / AnoGAN-tf

Unofficial Tensorflow Implementation of AnoGAN (Anomaly GAN)
MIT License
273 stars 90 forks source link

questions about the input_height and input_width #26

Open lzzlxxlsz opened 5 years ago

lzzlxxlsz commented 5 years ago

last time,I use my data with the size of 15441284,and I resized it with 178218 as the celeba dataset,but the result is not good.The generated picture is not clear,so I want to ask how to choose the size of data

LeeDoYup commented 5 years ago

When you train the GAN with your own data, hyperparameter tuning is needed. (because GAN is very sensitive to dataset and hyperparameters)

lzzlxxlsz commented 5 years ago

Thanks,that is to say,I don't need to resize my dataset,but when I set the input_height and width with a big size,it dosen't work.Another question is that the picture would lost some information after crop in the training test ,if I want the GAN learn the whole picture ,do I need crop?

lzzlxxlsz commented 5 years ago

Sorry,I wonder the hyperparameter, does it contains the learning rate,beta1 and other? Thanks for you if you could explain more detail,and I am feel confused about what aspects should I adjust to suit my dataset