search

elvisyjlin / AttGAN-PyTorch

AttGAN PyTorch Arbitrary Facial Attribute Editing: Only Change What You Want
MIT License
248 stars 61 forks source link

did you not use any initialization for Conv layer weight？ #20

Closed Johnson-yue closed 4 years ago

Johnson-yue commented 4 years ago

Hi， I check twice for your code 。 did you not use any initialization function for model weight ??

elvisyjlin commented 4 years ago

No, I do not use any fancy initialization function for all layers in AttGAN. I thought I tried two or three weight initializations and found that there was no much difference. If you want to know the exact initialization method for each layer, please check out the PyTorch documentation page (https://pytorch.org/docs/0.4.0/nn.html) and click the source of that layer.

For example, the convolutional layers is here: https://pytorch.org/docs/0.4.0/_modules/torch/nn/modules/conv.html#Conv1d The weight initialization is defined in reset_parameters().

Johnson-yue commented 4 years ago

ok， this is good result ！！ thank you