search

Newmu / dcgan_code

Deep Convolutional Generative Adversarial Networks
MIT License
3.42k stars 696 forks source link

ImageNet pretrained model layer dimensions #30

Open ch3njust1n opened 7 years ago

ch3njust1n commented 7 years ago

I'm studying this code and would like to know if my understanding is correct.

Generator:

layer gifn gain_ifn bias_ifn
1 (100, 128 8 4 * 4) 128 8 4 * 4 128 8 4 * 4
2 (1024, 512, 5, 5) 512 512
3 (512, 256, 5, 5) 256 256
4 (256, 128, 5, 5) 128 128
5 (128, 64, 5, 5) 64 64
6 (64, 32, 5, 5) 32 32
output (32, 3, 5, 5) ---- ----

Discriminator:

layer gifn gain_ifn bias_ifn
1 (32, 3, 5, 5) ---- ----
2 (64, 32, 5, 5) 64 64
3 (128, 64, 5, 5) 128 128
4 (256, 128, 5, 5) 256 256
5 (512, 256, 5, 5) 512 512
6 (1024, 512, 5, 5) 1024 1024
output (128 8 4 * 4, 1) ---- ----

Are these dimensions correct for the ImageNet model?

ch3njust1n commented 7 years ago

Also could you please upload the training code for ImageNet?