DCGAN-Pytorch-Alter

Please pay attention to me and star if you like, I will continue to update.

The default dcgan is used to generate 64 * 64 pictures. In this code repository, we modify the network structure to generate 128 * 128 or 256 * 256 pictures.

Environments

torch==1.2.0
torchvision==0.4.0
numpy==1.15.1
matplotlib==3.0.2
ipython==7.16.1

Run Example

First, choose the net in train.py to the model you want to use.

from nets.dcgan256 import get_DCGAN   

or  

from nets.dcgan128 import get_DCGAN  

...

Second, prepare the data set and start training.

Different parameters in the dcgan64、dcgan96、dcgan128、dcgan256

if use net/dcgan64.py    
python train.py --image_size=64 --ngf=64 --ndf=64  Training effect: good       

if use net/dcgan96.py  
python train.py --image_size=96 --ngf=64 --ndf=64  Training effect: good       

if use net/dcgan128.py  
python train.py --image_size=128 --ngf=32 --ndf=32  Training effect: good    
python train.py --image_size=128 --ngf=64 --ndf=64  Training effect: bad     

if use net/dcgan128_1.py  
python train.py --image_size=128 --ngf=64 --ndf=64  Training effect: good  

if use net/dcgan256.py  
python train.py --image_size=256 --ngf=64 --ndf=64  Training effect: good