codes of 4 or 6 layers PBGAN with applications
example: python main.py --dataset=celeba --input_height=108 --output_height=64 --train --crop --structure=0 --option=5
--dataset: svhn celeba cifar10 Imagenet1 --train: to train a new model or read checkpoint to do visualization --structure ==0 #for original,(default) ==1 #for binarized with batch norm, ==2 #for binarized without batch norm, ==3 #for XNOR with batch norm,(not available for 6 layers) ==4 #for XNOR without batch norm(not available for 6 layers)
--option: == 5 #generating sample outputs(default) == 6 #svhn feature maps, train,split into several files, each contains 10000 vectors == 7 #svhn test feature maps, test ,split into several files, each contains 10000 vectors == 8 #cifar10 feature maps, train, split into 5 files == 9 #cifar10 feature maps, test, split into 5 files == 10 #cifar10 feature maps visualizations,"horse picture"
note: when using celeba or other dataset with picture size not 32x32, use --input_height=xxx and output_height=64 and --crop as the example if using dataset with 32x32 picture, like svhn, cifar10 Imagenet1 simply run: python main.py --dataset=cifar10 --train --structure=0 --option=5
both read *.jpg data from ./../data/xxx for example,run: ~/PBGAN/4layers/python main.py --dataset=cifar10 --train --structure=0 --option=5 it reads all the *.jpg images from ~/PBGAN/data/cifar10/