Tensorflow implementation of Information Maximizing GAN MNIST handwritten digit dataset.
This code requires Tensorflow. The MNIST dataset is stored in the 'MNIST_data' directory. The files will be automatically downloaded if the dataset does not exist.
If you want to use --show_progress True option, you need to install python package progress.
$ pip install progress
To train a vanilla InfoGAN with z dimension 14, categorical code dimension 10, and continuous code dimension 2, run the following command:
$ python main.py --gan_type VanillaInfoGAN --z_dim 14 --c_cat 10 --c_cont 2
To see all training options, run:
$ python main.py --help
which will print:
usage: main.py [-h] [--input_dim INPUT_DIM] [--z_dim Z_DIM] [--c_cat C_CAT]
[--c_cont C_CONT] [--d_update D_UPDATE]
[--batch_size BATCH_SIZE] [--nepoch NEPOCH] [--lr LR]
[--max_grad_norm MAX_GRAD_NORM] [--gan_type GAN_TYPE]
[--checkpoint_dir CHECKPOINT_DIR] [--image_dir IMAGE_DIR]
[--use_adam [USE_ADAM]] [--nouse_adam]
[--show_progress [SHOW_PROGRESS]] [--noshow_progress]
optional arguments:
-h, --help show this help message and exit
--input_dim INPUT_DIM
dimension of the discriminator input placeholder [784]
--z_dim Z_DIM dimension of the generator input noise variable z [14]
--c_cat C_CAT dimension of the categorical latent code [10]
--c_cont C_CONT dimension of the continuous latent code [2]
--d_update D_UPDATE update the discriminator weights [d_update] times per
generator/Q network update [2]
--batch_size BATCH_SIZE
batch size to use during training [128]
--nepoch NEPOCH number of epochs to use during training [100]
--lr LR learning rate of the optimizer to use during training
[0.001]
--max_grad_norm MAX_GRAD_NORM
clip L2-norm of gradients to this threshold [40]
--gan_type GAN_TYPE input "VanillaInfoGAN" to use Vanilla InfoGAN;
otherwise, input "InfoDCGAN" [VanillaInfoGAN]
--checkpoint_dir CHECKPOINT_DIR
checkpoint directory [./checkpoints]
--image_dir IMAGE_DIR
directory to save generated images to [./images]
--use_adam [USE_ADAM]
if True, use Adam optimizer; otherwise, use SGD [True]
--nouse_adam
--show_progress [SHOW_PROGRESS]
print progress [False]
--noshow_progress
(Optional) If you want to see a progress bar, install progress with pip:
$ pip install progress
$ python main.py --gan_type VanillaInfoGAN --z_dim 14 --c_cat 10 --c_cont 2 --show_progress True
Majority of the source code in VanillaInfoGAN.py is from: Agustinus Kristiadi / @wiseodd.
I have also written the tensorflow implementation of the InfoGAN as introduced in the original paper (See Appendix C). However, due to the lack of computational power, I was not able to test it out; therefore, the convergence of the generator and the discriminator for InfoDCGAN is not guranteed.