hanyoseob / pytorch-sngan

Spectral Normalization for Generative Adversarial Networks

SN-GAN

Title

Spectral Normalization for Generative Adversarial Networks

Abstract

One of the challenges in the study of generative adversarial networks is the instability of its training. In this paper, we propose a novel weight normalization technique called spectral normalization to stabilize the training of the discriminator. Our new normalization technique is computationally light and easy to incorporate into existing implementations. We tested the efficacy of spectral normalization on CIFAR10, STL-10, and ILSVRC2012 dataset, and we experimentally confirmed that spectrally normalized GANs (SN-GANs) is capable of generating images of better or equal quality relative to the previous training stabilization techniques.

Train

$ python main.py --mode train \
                 --scope [scope name] \
                 --name_data [data name] \
                 --dir_data [data directory] \
                 --dir_log [log directory] \
                 --dir_checkpoint [checkpoint directory]
                 --gpu_ids [gpu id; '-1': no gpu, '0, 1, ..., N-1': gpus]

---

$ python main.py --mode train \
                 --scope sngan \
                 --name_data celeba \
                 --dir_data ./datasets \
                 --dir_log ./log \
                 --dir_checkpoint ./checkpoint
                 --gpu_ids 0

• Set [scope name] uniquely.
• Hyperparameters were written to arg.txt under the [log directory].
• To understand hierarchy of directories based on their arguments, see directories structure below.

Test

$ python main.py --mode test \
                 --scope [scope name] \
                 --name_data [data name] \
                 --dir_data [data directory] \
                 --dir_log [log directory] \
                 --dir_checkpoint [checkpoint directory] \
                 --dir_result [result directory]
                 --gpu_ids [gpu id; '-1': no gpu, '0, 1, ..., N-1': gpus]

---

$ python main.py --mode test \
                 --scope sngan \
                 --name_data celeba \
                 --dir_data ./datasets \
                 --dir_log ./log \
                 --dir_checkpoint ./checkpoints \
                 --dir_result ./results
                 --gpu_ids 0

• To test using trained network, set [scope name] defined in the train phase.
• Generated images are saved in the images subfolder along with [result directory] folder.
• index.html is also generated to display the generated images.

Tensorboard

$ tensorboard --logdir [log directory]/[scope name]/[data name] \
              --port [(optional) 4 digit port number]

---

$ tensorboard --logdir ./log/sngan/celeba \
              --port 6006

After the above comment executes, go http://localhost:6006

• You can change [(optional) 4 digit port number].
• Default 4 digit port number is 6006.

Results

• The results were generated by a network trained with celeba dataset during 10 epochs.
• After the Test phase runs, execute display_result.py to display the figure.

Directories structure

pytorch-SNGAN
+---[dir_checkpoint]
|   \---[scope]
|       \---[name_data]
|           +---model_epoch00000.pth
|           |   ...
|           \---model_epoch12345.pth
+---[dir_data]
|   \---[name_data]
|       +---000000.png
|       |   ...
|       \---12345.png
+---[dir_log]
|   \---[scope]
|       \---[name_data]
|           +---arg.txt
|           \---events.out.tfevents
\---[dir_result]
    \---[scope]
        \---[name_data]
            +---images
            |   +---00000-output.png
            |   |   ...
            |   +---12345-output.png
            \---index.html

---

pytorch-SNGAN
+---checkpoints
|   \---sngan
|       \---celeba
|           +---model_epoch00001.pth
|           |   ...
|           \---model_epoch0010.pth
+---datasets
|   \---celeba
|       +---000001.jpg
|       |   ...
|       \---202599.jpg
+---log
|   \---sngan
|       \---celeba
|           +---arg.txt
|           \---events.out.tfevents
\---results
    \---sngan
        \---celeba
            +---images
            |   +---0000-output.png
            |   |   ...
            |   +---0127-output.png
            \---index.html

• Above directory is created by setting arguments when main.py is executed.