sanghoon/prediction_gan

CIFAR-10 experiments with different LRs

sanghoon opened this issue · 0 comments

Notes

  • This work was done only to show some sample outputs. Different random seeds can lead to totally different outcomes. Therefore, we need to investigate outputs from repeated trials to correctly compare two GAN methods.

Large learning rate (0.01)

Vanilla DCGAN

After 2 epochs After 10 epochs After 25 epochs
ep02_cifar_base_lr 0 01 ep10_cifar_base_lr 0 01 ep25_cifar_base_lr 0 01

DCGAN w/ prediction

After 2 epochs After 10 epochs After 25 epochs
ep02_cifar_pred_lr 0 01 ep10_cifar_pred_lr 0 01 ep25_cifar_pred_lr 0 01

Medium learning rate (0.0001)

Vanilla DCGAN

After 2 epochs After 10 epochs After 25 epochs
ep02_cifar_base_lr 0 0001 ep10_cifar_base_lr 0 0001 ep25_cifar_base_lr 0 0001

DCGAN w/ prediction

After 2 epochs After 10 epochs After 25 epochs
ep02_cifar_pred_lr 0 0001 ep10_cifar_pred_lr 0 0001 ep25_cifar_pred_lr 0 0001

Small learning rate (1e-5)

Vanilla DCGAN

After 2 epochs After 10 epochs After 25 epochs
ep02_cifar_base_lr 0 00001 ep10_cifar_base_lr 0 00001 ep25_cifar_base_lr 0 00001

DCGAN w/ prediction

After 2 epochs After 10 epochs After 25 epochs
ep02_cifar_pred_lr 0 00001 ep10_cifar_pred_lr 0 00001 ep25_cifar_pred_lr 0 00001