CoGAN

Coupled Generative Adversarial Networks

Loss Function

• used sce loss at D/G nets.

Architecture Networks

• Same as CoGAN paper.

DIFFS	CoGAN Paper	ME
Pooling	max_pooling2d	conv2d pooling
G net	5 fc layers	2 fc + 3 deconv2d layers
conv2d filters	D[20, 50]	D[32, 64]
fc units	D[500]	D[512]

HE Initializer parameters : (factor = 1, FAN_AVG, uniform)

Tensorboard

Elapsed Time : 1h 3m 44s with GTX 1060 6GB x 1

Result

Name	Global Step 2.5k	Global Step 5k	Global Step 12.5k
Gen 1 (original)
Gen 2 (90° rotated)

To-Do