DRAGAN

On Convergence and Stability of Generative Adversarial Networks

Loss Function

• used sce loss at D/G nets.

Architecture Networks

• Same as DRAGAN paper.

DIFFS	DRAGAN Paper	ME
Weight initializer	xavier initializer	HE initializer
z noise	100	128
image scale	-1,1	0,1

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

Tensorboard

Elapsed Time : 3213s with GTX 1060 6GB x 1

Result

Name	Global Step 50k	Global Step 100k	Global Step 200k
DRAGAN

To-Do