Reconstruction of Generative Adversarial Networks in Cross Modal Image Generation with Canonical Polyadic Decomposition

Table 3

Comparison for different learning schemes of the reconstructed model.

Learning scheme	LR	FID	IS
Fixed learning rate	0.0002	99.54
Fixed learning rate	0.00017	120.86
Fixed learning rate	0.00014	106.45
CosineAnnealingWarmRestarts	0.00025~0.00012	96.95
CosineAnnealingWarmRestarts	0.0002~0.0001	119.12
CosineAnnealingWarmRestarts	0.0002~0.00014	91.08
CosineAnnealingWarmRestarts	0.0002~0.00017	92.88
CosineAnnealingWarmRestarts	0.0017~0.00014	98.09
MultistepLR	0.0002	131.57
MultistepLR	0.00017	65.05
MultistepLR	0.00014	84.27