GANs with Multiple Constraints for Image Translation

Table 4

This table shows that the results of CycleGAN VS MMGANs, in which the third and the fourth column, respectively, represent a percentage of better quality of generated images by CycleGAN and MMGANs. The last two columns mean the generated images by the two models generated good images and bad images in the testing sets.

Augmented domain	Task
Pear	Apple2orange	2.2	61.4	8.6	27.8
Pear	Orange2apple	2.4	39.6	4	54
Orange	Apple2pear	0.6	31.6	0.2	67.6
Orange	Pear2apple	2.4	43.2	0	54.4
Apple	Orange2pear	2.4	17.4	4.6	75.6
Apple	Pear2orange	3	36.4	11.8	48.8
Winter	Summer2autumn	6.4	11.4	32.8	49.4
Winter	Autumn2summer	0.4	54.2	2.6	42.8
Autumn	Summer2winter	14	14.6	27.4	44
Autumn	Winter2summer	7.2	11.6	34.8	46.4
Summer	Autumn2winter	4	28.2	31.2	36.6
Summer	Winter2autumn	4.4	16.4	37.4	41.8
	Average	4.12	30.5	16.28	49.1