Weakly Supervised GAN for Image-to-Image Translation in the Wild

Table 1

Quantitative evaluations in terms of cats to dogs (A), human faces to cats (B), human faces to dogs (C), and human faces to animes (D) task from different models.

Tasks	Baselines	AMT (%)	Cf	Cs
A	CycleGAN	2% ± 1.0	0.15	1.46
	Imo-GAN	26% ± 1.0	0.76	1.27
	MUNIT	23% ± 1.0	0.71	1.28
	Ours2	24% ± 1.0	0.73	1.20
	Ours	39% ± 1.0	0.92	1.06
B	CycleGAN	2% ± 1.0	0.15	1.48
	Imo-GAN	23% ± 1.0	0.72	1.36
	MUNIT	25% ± 1.0	0.76	1.34
	Ours2	26% ± 1.0	0.78	1.29
	Ours	28% ± 1.0	0.82	1.26
C	CycleGAN	2% ± 1.0	0.11	1.62
	Imo-GAN	20% ± 1.0	0.46	1.43
	MUNIT	23% ± 1.0	0.61	1.40
	Ours2	24% ± 1.0	0.66	1.38
	Ours	26% ± 1.0	0.79	1.24
D	CycleGAN	2% ± 1.0	0.11	1.63
	Imo-GAN	20% ± 1.0	0.46	1.43
	MUNIT	23% ± 1.0	0.61	1.40
	Ours2	25% ± 1.0	0.67	1.37
	Ours	26% ± 1.0	0.78	1.26