Baselines

We provide some sample baseline scores for our models. For the latest versions of the score and more details on the exact computation methods, please visit the official github webpage.

Metric	Dataset	Resolution	Models
			SNGAN	cGAN-PD	SSGAN	InfoMax-GAN
FID	LSUN-Bedroom	128 x 128	25.87 ± 0.03	N/A	12.02 ± 0.07	12.13 ± 0.16
	CelebA	128 x 128	12.93 ± 0.04	N/A	15.18 ± 0.10	9.50 ± 0.04
	CelebA	64 x 64	5.71 ± 0.02	N/A	6.03 ± 0.04	5.71 ± 0.06
	STL-10	48 x 48	39.56 ± 0.10	N/A	37.06 ± 0.19	35.52 ± 0.10
	CIFAR-10	32 x 32	16.77 ± 0.04	10.84 ± 0.03	14.65 ± 0.04	15.12 ± 0.10
	CIFAR-100	32 x 32	22.61 ± 0.06	14.16 ± 0.01	22.18 ± 0.10	18.94 ± 0.13
	ImageNet	32 x 32	23.04 ± 0.06	21.17 ± 0.05	21.79 ± 0.09	20.68 ± 0.02
KID	LSUN-Bedroom	128 x 128	0.0141 ± 0.0001	N/A	0.0077 ± 0.0001	0.0080 ± 0.0001
	CelebA	128 x 128	0.0076 ± 0.0001	N/A	0.0101 ± 0.0001	0.0063 ± 0.0001
	CelebA	64 x 64	0.0033 ± 0.0001	N/A	0.0036 ± 0.0001	0.0033 ± 0.0001
	STL-10	48 x 48	0.0369 ± 0.0002	N/A	0.0332 ± 0.0004	0.0326 ± 0.0002
	CIFAR-10	32 x 32	0.0125 ± 0.0001	0.0070 ± 0.0001	0.0101 ± 0.0002	0.0112 ± 0.0001
	CIFAR-100	32 x 32	0.0156 ± 0.0003	0.0085 ± 0.0002	0.0161 ± 0.0002	0.0135 ± 0.0004
	ImageNet	32 x 32	0.0157 ± 0.0002	0.0145 ± 0.0002	0.0152 ± 0.0002	0.0149 ± 0.0001
IS	LSUN-Bedroom	128 x 128	2.30 ± 0.01	N/A	2.12 ± 0.01	2.22 ± 0.01
	CelebA	128 x 128	2.72 ± 0.01	N/A	2.63 ± 0.01	2.84 ± 0.01
	CelebA	64 x 64	2.68 ± 0.01	N/A	2.67 ± 0.01	2.68 ± 0.01
	STL-10	48 x 48	8.04 ± 0.07	N/A	8.25 ± 0.06	8.54 ± 0.12
	CIFAR-10	32 x 32	7.97 ± 0.06	8.25 ± 0.13	8.17 ± 0.06	8.08± 0.08
	CIFAR-100	32 x 32	7.57 ± 0.11	8.92 ± 0.07	7.56 ± 0.07	7.86 ± 0.10
	ImageNet	32 x 32	8.97 ± 0.12	9.08 ± 0.17	9.11 ± 0.12	9.04 ± 0.10