Image Generation on FFHQ 256x256 (FID, Precision, Recall, Density, Coverage) benchmark leaderboard · SOTA2 Research

Image Generation on FFHQ 256x256 (FID, Precision, Recall, Density, Coverage)

2.19FIDSTYLEGAN-XL

Evaluation Results

Method
STYLEGAN-XL2023.05	2.19	79.91	38.79	0.86	0.73
PR-divergence fine-tuned BigGAN2023.05	35.24	78.41	9.66	0.89	0.6
PR-divergence fine-tuned BigGAN2023.05	35.66	78.7	9.45	0.88	0.6
PR-divergence fine-tuned BigGAN2023.05	35.91	78.95	8.32	0.9	0.57
PR-divergence fine-tuned BigGAN2023.05	36.33	81.1	8.69	1.05	0.64
PR-divergence fine-tuned BigGAN2023.05	38.16	84.31	8.52	1.15	0.63
BASELINE BIGGAN2023.05	41.41	65.57	10.17	0.52	0.47
HARD Truncation2023.05	43.32	68.84	8.66	0.58	0.47
HARD Truncation2023.05	56.19	76.44	4.76	0.75	0.44
SOFT Truncation2023.05	56.43	76.59	4.87	0.7	0.41
HARD Truncation2023.05	71.32	80.99	4.84	0.84	0.36
SOFT Truncation2023.05	82.05	84.48	1.58	0.89	0.32