Fake Image Detection on Cross-Generator Generalization Trained on ProGAN (test) benchmark leaderboard · SOTA2 Research

Fake Image Detection on Cross-Generator Generalization Trained on ProGAN (test)

100Pro-GAN Detection RateFusing

Evaluation Results

Method	Links
Fusing2025.12		100	95.5	90.76	99.48	99.82	88.32	71.12	77.33	65.3	65.62	74.85	77.44	69.91	64.53	75.42	81.03
Univfd2025.12		100	99.21	98.31	97.98	99.35	99.8	82.04	82.18	85.48	82.3	84.34	84.04	69.1	90.13	94.96	89.85
NPR2025.12		100	98.5	87.8	99.8	99.9	85.5	82.4	71.6	84	84.6	74.6	85.7	85.4	80.5	81.2	86.77
VIB-Net2025.12		100	99.8	99.29	98.79	99.72	99.99	92.64	91.62	87.24	86.98	87.88	88.53	75.68	90.92	96.51	93.04
CausalCLIP2025.12		100	99.27	99.31	99.7	99.89	99.72	90.78	93.14	89.91	87.59	88.92	92.41	82.72	93.52	97.12	94.27
CNNSpot2025.12		99.99	96.4	87.5	96.94	94.24	98.28	64.42	55.89	52.86	53.25	65.14	68.1	56.6	51.15	69.49	74.02
FreDect2025.12		99.99	84.77	93.62	88.97	99.48	82.85	70.77	49.5	38.5	38.41	63.72	54.73	47.25	40.44	86.01	69.27
Lgrad2025.12		99.9	94.01	90.75	99.8	99.98	79.29	71.71	45.09	70.9	71.72	71.83	75.96	71.42	66.51	70.23	78.61
CLIPping2025.12		99.85	94.03	91.27	95.24	99.05	89.52	73.89	56.62	58.28	57.85	77.25	77.93	52.43	60.65	80	77.59