Forensic Detection on FakeClass FakeBench (qualitative setting) benchmark leaderboard · SOTA2 Research

Forensic Detection on FakeClass FakeBench (qualitative setting)

99.5Detection Rate (Fake)FakeScope

Evaluation Results

Method	Links
FakeScope2025.03		99.5	99.87	99.73	99.63	100	100	99.67	100	100	100	98.67	100	97.67	99	99.68
FakeScope2025.03		98.67	97.93	98.67	97.93	99.67	98.67	99	99.67	96.67	97	99.33	100	98	98.67	98.3
Human (Best)2025.03		97	93	—	—	100	100	100	100	100	100	100	100	100	93	87.5
Human (Overall)2025.03		76.91	72.12	—	—	95.47	58.82	87.07	95.9	63.96	82.91	79.41	77.66	70.69	45.59	74.51
InstructBLIP2025.03		67.8	47.67	65.23	50.23	80.67	65.33	66	71.33	68	70.33	60.33	61	70.67	64.33	57.73
GPT-4V2025.03		59.87	96.2	76.33	79.73	95	66	66.67	81.33	36.67	49	45.67	44.33	61	53	78.03
mPLUG-Owl22025.03		49.6	93.97	71.87	71.7	56.33	10	72	75.33	19.33	32	56	55	64	56	71.78
Human (Worst)2025.03		43	13	—	—	50	0	46.15	77.78	33.33	16.67	41.67	7.14	14.29	0	55
Random guess2025.03		42.27	58.03	51.8	48.5	40.33	36	51.67	33.67	38	44.67	32.33	47.67	37.67	60.67	50.15
LLaVA-v1.52025.03		38	77.4	55.77	59.63	44.33	17.33	45.33	55.33	23	28.33	39.67	0	0	41	57.7
GeminiPro2025.03		35.83	99.27	61.27	73.83	45.33	15.67	47.33	52.67	19.33	28	35.33	35.67	41	38	67.5
InternLM-XC.2-vl2025.03		32.17	92.33	60.4	64.1	54	12.67	43.33	49	19	20.33	29.33	29.67	34.67	29.67	62.25
Qwen-VL2025.03		28.57	84.27	51.17	61.17	45.67	5.67	35	37.67	8.33	14.33	31.33	21	46.67	40	56.42
IDEFICS-Instruct2025.03		24.97	31.97	7	49.93	19.67	27.33	24.33	28.33	27.67	24	23.67	23.67	24	27	28.47
Claude3 Sonnet2025.03		12	98.23	55.97	54.27	7	0	20.33	13.67	0.67	2.67	15.33	9.33	30.33	20.67	55.12