Share your thoughts, 1 month free Claude Pro on usSee more

Image Deepfake Detection on DFBench AI-Edited (test)

57.32Object Enhance Accuracy

Qwen3-VL

Updated 1mo ago

Evaluation Results

Method	Links
Qwen3-VL 2026.03		57.32	56	50.01	59	52.21	47	60	61	90.01	70	61.91	58
Qwen3-VL 2026.03		55.97	51	48.82	57	51.89	47	58.01	57	88.81	62	60.69	54
Qwen3-VL 2026.03		55.77	50	46.32	55	50.14	41	57.98	53	87.78	61	59.59	52
Qwen3-VL 2026.03		53.43	49	45.54	54	49.09	40	57.54	51	87.66	56	58.65	50
Owl2.1 2026.03		36.32	32	23.45	21	27.32	31	30.02	25	82.32	53	39.89	32
Gemma-3 2026.03		21.34	21	23.32	21	15.65	12	19.92	19	85.54	61	33.15	27
Janus-Pro 2026.03		16.21	17	23.12	21	23.32	32	28.12	21	87.77	58	35.71	30
MiniCPM-V 2026.03		13.32	23	19.03	29	28.32	34	33.32	28	81.21	52	35.04	33
Kimi-VL 2026.03		11.21	19	13.21	23	21.21	34	19.32	16	83.32	64	29.65	31