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

Correspondence-Pointing on CrossPoint-Bench

95.87Object Accuracy

Human

Updated 1mo ago

Evaluation Results

Method	Links
Human 2025.12		95.87	91.18	93.63
CroPond-3B 2025.12		84.3	62.44	73.87
CroPond-7B 2025.12		83.88	71.49	77.97
Qwen3-VL-235B-A22B-Thinking 2025.12		50.83	26.7	39.31
Qwen3-VL-30B-A3B-Thinking 2025.12		32.23	19	25.92
Qwen2.5-VL-72B-Instruct 2025.12		23.14	10.86	17.28
Gemini-2.5-Pro 2025.12		20.25	12.22	16.41
Qwen2.5-VL-32B-Instruct 2025.12		19.83	5.88	13.17
o3 2025.12		10.74	9.5	10.15
Qwen3-VL-235B-A22B-Instruct 2025.12		7.85	4.98	6.48
Claude-Sonnet-4 2025.12		7.02	3.17	5.18
Qwen3-VL-30B-A3B-Instruct 2025.12		4.55	3.62	4.1
Molmo-7B-D 2025.12		4.13	1.81	3.02
Qwen2.5-VL-3B-Instruct 2025.12		2.89	2.26	2.59
RoboRefer-8B-SFT 2025.12		1.65	2.26	1.94
RoboPoint-v1-Vicuna-13B 2025.12		0.83	1.36	1.08
RoboBrain2.0-7B 2025.12		0.41	1.36	0.86
RoboBrain2.0-7B 2025.12		0.41	1.36	0.86
Qwen2.5-VL-7B-Instruct 2025.12		0	0	0