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

Fine-grained Grounding on CrossPoint-Bench

98.49Object Accuracy

Human

Updated 1mo ago

Evaluation Results

Method
Human 2025.12	98.49	96.63	97.44
CroPond-7B 2025.12	93.94	87.64	87.18
CroPond-3B 2025.12	90.91	73.03	76.92
Human 2025.12	83.57	76.37	79.51
Qwen3-VL-235B-A22B-Instruct 2025.12	79.17	54.63	62.18
CroPond-3B 2025.12	78.57	43.96	59.01
Qwen3-VL-235B-A22B-Thinking 2025.12	77.08	61.11	66.03
Gemini-2.5-Pro 2025.12	75	53.7	60.26
Qwen3-VL-235B-A22B-Thinking 2025.12	74.29	58.24	65.22
Qwen3-VL-30B-A3B-Thinking 2025.12	72.86	38.46	53.42
RoboBrain2.0-7B 2025.12	72.86	40.66	54.66
CroPond-7B 2025.12	71.43	51.65	60.25
o3 2025.12	70.83	60.19	63.46
Qwen3-VL-235B-A22B-Instruct 2025.12	70	52.75	60.25
Qwen2.5-VL-3B-Instruct 2025.12	67.14	41.76	52.8
Qwen2.5-VL-7B-Instruct 2025.12	65.71	35.16	48.45
Qwen2.5-VL-72B-Instruct 2025.12	65.71	46.15	54.66
RoboBrain2.0-7B 2025.12	64.29	41.76	51.55
RoboRefer-8B-SFT 2025.12	62.86	32.97	45.96
Molmo-7B-D 2025.12	62.86	36.26	47.83
Qwen3-VL-30B-A3B-Instruct 2025.12	55.71	26.37	39.13
Qwen2.5-VL-7B-Instruct 2025.12	54.55	37.08	41.03
Gemini-2.5-Pro 2025.12	50	19.78	32.92
Qwen3-VL-30B-A3B-Instruct 2025.12	50	41.67	44.23
Qwen2.5-VL-32B-Instruct 2025.12	47.92	40.74	42.95
Qwen2.5-VL-72B-Instruct 2025.12	45.83	51.85	50
Qwen3-VL-30B-A3B-Thinking 2025.12	45.83	45.37	45.51
Claude-Sonnet-4 2025.12	41.67	37.04	38.46
RoboBrain2.0-7B 2025.12	39.58	33.33	35.26
RoboPoint-v1-Vicuna-13B 2025.12	37.5	25.93	29.49
Molmo-7B-D 2025.12	37.5	35.19	35.9
Qwen2.5-VL-3B-Instruct 2025.12	33.33	33.71	33.97
Qwen2.5-VL-32B-Instruct 2025.12	32.86	14.29	22.36
RoboBrain2.0-7B 2025.12	29.17	41.67	37.82
RoboRefer-8B-SFT 2025.12	22.92	31.48	28.85
Claude-Sonnet-4 2025.12	17.14	7.69	11.8
o3 2025.12	11.43	6.59	8.7
RoboPoint-v1-Vicuna-13B 2025.12	7.14	7.69	7.45