Road Scene on RoadSceneBench benchmark leaderboard · SOTA2 Research

Road Scene on RoadSceneBench

82.8PrecisionMapVLM

Evaluation Results

Method	Links
MapVLM2025.11		82.8	82.36
MapVLM2025.11		81.62	82.65
MapVLM2025.11		75.78	72.17
InternVL3-8B2025.11		73.07	69.32
Qwen3-VL-8B2025.11		72.79	70.15
MapVLM2025.11		72.14	67.25
Qwen2.5-VL-3B2025.11		71.78	64.79
ERNIE-4.5-VL-28B-A3B2025.11		69.98	60.88
Qwen2.5-VL-7B2025.11		69.79	68.51
InternVL3-78B2025.11		69.02	69.92
Qwen2.5-VL-32B2025.11		68.07	63.02
Qwen2.5-VL-72B2025.11		67.85	67.4
GPT-4o2025.11		67.64	64.23
Gemini-2.5-pro2025.11		65.91	64.2
Qwen3-VL-32B2025.11		65.53	64.16
Claude-3.7V Sonnet2025.11		61.58	59.25
Gemini-2.5-pro2025.11		60.61	52.7
Qwen3-VL-8B2025.11		57.34	43.82
InternVL3-78B2025.11		55.53	45.27
Qwen3-VL-32B2025.11		53.54	43.69
InternVL3-1B2025.11		52.67	34.51
Qwen3-VL-2B2025.11		52.59	50.19
GPT-4o2025.11		51.79	42.08
Qwen2.5-VL-72B2025.11		50.57	40.18
Qwen2.5-VL-32B2025.11		50.4	37.5
ERNIE-4.5-VL-28B-A3B2025.11		50.11	33.93
Claude-3.7V Sonnet2025.11		47.3	41.38
LLaVA-Onevision / Next-7B2025.11		42.8	41.63
Qwen2.5-VL-7B2025.11		42.67	33.37
Qwen2.5-VL-3B2025.11		40.61	33.86
InternVL3-8B2025.11		39.24	35.51
LLaVA-Onevision / Next-7B2025.11		38.65	38.54
Qwen3-VL-2B2025.11		36.15	38.52
DeepSeek-VL22025.11		28.5	19.26
InternVL3-1B2025.11		26.11	23.19
DeepSeek-VL22025.11		16.63	14.57