Multi-entity reasoning grounding on ME-RSRG (test) benchmark leaderboard · SOTA2 Research

Multi-entity reasoning grounding on ME-RSRG (test)

44.4Subject Accuracy (IoU=0.5)Qwen2.5-VL-7B

Evaluation Results

Method	Links
Qwen2.5-VL-7B2026.03		44.4	48.9	46.72
InternVL3.5-8B2026.03		39.08	34.89	36.92
Qwen2.5-VL-3B2026.03		38.63	42.08	40.41
Qwen3-VL-4B2026.03		38.52	36.83	37.65
InternVL3.5-4B2026.03		38.02	35.31	36.62
Qwen3-VL-8B2026.03		37.96	36.94	37.43
Qwen3-VL-4B2026.03		36.39	35.89	36.13
InternVL3.5-8B2026.03		35.33	32.32	33.78
InternVL3.5-4B2026.03		34.49	31.22	32.8
Qwen3-VL-8B2026.03		33.76	34.89	34.34
Qwen2.5-VL-7B2026.03		31.75	34.84	33.34
Qwen2-VL-2B2026.03		27.55	22.93	25.16
Qwen2.5-VL-3B2026.03		27.04	30.43	28.79
Qwen3-VL-8B2026.03		26.82	16.95	21.72
Qwen2-VL-2B2026.03		23.18	23.08	23.13
Qwen3-VL-4B2026.03		20.95	13.8	17.25
LLaVA-OneVision2026.03		18.14	14.59	16.31
Qwen2.5-VL-7B2026.03		11.31	9.44	10.35
LLaVA-OneVision2026.03		9.8	6.24	7.96
InternVL2.5-4B2026.03		9.18	9.97	9.59
LLaVA-1.5-7b2026.03		7.95	7.97	7.96
Qwen2.5-VL-3B2026.03		7.45	10.81	9.18
InternVL2.5-4B2026.03		6.94	5.55	6.18
LLaVA-1.5-7b2026.03		3.42	3.25	3.33
InternVL3.5-8B2026.03		2.35	2.31	2.33
InternVL2.5-4B2026.03		1.68	1.57	1.63
Qwen2-VL-2B2026.03		1.4	1.15	1.27
LLaVA-1.5-7b2026.03		1.29	1.21	1.25
InternVL3.5-4B2026.03		0.5	0.52	0.51
LLaVA-OneVision2026.03		0	0.05	0.03