Image screening on SPR 128K 1.0 (test)

71.15Overall ScoreHCM-GRPO-3B

Evaluation results

Method
HCM-GRPO-3B2025.11	71.15	74.25	69.51	63.56	81.37
HCM-GRPO-2B2025.11	64.74	68.66	65.85	55.08	74.51
DPA-GRPO-2B2025.05	59.83	62.31	58.54	55.93	69.61
HCM-GRPO-1B2025.11	57.48	58.21	60.98	62.71	57.84
DPA-GRPO-1B2025.05	55.56	58.96	56.1	61.86	67.65
GPT5.22025.05	50	46.24	51.22	53.39	49.5
GPT5.22025.11	50	46.24	51.22	53.39	49.5
Gemini3-Pro2025.05	44.02	33.58	39.02	37.29	33.33
Gemini3-Pro2025.11	44.02	33.58	39.02	37.29	33.33
Qwen3.5-397B-A17B2025.05	43.8	35.82	37.8	33.05	39.22
Qwen3.5-397B-A17B2025.11	43.8	35.82	37.8	33.05	39.22
Qwen3.5-Plus2025.05	43.59	34.33	40.24	33.9	37.25
Qwen3.5-Plus2025.11	43.59	34.33	40.24	33.9	37.25
GPT5.22025.05	42.95	32.09	45.12	33.9	44.55
GPT5.22025.11	42.95	32.09	45.12	33.9	44.55
ClaudeSonnet4.52025.05	40.17	29.59	35.8	28.45	43.14
ClaudeSonnet4.52025.11	40.17	29.59	35.8	28.45	43.14
Qwen3.5-397B-A17B2025.05	39.96	31.34	35.37	31.36	32.35
Qwen3.5-397B-A17B2025.11	39.96	31.34	35.37	31.36	32.35
Qwen3.5-Plus2025.05	39.74	30.6	39.02	27.97	34.31
Qwen2.5-VL-72B-Instruct2025.05	39.74	29.85	40.24	32.2	30.39
Qwen3.5-Plus2025.11	39.74	30.6	39.02	27.97	34.31
Qwen2.5-VL-72B-Instruct2025.11	39.74	29.85	40.24	32.2	30.39
Qwen-VL-Max2025.05	38.25	29.1	37.8	25.42	28.43
Qwen-VL-Max2025.11	38.25	29.1	37.8	25.42	28.43
ClaudeSonnet4.52025.05	37.82	26.59	34.15	23.73	39.22
ClaudeSonnet4.52025.11	37.82	26.59	34.15	23.73	39.22
Gemini3-Pro2025.05	37.39	27.61	34.15	26.27	25.49
Gemini3-Pro2025.11	37.39	27.61	34.15	26.27	25.49
InternVL3-78B2025.05	36.75	26.49	31.71	29.66	28.43
InternVL3-38B2025.05	36.75	26.49	32.93	30.51	27.45
InternVL3-38B2025.11	36.75	26.49	32.93	30.51	27.45
Qwen3-VL-30B-A3B-Instruct2025.05	36.54	27.99	36.59	22.88	30.39
Qwen3-VL-30B-A3B-Instruct2025.11	36.54	27.99	36.59	22.88	30.39
Qwen2.5-VL-72B-Instruct2025.05	35.47	25	35.37	25.42	25.49
Qwen2.5-VL-72B-Instruct2025.11	35.47	25	35.37	25.42	25.49
GLM-4.5V2025.05	34.83	23.88	29.27	27.97	26.47
InternVL3-78B2025.05	34.83	23.88	31.71	24.58	27.45
GLM-4.5V2025.11	34.83	23.88	29.27	27.97	26.47
GPT4o2025.05	34.62	23.51	29.27	24.58	29.41
Qwen3-VL-8B-Instruct2025.05	34.62	23.88	32.93	24.58	24.51
GPT4o2025.11	34.62	23.51	29.27	24.58	29.41
Qwen3-VL-8B-Instruct2025.11	34.62	23.88	32.93	24.58	24.51
Qwen3-VL-8B-Instruct2025.05	34.4	21.64	29.27	25.42	24.51
Qwen3-VL-8B-Instruct2025.11	34.4	21.64	29.27	25.42	24.51
Qwen3-VL-30B-A3B-Instruct2025.05	33.97	23.51	25.61	25.42	26.47
Qwen3-VL-30B-A3B-Instruct2025.11	33.97	23.51	25.61	25.42	26.47
InternVL3.5-38B2025.05	33.33	20.9	26.83	25.42	25.49
InternVL3.5-38B2025.11	33.33	20.9	26.83	25.42	25.49
GLM-4.5V2025.05	32.69	20.9	30.49	22.03	23.53
GLM-4.5V2025.11	32.69	20.9	30.49	22.03	23.53
InternVL3.5-30B-A3B2025.05	32.48	20.9	26.83	22.03	26.47
InternVL3-38B2025.05	32.48	21.27	24.39	24.58	26.47
InternVL3.5-30B-A3B2025.11	32.48	20.9	26.83	22.03	26.47
InternVL3-38B2025.11	32.48	21.27	24.39	24.58	26.47
Qwen-VL-Max2025.05	32.05	21.64	26.83	25.42	15.67
Qwen-VL-Max2025.11	32.05	21.64	26.83	25.42	15.67
InternVL3.5-30B-A3B2025.05	27.56	18.28	21.95	22.03	25.49
InternVL3.5-30B-A3B2025.11	27.56	18.28	21.95	22.03	25.49
InternVL3.5-38B2025.05	20.73	13.06	21.95	13.56	15.69
InternVL3.5-38B2025.11	20.73	13.06	21.95	13.56	15.69
GPT4o2025.05	18.59	18.66	13.41	22.88	20.59
GPT4o2025.11	18.59	18.66	13.41	22.88	20.59