Volumetric Computed Tomography Multiple Choice Question Answering on AMOS-MM-MCQ (OOD)
66.13Existence AccuracyTIF-GRPO w/ Acc Reward (AMOS-MM)
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| TIF-GRPO w/ Acc Reward (AMOS-MM)#Params=4B2026.05 | 66.13 | 76.32 | 95.15 | 74.24 | 69.67 | |
| TIF-GRPO w/ Acc Reward (CT-RATE)#Params=4B2026.05 | 58.06 | 56.14 | 86.41 | 63.05 | 66.65 | |
| GRPO w/ Acc Reward (AMOS-MM)#Params=4B2026.05 | 57.83 | 68.86 | 91.26 | 67.34 | 63.91 | |
| Qwen3-VL-8B#Params=8B2026.05 | 56.1 | 19.74 | 61.65 | 48.02 | 50.63 | |
| InternVL3.5-8B#Params=8B2026.05 | 54.1 | 17.98 | 58.74 | 45.88 | 49.18 | |
| GRPO w/ Acc Reward (CT-RATE)#Params=4B2026.05 | 53.46 | 53.51 | 85.44 | 59.89 | 63.59 | |
| RadFM#Params=14B2026.05 | 52.55 | 22.81 | 27.67 | 39.1 | 39.53 | |
| Fleming-VL-8B#Params=8B2026.05 | 52.11 | 38.6 | 64.08 | 51.41 | 53.92 | |
| CT-CHAT#Params=8B2026.05 | 51.45 | 45.18 | 67.96 | 53.67 | 54.22 | |
| GPT-5#Params=*2026.05 | 50.89 | 51.59 | 80.61 | 59.82 | 60.54 | |
| Hulu-Med#Params=7B2026.05 | 50.33 | 37.47 | 78.58 | 53.59 | 57.03 | |
| Gemini-3-Pro#Params=*2026.05 | 47.32 | 53.16 | 80.68 | 56.42 | 56.7 | |
| M3D-LaMed#Params=7B2026.05 | 35.46 | 43.86 | 66.5 | 44.86 | 48.72 |