Medical Visual Question Answering on VQA-RAD (Closed)
1.3ECEQwen3-VL-2B
Evaluation Results
| Method | Links | |
|---|---|---|
| Qwen3-VL-2BCalibration Method=Platt Scaling, Confidence Type=Verbalized, 5-fold CV=true2026.04 | 1.3 | |
| Qwen3-VL-2BCalibration Method=Hist. Binning, Confidence Type=Verbalized, 5-fold CV=true2026.04 | 1.3 | |
| Qwen3-VL-2BCalibration Method=Isotonic Regr., Confidence Type=Verbalized, 5-fold CV=true2026.04 | 1.4 | |
| LLaVA-NeXT-34BCalibration Method=Platt Scaling, Confidence Type=Verbalized, 5-fold CV=true2026.04 | 1.9 | |
| InternVL3-8BCalibration Method=Platt Scaling, Confidence Type=Verbalized, 5-fold CV=true2026.04 | 2.3 | |
| InternVL3-8BCalibration Method=Isotonic Regr., Confidence Type=Verbalized, 5-fold CV=true2026.04 | 2.8 | |
| LLaVA-NeXT-7BCalibration Method=Platt Scaling, Confidence Type=Verbalized, 5-fold CV=true2026.04 | 2.9 | |
| LLaVA-NeXT-34BCalibration Method=HAC-Platt, Confidence Type=Verbalized, 5-fold CV=true2026.04 | 3 | |
| Qwen3-VL-2BCalibration Method=HAC-Gate, Confidence Type=Verbalized, 5-fold CV=true2026.04 | 3.2 | |
| InternVL3-8BCalibration Method=HAC-Platt, Confidence Type=Verbalized, 5-fold CV=true2026.04 | 3.3 | |
| InternVL3-8BCalibration Method=HAC-Gate, Confidence Type=Verbalized, 5-fold CV=true2026.04 | 3.5 | |
| Qwen3-VL-8BCalibration Method=Isotonic Regr., Confidence Type=Sampling, 5-fold CV=true2026.04 | 3.7 | |
| InternVL3-38BCalibration Method=HAC-Gate, Confidence Type=Verbalized, 5-fold CV=true2026.04 | 3.7 | |
| LLaVA-NeXT-7BCalibration Method=HAC-Platt, Confidence Type=Verbalized, 5-fold CV=true2026.04 | 3.8 | |
| LLaVA-NeXT-34BCalibration Method=Hist. Binning, Confidence Type=Verbalized, 5-fold CV=true2026.04 | 3.8 | |
| Qwen3-VL-8BCalibration Method=Hist. Binning, Confidence Type=Sampling, 5-fold CV=true2026.04 | 4.4 | |
| InternVL3-8BCalibration Method=Hist. Binning, Confidence Type=Verbalized, 5-fold CV=true2026.04 | 4.4 | |
| LLaVA-NeXT-34BCalibration Method=Isotonic Regr., Confidence Type=Verbalized, 5-fold CV=true2026.04 | 4.4 | |
| Qwen3-VL-32BCalibration Method=Isotonic Regr., Confidence Type=Verbalized, 5-fold CV=true2026.04 | 4.5 | |
| Qwen3-VL-32BCalibration Method=Isotonic Regr., Confidence Type=Sampling, 5-fold CV=true2026.04 | 4.6 | |
| Qwen3-VL-2BCalibration Method=HAC-Platt, Confidence Type=Verbalized, 5-fold CV=true2026.04 | 4.7 | |
| Qwen3-VL-8BCalibration Method=HAC-Platt, Confidence Type=Verbalized, 5-fold CV=true2026.04 | 4.8 | |
| Qwen3-VL-32BCalibration Method=Hist. Binning, Confidence Type=Verbalized, 5-fold CV=true2026.04 | 4.8 | |
| LLaVA-NeXT-7BCalibration Method=Isotonic Regr., Confidence Type=Verbalized, 5-fold CV=true2026.04 | 4.8 | |
| Qwen3-VL-8BCalibration Method=HAC-Gate, Confidence Type=Sampling, 5-fold CV=true2026.04 | 4.9 | |
| LLaVA-NeXT-34BCalibration Method=HAC-Gate, Confidence Type=Verbalized, 5-fold CV=true2026.04 | 4.9 | |
| Qwen3-VL-32BCalibration Method=HAC-Gate, Confidence Type=Verbalized, 5-fold CV=true2026.04 | 5.1 | |
| InternVL3-2BCalibration Method=Isotonic Regr., Confidence Type=Verbalized, 5-fold CV=true2026.04 | 5.2 | |
| Qwen3-VL-32BCalibration Method=HAC-Gate, Confidence Type=Sampling, 5-fold CV=true2026.04 | 5.4 | |
| Qwen3-VL-8BCalibration Method=Isotonic Regr., Confidence Type=Verbalized, 5-fold CV=true2026.04 | 5.5 | |
| Qwen3-VL-8BCalibration Method=HAC-Platt, Confidence Type=Sampling, 5-fold CV=true2026.04 | 5.5 | |
| InternVL3-2BCalibration Method=Platt Scaling, Confidence Type=Verbalized, 5-fold CV=true2026.04 | 5.5 | |
| Qwen3-VL-8BCalibration Method=Platt Scaling, Confidence Type=Sampling, 5-fold CV=true2026.04 | 5.7 | |
| Qwen3-VL-32BCalibration Method=HAC-Platt, Confidence Type=Sampling, 5-fold CV=true2026.04 | 5.7 | |
| Qwen3-VL-32BCalibration Method=HAC-Platt, Confidence Type=Verbalized, 5-fold CV=true2026.04 | 5.7 | |
| InternVL3-38BCalibration Method=Hist. Binning, Confidence Type=Verbalized, 5-fold CV=true2026.04 | 5.7 | |
| Qwen3-VL-32BCalibration Method=Hist. Binning, Confidence Type=Sampling, 5-fold CV=true2026.04 | 5.8 | |
| Qwen3-VL-8BCalibration Method=Platt Scaling, Confidence Type=Verbalized, 5-fold CV=true2026.04 | 5.9 | |
| Qwen3-VL-32BCalibration Method=Platt Scaling, Confidence Type=Sampling, 5-fold CV=true2026.04 | 6 | |
| Qwen3-VL-32BCalibration Method=Platt Scaling, Confidence Type=Verbalized, 5-fold CV=true2026.04 | 6.1 | |
| InternVL3-38BCalibration Method=Isotonic Regr., Confidence Type=Verbalized, 5-fold CV=true2026.04 | 6.1 | |
| InternVL3-2BCalibration Method=Hist. Binning, Confidence Type=Verbalized, 5-fold CV=true2026.04 | 6.2 | |
| Qwen3-VL-8BCalibration Method=HAC-Gate, Confidence Type=Verbalized, 5-fold CV=true2026.04 | 6.4 | |
| Qwen3-VL-8BCalibration Method=Hist. Binning, Confidence Type=Verbalized, 5-fold CV=true2026.04 | 6.7 | |
| InternVL3-38BCalibration Method=Isotonic Regr., Confidence Type=Sampling, 5-fold CV=true2026.04 | 7.1 | |
| InternVL3-2BCalibration Method=HAC-Gate, Confidence Type=Verbalized, 5-fold CV=true2026.04 | 7.8 | |
| LLaVA-NeXT-7BCalibration Method=Hist. Binning, Confidence Type=Verbalized, 5-fold CV=true2026.04 | 7.8 | |
| InternVL3-38BCalibration Method=Platt Scaling, Confidence Type=Verbalized, 5-fold CV=true2026.04 | 8.3 | |
| InternVL3-38BCalibration Method=HAC-Platt, Confidence Type=Verbalized, 5-fold CV=true2026.04 | 8.3 | |
| InternVL3-2BCalibration Method=HAC-Platt, Confidence Type=Verbalized, 5-fold CV=true2026.04 | 8.5 | |
| Qwen3-VL-2BCalibration Method=Isotonic Regr., Confidence Type=Sampling, 5-fold CV=true2026.04 | 8.6 | |
| InternVL3-2BCalibration Method=Isotonic Regr., Confidence Type=Sampling, 5-fold CV=true2026.04 | 9.9 | |
| InternVL3-38BCalibration Method=Hist. Binning, Confidence Type=Sampling, 5-fold CV=true2026.04 | 10.3 | |
| LLaVA-NeXT-7BCalibration Method=Platt Scaling, Confidence Type=Sampling, 5-fold CV=true2026.04 | 10.4 | |
| LLaVA-NeXT-7BCalibration Method=HAC-Platt, Confidence Type=Sampling, 5-fold CV=true2026.04 | 10.4 | |
| LLaVA-NeXT-7BCalibration Method=Isotonic Regr., Confidence Type=Sampling, 5-fold CV=true2026.04 | 10.7 | |
| InternVL3-38BCalibration Method=Platt Scaling, Confidence Type=Sampling, 5-fold CV=true2026.04 | 10.8 | |
| InternVL3-38BCalibration Method=HAC-Platt, Confidence Type=Sampling, 5-fold CV=true2026.04 | 10.8 | |
| Qwen3-VL-2BCalibration Method=HAC-Gate, Confidence Type=Sampling, 5-fold CV=true2026.04 | 11 | |
| LLaVA-NeXT-7BCalibration Method=HAC-Gate, Confidence Type=Verbalized, 5-fold CV=true2026.04 | 11.2 | |
| Qwen3-VL-2BCalibration Method=Platt Scaling, Confidence Type=Sampling, 5-fold CV=true2026.04 | 11.5 | |
| Qwen3-VL-2BCalibration Method=Hist. Binning, Confidence Type=Sampling, 5-fold CV=true2026.04 | 11.7 | |
| LLaVA-NeXT-34BCalibration Method=Hist. Binning, Confidence Type=Sampling, 5-fold CV=true2026.04 | 11.7 | |
| Qwen3-VL-2BCalibration Method=HAC-Platt, Confidence Type=Sampling, 5-fold CV=true2026.04 | 11.8 | |
| LLaVA-NeXT-7BCalibration Method=HAC-Gate, Confidence Type=Sampling, 5-fold CV=true2026.04 | 11.8 | |
| InternVL3-38BCalibration Method=HAC-Gate, Confidence Type=Sampling, 5-fold CV=true2026.04 | 11.9 | |
| LLaVA-NeXT-7BCalibration Method=Hist. Binning, Confidence Type=Sampling, 5-fold CV=true2026.04 | 12 | |
| InternVL3-8BCalibration Method=Hist. Binning, Confidence Type=Sampling, 5-fold CV=true2026.04 | 12.1 | |
| LLaVA-NeXT-34BCalibration Method=Platt Scaling, Confidence Type=Sampling, 5-fold CV=true2026.04 | 12.3 | |
| LLaVA-NeXT-34BCalibration Method=HAC-Platt, Confidence Type=Sampling, 5-fold CV=true2026.04 | 12.3 | |
| InternVL3-38BCalibration Method=Uncalibrated, Confidence Type=Sampling, 5-fold CV=true2026.04 | 12.4 | |
| InternVL3-8BCalibration Method=Platt Scaling, Confidence Type=Sampling, 5-fold CV=true2026.04 | 12.8 | |
| InternVL3-8BCalibration Method=HAC-Platt, Confidence Type=Sampling, 5-fold CV=true2026.04 | 12.9 | |
| LLaVA-NeXT-34BCalibration Method=HAC-Gate, Confidence Type=Sampling, 5-fold CV=true2026.04 | 12.9 | |
| LLaVA-NeXT-34BCalibration Method=Isotonic Regr., Confidence Type=Sampling, 5-fold CV=true2026.04 | 13 | |
| InternVL3-2BCalibration Method=HAC-Gate, Confidence Type=Sampling, 5-fold CV=true2026.04 | 13.1 | |
| InternVL3-8BCalibration Method=Isotonic Regr., Confidence Type=Sampling, 5-fold CV=true2026.04 | 13.3 | |
| LLaVA-NeXT-7BCalibration Method=Uncalibrated, Confidence Type=Sampling, 5-fold CV=true2026.04 | 13.3 | |
| InternVL3-2BCalibration Method=HAC-Platt, Confidence Type=Sampling, 5-fold CV=true2026.04 | 14.6 | |
| InternVL3-2BCalibration Method=Platt Scaling, Confidence Type=Sampling, 5-fold CV=true2026.04 | 14.7 | |
| InternVL3-8BCalibration Method=HAC-Gate, Confidence Type=Sampling, 5-fold CV=true2026.04 | 14.9 | |
| InternVL3-2BCalibration Method=Hist. Binning, Confidence Type=Sampling, 5-fold CV=true2026.04 | 15.6 | |
| Qwen3-VL-32BCalibration Method=Uncalibrated, Confidence Type=Verbalized, 5-fold CV=true2026.04 | 17.4 | |
| InternVL3-38BCalibration Method=Uncalibrated, Confidence Type=Verbalized, 5-fold CV=true2026.04 | 18.3 | |
| InternVL3-8BCalibration Method=Uncalibrated, Confidence Type=Sampling, 5-fold CV=true2026.04 | 18.8 | |
| InternVL3-8BCalibration Method=Uncalibrated, Confidence Type=Verbalized, 5-fold CV=true2026.04 | 20.3 | |
| LLaVA-NeXT-34BCalibration Method=Uncalibrated, Confidence Type=Sampling, 5-fold CV=true2026.04 | 21.2 | |
| Qwen3-VL-8BCalibration Method=Uncalibrated, Confidence Type=Verbalized, 5-fold CV=true2026.04 | 23 | |
| Qwen3-VL-32BCalibration Method=Uncalibrated, Confidence Type=Sampling, 5-fold CV=true2026.04 | 23 | |
| InternVL3-2BCalibration Method=Uncalibrated, Confidence Type=Sampling, 5-fold CV=true2026.04 | 23.2 | |
| InternVL3-2BCalibration Method=Uncalibrated, Confidence Type=Verbalized, 5-fold CV=true2026.04 | 23.2 | |
| Qwen3-VL-2BCalibration Method=Uncalibrated, Confidence Type=Verbalized, 5-fold CV=true2026.04 | 24.5 | |
| Qwen3-VL-8BCalibration Method=Uncalibrated, Confidence Type=Sampling, 5-fold CV=true2026.04 | 24.7 | |
| LLaVA-NeXT-34BCalibration Method=Uncalibrated, Confidence Type=Verbalized, 5-fold CV=true2026.04 | 27 | |
| Qwen3-VL-2BCalibration Method=Uncalibrated, Confidence Type=Sampling, 5-fold CV=true2026.04 | 28.8 | |
| LLaVA-NeXT-7BCalibration Method=Uncalibrated, Confidence Type=Verbalized, 5-fold CV=true2026.04 | 35.7 |