Medical Visual Question Answering on VQA-Med Open
0.006ECEInternVL3-2B
Evaluation Results
| Method | Links | |
|---|---|---|
| InternVL3-2BCalibration Strategy=Platt Scaling, Confidence Mode=Verbalized2026.04 | 0.006 | |
| Qwen3-VL-2BCalibration Strategy=Hist. Binning, Confidence Mode=Verbalized2026.04 | 0.012 | |
| Qwen3-VL-2BCalibration Strategy=Isotonic Regr., Confidence Mode=Verbalized2026.04 | 0.02 | |
| InternVL3-8BCalibration Strategy=Isotonic Regr., Confidence Mode=Verbalized2026.04 | 0.024 | |
| InternVL3-8BCalibration Strategy=Hist. Binning, Confidence Mode=Verbalized2026.04 | 0.024 | |
| InternVL3-38BCalibration Strategy=Isotonic Regr., Confidence Mode=Verbalized2026.04 | 0.026 | |
| LLaVA-NeXT-7BCalibration Strategy=Isotonic Regr., Confidence Mode=Verbalized2026.04 | 0.026 | |
| LLaVA-NeXT-7BCalibration Strategy=Hist. Binning, Confidence Mode=Verbalized2026.04 | 0.026 | |
| InternVL3-8BCalibration Strategy=Platt Scaling, Confidence Mode=Verbalized2026.04 | 0.028 | |
| LLaVA-NeXT-7BCalibration Strategy=Platt Scaling, Confidence Mode=Verbalized2026.04 | 0.028 | |
| InternVL3-38BCalibration Strategy=Hist. Binning, Confidence Mode=Verbalized2026.04 | 0.029 | |
| Qwen3-VL-32BCalibration Strategy=Isotonic Regr., Confidence Mode=Sampling2026.04 | 0.031 | |
| Qwen3-VL-2BCalibration Strategy=Platt Scaling, Confidence Mode=Verbalized2026.04 | 0.032 | |
| InternVL3-2BCalibration Strategy=Isotonic Regr., Confidence Mode=Verbalized2026.04 | 0.033 | |
| InternVL3-2BCalibration Strategy=Hist. Binning, Confidence Mode=Verbalized2026.04 | 0.034 | |
| Qwen3-VL-32BCalibration Strategy=Hist. Binning, Confidence Mode=Verbalized2026.04 | 0.035 | |
| Qwen3-VL-32BCalibration Strategy=Isotonic Regr., Confidence Mode=Verbalized2026.04 | 0.036 | |
| Qwen3-VL-32BCalibration Strategy=Hist. Binning, Confidence Mode=Sampling2026.04 | 0.041 | |
| Qwen3-VL-8BCalibration Strategy=Hist. Binning, Confidence Mode=Sampling2026.04 | 0.042 | |
| Qwen3-VL-8BCalibration Strategy=Hist. Binning, Confidence Mode=Verbalized2026.04 | 0.042 | |
| Qwen3-VL-8BCalibration Strategy=Isotonic Regr., Confidence Mode=Sampling2026.04 | 0.044 | |
| LLaVA-NeXT-34BCalibration Strategy=Isotonic Regr., Confidence Mode=Sampling2026.04 | 0.047 | |
| LLaVA-NeXT-34BCalibration Strategy=Isotonic Regr., Confidence Mode=Verbalized2026.04 | 0.047 | |
| LLaVA-NeXT-34BCalibration Strategy=Hist. Binning, Confidence Mode=Verbalized2026.04 | 0.047 | |
| LLaVA-NeXT-7BCalibration Strategy=Isotonic Regr., Confidence Mode=Sampling2026.04 | 0.049 | |
| Qwen3-VL-8BCalibration Strategy=Platt Scaling, Confidence Mode=Sampling2026.04 | 0.051 | |
| Qwen3-VL-32BCalibration Strategy=Platt Scaling, Confidence Mode=Sampling2026.04 | 0.055 | |
| InternVL3-38BCalibration Strategy=Isotonic Regr., Confidence Mode=Sampling2026.04 | 0.055 | |
| Qwen3-VL-2BCalibration Strategy=Isotonic Regr., Confidence Mode=Sampling2026.04 | 0.059 | |
| LLaVA-NeXT-34BCalibration Strategy=Platt Scaling, Confidence Mode=Verbalized2026.04 | 0.064 | |
| InternVL3-38BCalibration Strategy=Platt Scaling, Confidence Mode=Sampling2026.04 | 0.065 | |
| Qwen3-VL-2BCalibration Strategy=Platt Scaling, Confidence Mode=Sampling2026.04 | 0.067 | |
| InternVL3-38BCalibration Strategy=Hist. Binning, Confidence Mode=Sampling2026.04 | 0.067 | |
| LLaVA-NeXT-34BCalibration Strategy=Platt Scaling, Confidence Mode=Sampling2026.04 | 0.072 | |
| Qwen3-VL-2BCalibration Strategy=HAC-Platt, Confidence Mode=Sampling2026.04 | 0.076 | |
| Qwen3-VL-2BCalibration Strategy=Hist. Binning, Confidence Mode=Sampling2026.04 | 0.077 | |
| InternVL3-8BCalibration Strategy=Isotonic Regr., Confidence Mode=Sampling2026.04 | 0.078 | |
| LLaVA-NeXT-7BCalibration Strategy=Hist. Binning, Confidence Mode=Sampling2026.04 | 0.079 | |
| Qwen3-VL-8BCalibration Strategy=Platt Scaling, Confidence Mode=Verbalized2026.04 | 0.083 | |
| Qwen3-VL-8BCalibration Strategy=Isotonic Regr., Confidence Mode=Verbalized2026.04 | 0.085 | |
| InternVL3-38BCalibration Strategy=HAC-Platt, Confidence Mode=Sampling2026.04 | 0.085 | |
| InternVL3-38BCalibration Strategy=HAC-Gate, Confidence Mode=Sampling2026.04 | 0.087 | |
| LLaVA-NeXT-34BCalibration Strategy=HAC-Gate, Confidence Mode=Sampling2026.04 | 0.089 | |
| LLaVA-NeXT-7BCalibration Strategy=HAC-Platt, Confidence Mode=Verbalized2026.04 | 0.09 | |
| LLaVA-NeXT-34BCalibration Strategy=Hist. Binning, Confidence Mode=Sampling2026.04 | 0.091 | |
| Qwen3-VL-2BCalibration Strategy=HAC-Gate, Confidence Mode=Sampling2026.04 | 0.093 | |
| InternVL3-8BCalibration Strategy=Hist. Binning, Confidence Mode=Sampling2026.04 | 0.098 | |
| LLaVA-NeXT-7BCalibration Strategy=HAC-Gate, Confidence Mode=Sampling2026.04 | 0.098 | |
| LLaVA-NeXT-7BCalibration Strategy=HAC-Gate, Confidence Mode=Verbalized2026.04 | 0.098 | |
| LLaVA-NeXT-34BCalibration Strategy=HAC-Platt, Confidence Mode=Sampling2026.04 | 0.099 | |
| InternVL3-2BCalibration Strategy=HAC-Platt, Confidence Mode=Sampling2026.04 | 0.104 | |
| Hist. BinningBackbone=InternVL3-38B, Confidence Type=Sampling, Protocol=5-fold CV2026.04 | 0.106 | |
| InternVL3-2BCalibration Strategy=HAC-Gate, Confidence Mode=Sampling2026.04 | 0.107 | |
| Qwen3-VL-32BCalibration Strategy=Platt Scaling, Confidence Mode=Verbalized2026.04 | 0.108 | |
| Qwen3-VL-2BCalibration Strategy=HAC-Gate, Confidence Mode=Verbalized2026.04 | 0.112 | |
| InternVL3-2BCalibration Strategy=Hist. Binning, Confidence Mode=Sampling2026.04 | 0.113 | |
| InternVL3-2BCalibration Strategy=Isotonic Regr., Confidence Mode=Sampling2026.04 | 0.114 | |
| Hist. BinningBackbone=LLaVA-NeXT-7B, Confidence Type=Verbalized, Protocol=5-fold CV2026.04 | 0.114 | |
| Qwen3-VL-32BCalibration Strategy=HAC-Gate, Confidence Mode=Sampling2026.04 | 0.116 | |
| LLaVA-NeXT-7BCalibration Strategy=Platt Scaling, Confidence Mode=Sampling2026.04 | 0.116 | |
| Isotonic Regr.Backbone=InternVL3-38B, Confidence Type=Sampling, Protocol=5-fold CV2026.04 | 0.116 | |
| Qwen3-VL-32BCalibration Strategy=HAC-Platt, Confidence Mode=Sampling2026.04 | 0.117 | |
| Platt ScalingBackbone=InternVL3-38B, Confidence Type=Sampling, Protocol=5-fold CV2026.04 | 0.117 | |
| Qwen3-VL-8BCalibration Strategy=HAC-Gate, Confidence Mode=Verbalized2026.04 | 0.118 | |
| HAC-PlattBackbone=InternVL3-38B, Confidence Type=Sampling, Protocol=5-fold CV2026.04 | 0.118 | |
| HAC-GateBackbone=InternVL3-38B, Confidence Type=Sampling, Protocol=5-fold CV2026.04 | 0.118 | |
| InternVL3-2BCalibration Strategy=Platt Scaling, Confidence Mode=Sampling2026.04 | 0.119 | |
| Qwen3-VL-2BCalibration Strategy=HAC-Platt, Confidence Mode=Verbalized2026.04 | 0.121 | |
| HAC-GateBackbone=InternVL3-2B, Confidence Type=Sampling, Protocol=5-fold CV2026.04 | 0.122 | |
| Hist. BinningBackbone=LLaVA-NeXT-7B, Confidence Type=Sampling, Protocol=5-fold CV2026.04 | 0.122 | |
| Isotonic Regr.Backbone=LLaVA-NeXT-7B, Confidence Type=Verbalized, Protocol=5-fold CV2026.04 | 0.124 | |
| Isotonic Regr.Backbone=Qwen3-VL-2B, Confidence Type=Sampling, Protocol=5-fold CV2026.04 | 0.125 | |
| Platt ScalingBackbone=LLaVA-NeXT-7B, Confidence Type=Verbalized, Protocol=5-fold CV2026.04 | 0.125 | |
| Qwen3-VL-8BCalibration Strategy=HAC-Platt, Confidence Mode=Sampling2026.04 | 0.127 | |
| Platt ScalingBackbone=Qwen3-VL-8B, Confidence Type=Sampling, Protocol=5-fold CV2026.04 | 0.127 | |
| Isotonic Regr.Backbone=LLaVA-NeXT-7B, Confidence Type=Sampling, Protocol=5-fold CV2026.04 | 0.128 | |
| LLaVA-NeXT-34BCalibration Strategy=HAC-Gate, Confidence Mode=Verbalized2026.04 | 0.129 | |
| Platt ScalingBackbone=InternVL3-2B, Confidence Type=Sampling, Protocol=5-fold CV2026.04 | 0.13 | |
| HAC-PlattBackbone=InternVL3-2B, Confidence Type=Sampling, Protocol=5-fold CV2026.04 | 0.13 | |
| Qwen3-VL-32BCalibration Strategy=HAC-Gate, Confidence Mode=Verbalized2026.04 | 0.131 | |
| Qwen3-VL-8BCalibration Strategy=HAC-Platt, Confidence Mode=Verbalized2026.04 | 0.132 | |
| Platt ScalingBackbone=Qwen3-VL-2B, Confidence Type=Sampling, Protocol=5-fold CV2026.04 | 0.132 | |
| InternVL3-8BCalibration Strategy=Platt Scaling, Confidence Mode=Sampling2026.04 | 0.133 | |
| Isotonic Regr.Backbone=LLaVA-NeXT-34B, Confidence Type=Sampling, Protocol=5-fold CV2026.04 | 0.133 | |
| Isotonic Regr.Backbone=Qwen3-VL-8B, Confidence Type=Sampling, Protocol=5-fold CV2026.04 | 0.134 | |
| Isotonic Regr.Backbone=LLaVA-NeXT-34B, Confidence Type=Verbalized, Protocol=5-fold CV2026.04 | 0.134 | |
| Hist. BinningBackbone=LLaVA-NeXT-34B, Confidence Type=Verbalized, Protocol=5-fold CV2026.04 | 0.134 | |
| Qwen3-VL-8BCalibration Strategy=HAC-Gate, Confidence Mode=Sampling2026.04 | 0.135 | |
| HAC-PlattBackbone=Qwen3-VL-2B, Confidence Type=Sampling, Protocol=5-fold CV2026.04 | 0.135 | |
| Isotonic Regr.Backbone=InternVL3-2B, Confidence Type=Sampling, Protocol=5-fold CV2026.04 | 0.135 | |
| Qwen3-VL-32BCalibration Strategy=HAC-Platt, Confidence Mode=Verbalized2026.04 | 0.137 | |
| HAC-PlattBackbone=Qwen3-VL-8B, Confidence Type=Verbalized, Protocol=5-fold CV2026.04 | 0.137 | |
| Isotonic Regr.Backbone=InternVL3-8B, Confidence Type=Sampling, Protocol=5-fold CV2026.04 | 0.139 | |
| HAC-PlattBackbone=LLaVA-NeXT-7B, Confidence Type=Verbalized, Protocol=5-fold CV2026.04 | 0.139 | |
| HAC-GateBackbone=LLaVA-NeXT-34B, Confidence Type=Sampling, Protocol=5-fold CV2026.04 | 0.14 | |
| InternVL3-8BCalibration Strategy=HAC-Platt, Confidence Mode=Verbalized2026.04 | 0.141 | |
| InternVL3-38BCalibration Strategy=Platt Scaling, Confidence Mode=Verbalized2026.04 | 0.141 | |
| LLaVA-NeXT-34BCalibration Strategy=HAC-Platt, Confidence Mode=Verbalized2026.04 | 0.141 | |
| HAC-GateBackbone=LLaVA-NeXT-7B, Confidence Type=Sampling, Protocol=5-fold CV2026.04 | 0.141 | |
| LLaVA-NeXT-7BCalibration Strategy=HAC-Platt, Confidence Mode=Sampling2026.04 | 0.142 |