Visual Question Answering on VQA-RAD (Open)
0.819AUROCLLaVA-NeXT-34B
Evaluation Results
| Method | Links | |
|---|---|---|
| LLaVA-NeXT-34BCalibration Method=HAC-Gate, Confidence Type=Verbalized2026.04 | 0.819 | |
| InternVL3-2BCalibration Method=HAC-Gate, Confidence Type=Sampling2026.04 | 0.815 | |
| InternVL3-2BCalibration Method=HAC-Gate, Confidence Type=Verbalized2026.04 | 0.808 | |
| LLaVA-NeXT-34BCalibration Method=HAC-Platt, Confidence Type=Verbalized2026.04 | 0.807 | |
| InternVL3-2BCalibration Method=HAC-Platt, Confidence Type=Verbalized2026.04 | 0.806 | |
| InternVL3-2BCalibration Method=HAC-Platt, Confidence Type=Sampling2026.04 | 0.803 | |
| Qwen3-VL-32BCalibration Method=HAC-Gate, Confidence Type=Verbalized2026.04 | 0.798 | |
| InternVL3-38BCalibration Method=HAC-Platt, Confidence Type=Sampling2026.04 | 0.796 | |
| InternVL3-38BCalibration Method=HAC-Gate, Confidence Type=Sampling2026.04 | 0.794 | |
| Qwen3-VL-32BCalibration Method=HAC-Platt, Confidence Type=Verbalized2026.04 | 0.788 | |
| LLaVA-NeXT-7BCalibration Method=HAC-Platt, Confidence Type=Sampling2026.04 | 0.788 | |
| LLaVA-NeXT-7BCalibration Method=HAC-Gate, Confidence Type=Sampling2026.04 | 0.788 | |
| Qwen3-VL-8BCalibration Method=HAC-Platt, Confidence Type=Verbalized2026.04 | 0.774 | |
| InternVL3-38BCalibration Method=HAC-Platt, Confidence Type=Verbalized2026.04 | 0.771 | |
| InternVL3-38BCalibration Method=Uncalibrated, Confidence Type=Sampling2026.04 | 0.769 | |
| InternVL3-38BCalibration Method=Platt Scaling, Confidence Type=Sampling2026.04 | 0.769 | |
| InternVL3-38BCalibration Method=Isotonic Regr., Confidence Type=Sampling2026.04 | 0.768 | |
| Qwen3-VL-8BCalibration Method=Uncalibrated, Confidence Type=Verbalized2026.04 | 0.753 | |
| Qwen3-VL-8BCalibration Method=Platt Scaling, Confidence Type=Verbalized2026.04 | 0.753 | |
| Qwen3-VL-8BCalibration Method=Isotonic Regr., Confidence Type=Verbalized2026.04 | 0.753 | |
| Qwen3-VL-32BCalibration Method=HAC-Platt, Confidence Type=Sampling2026.04 | 0.753 | |
| LLaVA-NeXT-7BCalibration Method=Isotonic Regr., Confidence Type=Sampling2026.04 | 0.749 | |
| Qwen3-VL-8BCalibration Method=HAC-Gate, Confidence Type=Verbalized2026.04 | 0.746 | |
| Qwen3-VL-32BCalibration Method=HAC-Gate, Confidence Type=Sampling2026.04 | 0.745 | |
| LLaVA-NeXT-7BCalibration Method=Uncalibrated, Confidence Type=Sampling2026.04 | 0.745 | |
| LLaVA-NeXT-7BCalibration Method=Platt Scaling, Confidence Type=Sampling2026.04 | 0.745 | |
| InternVL3-8BCalibration Method=HAC-Platt, Confidence Type=Verbalized2026.04 | 0.739 | |
| InternVL3-38BCalibration Method=Hist. Binning, Confidence Type=Sampling2026.04 | 0.738 | |
| LLaVA-NeXT-7BCalibration Method=HAC-Platt, Confidence Type=Verbalized2026.04 | 0.737 | |
| LLaVA-NeXT-7BCalibration Method=HAC-Gate, Confidence Type=Verbalized2026.04 | 0.737 | |
| Qwen3-VL-8BCalibration Method=Hist. Binning, Confidence Type=Verbalized2026.04 | 0.735 | |
| LLaVA-NeXT-34BCalibration Method=Isotonic Regr., Confidence Type=Sampling2026.04 | 0.728 | |
| InternVL3-2BCalibration Method=Uncalibrated, Confidence Type=Sampling2026.04 | 0.727 | |
| InternVL3-2BCalibration Method=Platt Scaling, Confidence Type=Sampling2026.04 | 0.727 | |
| InternVL3-38BCalibration Method=HAC-Gate, Confidence Type=Verbalized2026.04 | 0.726 | |
| InternVL3-38BCalibration Method=Uncalibrated, Confidence Type=Verbalized2026.04 | 0.72 | |
| InternVL3-38BCalibration Method=Platt Scaling, Confidence Type=Verbalized2026.04 | 0.72 | |
| Qwen3-VL-2BCalibration Method=HAC-Gate, Confidence Type=Verbalized2026.04 | 0.717 | |
| Qwen3-VL-8BCalibration Method=HAC-Platt, Confidence Type=Sampling2026.04 | 0.717 | |
| Qwen3-VL-8BCalibration Method=HAC-Gate, Confidence Type=Sampling2026.04 | 0.717 | |
| Qwen3-VL-2BCalibration Method=HAC-Platt, Confidence Type=Verbalized2026.04 | 0.716 | |
| InternVL3-38BCalibration Method=Isotonic Regr., Confidence Type=Verbalized2026.04 | 0.716 | |
| InternVL3-38BCalibration Method=Hist. Binning, Confidence Type=Verbalized2026.04 | 0.715 | |
| InternVL3-8BCalibration Method=HAC-Gate, Confidence Type=Verbalized2026.04 | 0.713 | |
| LLaVA-NeXT-34BCalibration Method=Uncalibrated, Confidence Type=Sampling2026.04 | 0.713 | |
| LLaVA-NeXT-34BCalibration Method=Platt Scaling, Confidence Type=Sampling2026.04 | 0.713 | |
| Qwen3-VL-32BCalibration Method=Uncalibrated, Confidence Type=Verbalized2026.04 | 0.708 | |
| Qwen3-VL-32BCalibration Method=Platt Scaling, Confidence Type=Verbalized2026.04 | 0.708 | |
| InternVL3-2BCalibration Method=Isotonic Regr., Confidence Type=Sampling2026.04 | 0.705 | |
| InternVL3-8BCalibration Method=HAC-Gate, Confidence Type=Sampling2026.04 | 0.704 | |
| LLaVA-NeXT-34BCalibration Method=HAC-Platt, Confidence Type=Sampling2026.04 | 0.704 | |
| LLaVA-NeXT-34BCalibration Method=Uncalibrated, Confidence Type=Verbalized2026.04 | 0.701 | |
| LLaVA-NeXT-34BCalibration Method=Platt Scaling, Confidence Type=Verbalized2026.04 | 0.701 | |
| InternVL3-8BCalibration Method=HAC-Platt, Confidence Type=Sampling2026.04 | 0.699 | |
| LLaVA-NeXT-34BCalibration Method=Isotonic Regr., Confidence Type=Verbalized2026.04 | 0.699 | |
| Qwen3-VL-32BCalibration Method=Isotonic Regr., Confidence Type=Verbalized2026.04 | 0.698 | |
| LLaVA-NeXT-34BCalibration Method=HAC-Gate, Confidence Type=Sampling2026.04 | 0.696 | |
| LLaVA-NeXT-34BCalibration Method=Hist. Binning, Confidence Type=Sampling2026.04 | 0.69 | |
| Qwen3-VL-2BCalibration Method=HAC-Gate, Confidence Type=Sampling2026.04 | 0.676 | |
| Qwen3-VL-8BCalibration Method=Isotonic Regr., Confidence Type=Sampling2026.04 | 0.676 | |
| Qwen3-VL-8BCalibration Method=Uncalibrated, Confidence Type=Sampling2026.04 | 0.675 | |
| Qwen3-VL-8BCalibration Method=Platt Scaling, Confidence Type=Sampling2026.04 | 0.675 | |
| Qwen3-VL-2BCalibration Method=HAC-Platt, Confidence Type=Sampling2026.04 | 0.672 | |
| Qwen3-VL-2BCalibration Method=Uncalibrated, Confidence Type=Sampling2026.04 | 0.666 | |
| Qwen3-VL-2BCalibration Method=Platt Scaling, Confidence Type=Sampling2026.04 | 0.666 | |
| InternVL3-2BCalibration Method=Hist. Binning, Confidence Type=Sampling2026.04 | 0.663 | |
| LLaVA-NeXT-34BCalibration Method=Hist. Binning, Confidence Type=Verbalized2026.04 | 0.662 | |
| InternVL3-8BCalibration Method=Uncalibrated, Confidence Type=Sampling2026.04 | 0.654 | |
| InternVL3-8BCalibration Method=Platt Scaling, Confidence Type=Sampling2026.04 | 0.654 | |
| Qwen3-VL-32BCalibration Method=Hist. Binning, Confidence Type=Verbalized2026.04 | 0.653 | |
| Qwen3-VL-2BCalibration Method=Isotonic Regr., Confidence Type=Sampling2026.04 | 0.65 | |
| InternVL3-8BCalibration Method=Isotonic Regr., Confidence Type=Sampling2026.04 | 0.64 | |
| LLaVA-NeXT-7BCalibration Method=Hist. Binning, Confidence Type=Sampling2026.04 | 0.631 | |
| Qwen3-VL-2BCalibration Method=Uncalibrated, Confidence Type=Verbalized2026.04 | 0.625 | |
| Qwen3-VL-2BCalibration Method=Platt Scaling, Confidence Type=Verbalized2026.04 | 0.625 | |
| InternVL3-8BCalibration Method=Isotonic Regr., Confidence Type=Verbalized2026.04 | 0.617 | |
| InternVL3-8BCalibration Method=Uncalibrated, Confidence Type=Verbalized2026.04 | 0.616 | |
| InternVL3-8BCalibration Method=Platt Scaling, Confidence Type=Verbalized2026.04 | 0.616 | |
| Qwen3-VL-8BCalibration Method=Hist. Binning, Confidence Type=Sampling2026.04 | 0.614 | |
| InternVL3-8BCalibration Method=Hist. Binning, Confidence Type=Verbalized2026.04 | 0.604 | |
| Qwen3-VL-2BCalibration Method=Hist. Binning, Confidence Type=Sampling2026.04 | 0.601 | |
| Qwen3-VL-2BCalibration Method=Isotonic Regr., Confidence Type=Verbalized2026.04 | 0.599 | |
| InternVL3-8BCalibration Method=Hist. Binning, Confidence Type=Sampling2026.04 | 0.598 | |
| LLaVA-NeXT-7BCalibration Method=Uncalibrated, Confidence Type=Verbalized2026.04 | 0.589 | |
| LLaVA-NeXT-7BCalibration Method=Platt Scaling, Confidence Type=Verbalized2026.04 | 0.589 | |
| Qwen3-VL-2BCalibration Method=Hist. Binning, Confidence Type=Verbalized2026.04 | 0.587 | |
| LLaVA-NeXT-7BCalibration Method=Isotonic Regr., Confidence Type=Verbalized2026.04 | 0.581 | |
| Qwen3-VL-32BCalibration Method=Isotonic Regr., Confidence Type=Sampling2026.04 | 0.579 | |
| Qwen3-VL-32BCalibration Method=Platt Scaling, Confidence Type=Sampling2026.04 | 0.578 | |
| Qwen3-VL-32BCalibration Method=Uncalibrated, Confidence Type=Sampling2026.04 | 0.577 | |
| Qwen3-VL-32BCalibration Method=Hist. Binning, Confidence Type=Sampling2026.04 | 0.548 | |
| LLaVA-NeXT-7BCalibration Method=Hist. Binning, Confidence Type=Verbalized2026.04 | 0.536 | |
| InternVL3-2BCalibration Method=Uncalibrated, Confidence Type=Verbalized2026.04 | 0.514 | |
| InternVL3-2BCalibration Method=Platt Scaling, Confidence Type=Verbalized2026.04 | 0.514 | |
| InternVL3-2BCalibration Method=Isotonic Regr., Confidence Type=Verbalized2026.04 | 0.468 | |
| InternVL3-2BCalibration Method=Hist. Binning, Confidence Type=Verbalized2026.04 | 0.439 |