Research map / Vision-language and multimodal / Visual question answering
Dual-Level Confidence based Implicit Self-Refinement for Medical Visual Question Answering
Meihong Pan, Yefeng Zheng
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add naive pseudo-labeled test samples as ground truth (ablation) | DuCoR baseline (Llama2) · PathVQA closed | accuracy: 87.22 -> 85.41 | Hurt |
| add naive pseudo-labels (ablation) | DuCoR baseline (Llama2) · PathVQA open | recall: 39.73 -> 40.16 | No clear effect |
| add loss-level confidence weighting (ablation) | DuCoR (Llama2) · PathVQA closed | accuracy: 85.41 -> 87.95 | Improved |
| add feature-level confidence (ablation) | DuCoR (Llama2) · PathVQA closed | accuracy: 87.95 -> 89.37 | Improved |
| add feature-level confidence (ablation) | DuCoR (Llama2) · PathVQA | F1: 41.97 -> 41.45 | Mixed |
| add prototype contrastive regularization (ablation) | DuCoR (Llama2) · PathVQA closed | accuracy: 89.37 -> 89.84 | Improved |
| use DuCoR instead of AdaContrast for pseudo-labeling | DuCoR · VQA-RAD | overall accuracy: 74.70 -> 77.56 | Improved |
| use DuCoR adaptation instead of zero-shot transfer | DuCoR · SLAKE to PathVQA | recall: 17.36 -> 24.56 | Improved |
| use DuCoR adaptation instead of zero-shot transfer | DuCoR · PathVQA to SLAKE | recall: 28.74 -> 42.32 | Improved |
| use DuCoR adaptation instead of zero-shot transfer | DuCoR · SLAKE to VQA-RAD | recall: 45.02 -> 51.81 | Improved |
| add DuCoR training | DuCoR · pseudo-label reliability | AUROC: 0.6827 -> 0.8575 | Improved |
| scale from GPT2 to Llama2 backbone | DuCoR · PathVQA closed | accuracy: 88.89 -> 89.84 | No clear effect |
Closest papers by meaning
- DiN: Diffusion Model for Robust Medical VQA with Semantic Noisy Labels (CVPR 2025 · similarity 0.89)
- Alignment, Mining and Fusion: Representation Alignment with Hard Negative Mining and Selective Knowledge Fusion for Medical Visual Question Answering (CVPR 2025 · similarity 0.88)
- Towards Self-Refinement of Vision-Language Models with Triangular Consistency (NeurIPS 2025 · similarity 0.88)
- Visual and Domain Knowledge for Professional-level Graph-of-Thought Medical Reasoning (ICML 2025 · similarity 0.87)
- Fine-grained Adaptive Visual Prompt for Generative Medical Visual Question Answering (AAAI 2025 · similarity 0.87)
- Towards a Multimodal Large Language Model with Pixel-Level Insight for Biomedicine (AAAI 2025 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
