Research map / Vision-language and multimodal / Multimodal LLMs
Proximity-Constrained Counterfactual Decoding for Hallucination-Robust Medical VQA
Dwarikanath Mahapatra, Abhijit Das, Manish Pandey, Joy Dhar, Sudipta Roy, Zongyuan Ge, Behzad Bozorgtabar, Imran Razzak
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 dual proximity guards (s, s_obj) (ablation) | PCCD · MIMIC-Diff-VQA | AUC: 87.50 -> 89.59 | Improved |
| add dual proximity guards (s, s_obj) (ablation) | PCCD · Gut-VLM description | QAAS / R-Sim: -1.55 QAAS / -0.14 R-Sim when removed | Improved |
| add object-masked proximity guard s_obj (ablation) | PCCD · MIMIC-Diff-VQA | AUC: 88.10 -> 89.59 | Improved |
| add LD-VCD branch to OA-only (ablation) | PCCD · MIMIC-Diff-VQA | AUC: 88.55 -> 89.59 | Improved |
| add OA-VCD branch to LD-only (ablation) | PCCD · MIMIC-Diff-VQA | AUC: 87.95 -> 89.59 | Improved |
| LD-only vs OA-only on attribute-heavy description (ablation) | PCCD · Gut-VLM description | R-Sim: 4.08 vs 4.05 | Mixed |
| use adaptive contrast weight instead of fixed (ablation) | PCCD · MIMIC-Diff-VQA | AUC: 88.35 -> 89.59 | Improved |
| add per-token branch selector over always-OA (ablation) | PCCD · MIMIC-Diff-VQA | AUC: 88.30 -> 89.59 | Improved |
| add per-token branch selector over always-LD (ablation) | PCCD · MIMIC-Diff-VQA | AUC: 88.00 -> 89.59 | Improved |
| add attribute lexicon routing (ablation) | PCCD · Gut-VLM description | QAAS: -1.40 QAAS when removed | Improved |
| apply PCCD vs greedy under poor attribution (IoU<0.2) | PCCD · MIMIC-Diff-VQA | delta AUC: approximately 0 vs AGLA -1.9 pp | No clear effect |
| apply PCCD vs VASparse on Med-POPE adversarial | PCCD · Med-POPE (Adversarial) | F1: 77.4 -> 79.7 (+2.3 pp) | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- VASparse — VASparse: Towards Efficient Visual Hallucination Mitigation via Visual-Aware Token Sparsification (CVPR 2025) · better on 14 of 14 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Anatomical Region-Guided Contrastive Decoding: A Plug-and-Play Strategy for Mitigating Hallucinations in Medical VLMs (AAAI 2026 · similarity 0.92)
- Mitigating Object Hallucinations in Large Vision-Language Models through Visual Contrastive Decoding (CVPR 2024 · similarity 0.90)
- HulluEdit: Single-Pass Evidence-Consistent Subspace Editing for Mitigating Hallucinations in Large Vision-Language Models (CVPR 2026 · similarity 0.89)
- SECOND: Mitigating Perceptual Hallucination in Vision-Language Models via Selective and Contrastive Decoding (ICML 2025 · similarity 0.89)
- ConVis: Contrastive Decoding with Hallucination Visualization for Mitigating Hallucinations in Multimodal Large Language Models (AAAI 2025 · similarity 0.89)
- Mitigating Object Hallucination in Large Vision-Language Models via False Discovery Controlled Visual Data Splitting (NeurIPS 2026 · similarity 0.89)
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.
