A preference optimization strategy using confidence-based hard example mining, similarity retrieval, and synthetic counterfactual rationales improves chest X-ray VQA accuracy by 8.93% relative over supervised fine-tuning.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
CheXPO: Preference Optimization for Chest X-ray VLMs with Counterfactual Rationale
A preference optimization strategy using confidence-based hard example mining, similarity retrieval, and synthetic counterfactual rationales improves chest X-ray VQA accuracy by 8.93% relative over supervised fine-tuning.