Object-Level Verbalized Confidence Calibration in Vision-Language Models via Semantic Perturbation

2025-04-21

AI Summary

Vision-language models (VLMs) suffer from poor calibration in object-level queries, where verbalized confidence scores fail to reflect actual correctness—undermining reliability. To address this, we propose an object-region–oriented semantic perturbation strategy: key object regions are perturbed with Gaussian noise to explicitly model the quantitative relationship between visual uncertainty and confidence levels. We further introduce a two-stage calibration paradigm: (1) supervised fine-tuning to learn the confidence–correctness mapping, followed by (2) PPO-based optimization of the output distribution using human feedback. Our method reduces Expected Calibration Error (ECE) by up to 42% across multiple benchmarks while preserving or improving task accuracy, thereby significantly enhancing VLM calibration, trustworthiness, and interpretability. To our knowledge, this is the first work to employ fine-grained, object-centric visual perturbations for confidence calibration in VLMs.

Technology Category

• Machine Learning: Calibration & Uncertainty Quantification
• Computer Vision: Large Vision Models
• Reasoning under Uncertainty: Uncertainty Representations

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems

Abstract

Vision-language models (VLMs) excel in various multimodal tasks but frequently suffer from poor calibration, resulting in misalignment between their verbalized confidence and response correctness. This miscalibration undermines user trust, especially when models confidently provide incorrect or fabricated information. In this work, we propose a novel Confidence Calibration through Semantic Perturbation (CSP) framework to improve the calibration of verbalized confidence for VLMs in response to object-centric queries. We first introduce a perturbed dataset where Gaussian noise is applied to the key object regions to simulate visual uncertainty at different confidence levels, establishing an explicit mapping between visual ambiguity and confidence levels. We further enhance calibration through a two-stage training process combining supervised fine-tuning on the perturbed dataset with subsequent preference optimization. Extensive experiments on popular benchmarks demonstrate that our method significantly improves the alignment between verbalized confidence and response correctness while maintaining or enhancing overall task performance. These results highlight the potential of semantic perturbation as a practical tool for improving the reliability and interpretability of VLMs.

Problem

• Improving verbalized confidence calibration in vision-language models
• Addressing misalignment between confidence and response correctness
• Enhancing reliability and interpretability of object-centric queries

Innovation

• Semantic perturbation maps visual ambiguity to confidence
• Two-stage training combines fine-tuning and preference optimization
• Gaussian noise simulates uncertainty for object-centric queries