Understanding and Mitigating Miscalibration in Prompt Tuning for Vision-Language Models

TL;DR

This paper investigates calibration issues in vision-language models like CLIP after prompt tuning, identifies causes of overconfidence and underconfidence, and proposes a new method called Dynamic Outlier Regularization (DOR) to improve confidence calibration across classes.

Contribution

The paper introduces DOR, a novel regularization technique that improves calibration in vision-language models by balancing confidence levels for base and new classes.

Findings

DOR improves calibration performance on both base and new classes.

Prompt tuning methods exhibit trade-offs between overconfidence and underconfidence.

Extensive experiments validate the effectiveness of DOR in real-world scenarios.

Abstract

Confidence calibration is critical for the safe deployment of machine learning models in the real world. However, such issue in vision-language models like CLIP, particularly after fine-tuning, has not been fully addressed. In this work, we demonstrate that existing prompt tuning methods usually lead to a trade-off of calibration between base and new classes: the cross-entropy loss in CoOp causes overconfidence in new classes by increasing textual label divergence, whereas the regularization of KgCoOp maintains the confidence level but results in underconfidence in base classes due to the improved accuracy. Inspired by the observations, we introduce Dynamic Outlier Regularization (DOR) to ensure the confidence calibration on both base and new classes after fine-tuning. In particular, we propose to minimize the feature deviation of novel textual labels (instead of base classes) sampled from a large vocabulary. In effect, DOR prevents the increase in textual divergence for new labels while easing restrictions on base classes. Extensive experiments demonstrate that DOR can enhance the calibration performance of current fine-tuning methods on base and new classes.