Self-Calibrated Tuning of Vision-Language Models for Out-of-Distribution Detection

TL;DR

This paper introduces Self-Calibrated Tuning (SCT), a novel framework that improves out-of-distribution detection in vision-language models by adaptively balancing regularization and calibration during prompt tuning with limited ID data.

Contribution

The paper proposes SCT, a new method that dynamically calibrates OOD regularization in prompt tuning, enhancing detection performance with minimal ID data.

Findings

SCT improves OOD detection accuracy across multiple benchmarks.

The method effectively balances regularization and calibration during training.

Extensive experiments validate the robustness of SCT in various scenarios.

Abstract

Out-of-distribution (OOD) detection is crucial for deploying reliable machine learning models in open-world applications. Recent advances in CLIP-based OOD detection have shown promising results via regularizing prompt tuning with OOD features extracted from ID data. However, the irrelevant context mined from ID data can be spurious due to the inaccurate foreground-background decomposition, thus limiting the OOD detection performance. In this work, we propose a novel framework, namely, Self-Calibrated Tuning (SCT), to mitigate this problem for effective OOD detection with only the given few-shot ID data. Specifically, SCT introduces modulating factors respectively on the two components of the original learning objective. It adaptively directs the optimization process between the two tasks during training on data with different prediction uncertainty to calibrate the influence of OOD regularization, which is compatible with many prompt tuning based OOD detection methods. Extensive experiments and analyses have been conducted to characterize and demonstrate the effectiveness of the proposed SCT. The code is publicly available.