SoC: Semantic Orthogonal Calibration for Test-Time Prompt Tuning

TL;DR

This paper introduces Semantic Orthogonal Calibration (SoC), a novel regularizer for test-time prompt tuning in vision-language models that improves uncertainty calibration by balancing semantic proximity and class separability.

Contribution

The work proposes SoC, a new regularizer that enhances calibration in VLMs by enforcing smooth semantic separation, addressing limitations of fully orthogonal constraints.

Findings

SoC improves calibration performance across multiple benchmarks.

It maintains competitive discriminative accuracy.

Theoretical analysis explains the overconfidence issue with full orthogonality.

Abstract

With the increasing adoption of vision-language models (VLMs) in critical decision-making systems such as healthcare or autonomous driving, the calibration of their uncertainty estimates becomes paramount. Yet, this dimension has been largely underexplored in the VLM test-time prompt-tuning (TPT) literature, which has predominantly focused on improving their discriminative performance. Recent state-of-the-art advocates for enforcing full orthogonality over pairs of text prompt embeddings to enhance separability, and therefore calibration. Nevertheless, as we theoretically show in this work, the inherent gradients from fully orthogonal constraints will strongly push semantically related classes away, ultimately making the model overconfident. Based on our findings, we propose Semantic Orthogonal Calibration (SoC), a Huber-based regularizer that enforces smooth prototype separation while preserving semantic proximity, thereby improving calibration compared to prior orthogonality-based approaches. Across a comprehensive empirical validation, we demonstrate that SoC consistently improves calibration performance, while also maintaining competitive discriminative capabilities.