Mind the Way You Select Negative Texts: Pursuing the Distance Consistency in OOD Detection with VLMs

TL;DR

This paper introduces InterNeg, a framework that enhances OOD detection with VLMs by enforcing inter-modal distance consistency through negative text selection and image inversion, achieving state-of-the-art results.

Contribution

It systematically aligns intra- and inter-modal distances in VLMs for improved OOD detection, a novel approach compared to prior methods.

Findings

Achieves 3.47% lower FPR95 on ImageNet

Improves AUROC by 5.50% on Near-OOD benchmark

Outperforms existing methods across multiple benchmarks

Abstract

Out-of-distribution (OOD) detection seeks to identify samples from unknown classes, a critical capability for deploying machine learning models in open-world scenarios. Recent research has demonstrated that Vision-Language Models (VLMs) can effectively leverage their multi-modal representations for OOD detection. However, current methods often incorporate intra-modal distance during OOD detection, such as comparing negative texts with ID labels or comparing test images with image proxies. This design paradigm creates an inherent inconsistency against the inter-modal distance that CLIP-like VLMs are optimized for, potentially leading to suboptimal performance. To address this limitation, we propose InterNeg, a simple yet effective framework that systematically utilizes consistent inter-modal distance enhancement from textual and visual perspectives. From the textual perspective, we devise an inter-modal criterion for selecting negative texts. From the visual perspective, we dynamically identify high-confidence OOD images and invert them into the textual space, generating extra negative text embeddings guided by inter-modal distance. Extensive experiments across multiple benchmarks demonstrate the superiority of our approach. Notably, our InterNeg achieves state-of-the-art performance compared to existing works, with a 3.47% reduction in FPR95 on the large-scale ImageNet benchmark and a 5.50% improvement in AUROC on the challenging Near-OOD benchmark.