Language-Enhanced Latent Representations for Out-of-Distribution Detection in Autonomous Driving

TL;DR

This paper introduces a language-enhanced approach for out-of-distribution detection in autonomous driving, leveraging multimodal models like CLIP to improve transparency and performance over traditional encoder-based methods.

Contribution

It proposes using cosine similarity of image and text representations from CLIP for more transparent and controllable OOD detection in autonomous driving.

Findings

Language-based representations outperform traditional vision encoder features.

Combining language and visual features improves detection accuracy.

The approach enhances transparency and user interpretability in OOD detection.

Abstract

Out-of-distribution (OOD) detection is essential in autonomous driving, to determine when learning-based components encounter unexpected inputs. Traditional detectors typically use encoder models with fixed settings, thus lacking effective human interaction capabilities. With the rise of large foundation models, multimodal inputs offer the possibility of taking human language as a latent representation, thus enabling language-defined OOD detection. In this paper, we use the cosine similarity of image and text representations encoded by the multimodal model CLIP as a new representation to improve the transparency and controllability of latent encodings used for visual anomaly detection. We compare our approach with existing pre-trained encoders that can only produce latent representations that are meaningless from the user's standpoint. Our experiments on realistic driving data show that the language-based latent representation performs better than the traditional representation of the vision encoder and helps improve the detection performance when combined with standard representations.