Layer-Wise Modality Decomposition for Interpretable Multimodal Sensor Fusion

TL;DR

This paper introduces Layer-Wise Modality Decomposition (LMD), a novel interpretability method that disentangles and attributes sensor modality contributions in multimodal perception models for autonomous driving, enhancing transparency and safety.

Contribution

LMD is the first approach to attribute predictions to individual sensor modalities in a pretrained multimodal fusion model for autonomous driving.

Findings

LMD effectively disentangles modality-specific information across all layers.

It accurately attributes model predictions to input modalities.

Demonstrates practical interpretability in complex multimodal architectures.

Abstract

In autonomous driving, transparency in the decision-making of perception models is critical, as even a single misperception can be catastrophic. Yet with multi-sensor inputs, it is difficult to determine how each modality contributes to a prediction because sensor information becomes entangled within the fusion network. We introduce Layer-Wise Modality Decomposition (LMD), a post-hoc, model-agnostic interpretability method that disentangles modality-specific information across all layers of a pretrained fusion model. To our knowledge, LMD is the first approach to attribute the predictions of a perception model to individual input modalities in a sensor-fusion system for autonomous driving. We evaluate LMD on pretrained fusion models under camera-radar, camera-LiDAR, and camera-radar-LiDAR settings for autonomous driving. Its effectiveness is validated using structured perturbation-based metrics and modality-wise visual decompositions, demonstrating practical applicability to interpreting high-capacity multimodal architectures. Code is available at https://github.com/detxter-jvb/Layer-Wise-Modality-Decomposition.