H-MBA: Hierarchical MamBa Adaptation for Multi-Modal Video Understanding in Autonomous Driving

TL;DR

This paper introduces H-MBA, a hierarchical adaptation framework for multi-modal video understanding in autonomous driving, improving model generalization to complex spatial-temporal scenes by capturing multi-scale context.

Contribution

The novel H-MBA framework with Context and Query Mamba modules effectively captures multi-granularity temporal context, enhancing multi-modal video understanding in autonomous driving.

Findings

Outperforms previous SOTA with 5.5% mIoU improvement in risk object detection

Effectively captures multi-scale temporal context in complex driving scenes

Demonstrates robustness across various multi-modal video tasks

Abstract

With the prevalence of Multimodal Large Language Models(MLLMs), autonomous driving has encountered new opportunities and challenges. In particular, multi-modal video understanding is critical to interactively analyze what will happen in the procedure of autonomous driving. However, videos in such a dynamical scene that often contains complex spatial-temporal movements, which restricts the generalization capacity of the existing MLLMs in this field. To bridge the gap, we propose a novel Hierarchical Mamba Adaptation (H-MBA) framework to fit the complicated motion changes in autonomous driving videos. Specifically, our H-MBA consists of two distinct modules, including Context Mamba (C-Mamba) and Query Mamba (Q-Mamba). First, C-Mamba contains various types of structure state space models, which can effectively capture multi-granularity video context for different temporal resolutions. Second, Q-Mamba flexibly transforms the current frame as the learnable query, and attentively selects multi-granularity video context into query. Consequently, it can adaptively integrate all the video contexts of multi-scale temporal resolutions to enhance video understanding. Via a plug-and-play paradigm in MLLMs, our H-MBA shows the remarkable performance on multi-modal video tasks in autonomous driving, e.g., for risk object detection, it outperforms the previous SOTA method with 5.5% mIoU improvement.