LIDAR: Lightweight Adaptive Cue-Aware Fusion Vision Mamba for Multimodal Segmentation of Structural Cracks

TL;DR

LIDAR is a lightweight, adaptive multimodal segmentation network that effectively combines morphological and textural cues for pixel-level crack detection with low computational cost.

Contribution

The paper introduces LIDAR, a novel fusion network with adaptive cue modeling and efficient modules, advancing crack segmentation performance and efficiency.

Findings

Outperforms state-of-the-art methods on three datasets.

Achieves high F1 and mIoU scores with minimal parameters.

Demonstrates real-time applicability due to low computational overhead.

Abstract

Achieving pixel-level segmentation with low computational cost using multimodal data remains a key challenge in crack segmentation tasks. Existing methods lack the capability for adaptive perception and efficient interactive fusion of cross-modal features. To address these challenges, we propose a Lightweight Adaptive Cue-Aware Vision Mamba network (LIDAR), which efficiently perceives and integrates morphological and textural cues from different modalities under multimodal crack scenarios, generating clear pixel-level crack segmentation maps. Specifically, LIDAR is composed of a Lightweight Adaptive Cue-Aware Visual State Space module (LacaVSS) and a Lightweight Dual Domain Dynamic Collaborative Fusion module (LD3CF). LacaVSS adaptively models crack cues through the proposed mask-guided Efficient Dynamic Guided Scanning Strategy (EDG-SS), while LD3CF leverages an Adaptive Frequency Domain Perceptron (AFDP) and a dual-pooling fusion strategy to effectively capture spatial and frequency-domain cues across modalities. Moreover, we design a Lightweight Dynamically Modulated Multi-Kernel convolution (LDMK) to perceive complex morphological structures with minimal computational overhead, replacing most convolutional operations in LIDAR. Experiments on three datasets demonstrate that our method outperforms other state-of-the-art (SOTA) methods. On the light-field depth dataset, our method achieves 0.8204 in F1 and 0.8465 in mIoU with only 5.35M parameters. Code and datasets are available at https://github.com/Karl1109/LIDAR-Mamba.