RoadFusion: Latent Diffusion Model for Pavement Defect Detection

2025-07-21

AI Summary

Road surface defect detection faces three key challenges: scarcity of annotated data, train-deployment domain shift, and high visual variability of defects. To address these, we propose RoadFusion—a novel latent diffusion model featuring dual-path feature adaptation, integrating text guidance and spatial mask control to synthesize diverse, high-fidelity defect samples. We further design a dual-branch adapter specialized for normal and anomalous features, coupled with a lightweight patch-level image discriminator, jointly enhancing domain robustness and fine-grained defect recognition. Evaluated on six benchmark datasets, RoadFusion achieves state-of-the-art performance in both classification and localization tasks, significantly outperforming prior methods across multiple metrics. Our approach advances reliable, real-world deployment of road defect detection systems.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Adversarial Learning & Robustness
• Natural Language Processing: Safety and Robustness

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Pavement defect detection faces critical challenges including limited annotated data, domain shift between training and deployment environments, and high variability in defect appearances across different road conditions. We propose RoadFusion, a framework that addresses these limitations through synthetic anomaly generation with dual-path feature adaptation. A latent diffusion model synthesizes diverse, realistic defects using text prompts and spatial masks, enabling effective training under data scarcity. Two separate feature adaptors specialize representations for normal and anomalous inputs, improving robustness to domain shift and defect variability. A lightweight discriminator learns to distinguish fine-grained defect patterns at the patch level. Evaluated on six benchmark datasets, RoadFusion achieves consistently strong performance across both classification and localization tasks, setting new state-of-the-art in multiple metrics relevant to real-world road inspection.

Problem

• Addresses limited annotated pavement defect data
• Overcomes domain shift in training and deployment
• Handles high variability in defect appearances

Innovation

• Latent diffusion model generates synthetic defects
• Dual-path feature adaptors enhance domain robustness
• Lightweight discriminator detects fine-grained defect patterns