# RoadFusion: Latent Diffusion Model for Pavement Defect Detection

arXiv: https://arxiv.org/abs/2507.15346 · HTML: https://mlsanity.com/paper/2507.15346/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Placed in Flow matching, time series and science, one of 14 directions in Diffusion models.
- Builds on 10 of its 30 references that are in the map.
- First posted on arXiv 21 Jul 2025.

## Placed in

- Diffusion models / [Flow matching, time series and science](https://mlsanity.com/diffusion/flow-matching-time-series-and-science/)

## Builds on (most cited first, 10 of 10)

- [Denoising Diffusion Probabilistic Models](https://mlsanity.com/paper/2006.11239/) (2020)
- [Towards Total Recall in Industrial Anomaly Detection](https://mlsanity.com/paper/2106.08265/) (2021)
- [Feature Pyramid and Hierarchical Boosting Network for Pavement Crack Detection](https://mlsanity.com/paper/1901.06340/) (2019)
- [Road Damage Detection Using Deep Neural Networks with Images Captured Through a Smartphone](https://mlsanity.com/paper/1801.09454/) (2018)
- [We Learn Better Road Pothole Detection: from Attention Aggregation to Adversarial Domain Adaptation](https://mlsanity.com/paper/2008.06840/) (2020)
- [Diffusion-based Image Generation for In-distribution Data Augmentation in Surface Defect Detection](https://mlsanity.com/paper/2406.00501/) (2024)
- [Leveraging Latent Diffusion Models for Training-Free In-Distribution Data Augmentation for Surface Defect Detection](https://mlsanity.com/paper/2407.03961/) (2024)
- [SHREC 2022: pothole and crack detection in the road pavement using images and RGB-D data](https://mlsanity.com/paper/2205.13326/) (2022)
- [Self-Supervised Iterative Refinement for Anomaly Detection in Industrial Quality Control](https://mlsanity.com/paper/2408.11561/) (2024)
- [Meta Learning-Driven Iterative Refinement for Robust Anomaly Detection in Industrial Inspection](https://mlsanity.com/paper/2503.01569/) (2025)

## Abstract (arXiv)

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.