TD-RD: A Top-Down Benchmark with Real-Time Framework for Road Damage Detection

Xi Xiao, Zhengji Li, Wentao Wang, Jiacheng Xie, Houjie Lin, Swalpa Kumar Roy, Tianyang Wang, Min Xu

Abstract

Object detection has witnessed remarkable advancements over the past decade, largely driven by breakthroughs in deep learning and the proliferation of large-scale datasets. However, the domain of road damage detection remains relatively under-explored, despite its critical significance for applications such as infrastructure maintenance and road safety. This paper addresses this gap by introducing a novel top-down benchmark that offers a complementary perspective to existing datasets, specifically tailored for road damage detection. Our proposed Top-Down Road Damage Detection Dataset (TD-RD) includes three primary categories of road damage—cracks, potholes, and patches—captured from a top-down viewpoint. The dataset consists of 7,088 high-resolution images, encompassing 12,882 annotated instances of road damage. Additionally, we present a novel real-time object detection framework, TD-YOLOv10, designed to handle the unique challenges posed by the TD-RD dataset. Comparative studies with state-of-the-art models demonstrate competitive baseline results. By releasing TD-RD, we aim to accelerate research in this crucial area. A sample of the dataset will be made publicly available upon the paper's acceptance.

Introduction

Road maintenance is a critical infrastructure challenge, requiring the efficient identification of hazards like potholes and cracks across vast networks of asphalt. While computer vision has significantly improved automated detection, most existing models rely on images captured from vehicle dashboards. This oblique perspective often distorts the size and shape of damage, complicating accurate engineering assessments.

In "TD-RD: A Top-Down Benchmark with Real-Time Framework for Road Damage Detection," researchers propose a shift to a top-down (nadir) perspective. By introducing a specialized dataset and a high-performance deep learning architecture, the authors aim to provide a more accurate "map-like" representation of road conditions suitable for professional inspection.

The Challenge of Perspective

Most road damage datasets, such as CRDDC or CNRDD, utilize dashcam imagery. While accessible, this viewpoint suffers from perspective distortion, where damage further from the vehicle appears compressed and smaller than its actual dimensions.

Traditional dashcam perspective showing perspective distortion in road damage detection.

A top-down view is superior for modeling road surface conditions because it allows for direct measurement of a pothole's area or a crack's length. However, top-down data is historically scarce, and standard object detectors often struggle with the fine-grained, multi-scale textures found on pavement.

The TD-RD Dataset

The primary contribution of this work is the TD-RD dataset, the first high-resolution, top-down benchmark for road damage. It contains 7,088 images at 3840 × 2160 resolution, featuring 12,882 annotated instances. Unlike many datasets skewed toward cracks, TD-RD is curated to be balanced across three essential categories: cracks, potholes, and patches (repairs).

Annotated samples from the TD-RD dataset illustrating the top-down perspective for cracks, potholes, and repairs.

TD-YOLOv10: A Real-Time Framework

To process this data, the researchers developed TD-YOLOv10, a framework optimized for the unique textures of asphalt. The architecture introduces several key innovations:

  1. MAPSE (Multi-Scale Attention with Positional Squeeze-and-Excitation): Integrated into the backbone, this module uses multi-head attention with varying dilation rates (r=1, 2, 4). This allows the model to detect both hairline cracks and large potholes simultaneously by capturing features at different scales.
  2. VGAU (Visual Global Attention Upsampling): This module incorporates global average pooling during upsampling in the model's "neck." This helps the detector distinguish between actual damage and harmless road textures or shadows by maintaining semantic context.
  3. Efficiency via DSC: To ensure real-time performance, the model utilizes Depthwise Separable Convolutions (DSC), which significantly reduces parameter count and computational load.

Results and Performance

TD-YOLOv10 achieves a superior balance between accuracy and speed compared to standard YOLO variants:

  • Accuracy: The model achieved 88.1% mAP (Mean Average Precision) on the TD-RD dataset, outperforming YOLOv8-s by 3.0% and the state-of-the-art RT-DETR by 0.4%.
  • Speed: It sustains a frame rate of 200 FPS on an Nvidia A100 GPU, making it ideal for deployment on high-speed inspection vehicles.
  • Robustness: When tested on the dashcam-based CRDDC’22 dataset, it achieved 48.0% mAP, demonstrating strong generalization capabilities across different camera angles.

Conclusion

The TD-RD benchmark and TD-YOLOv10 framework represent a significant step forward in automated road maintenance. By prioritizing a top-down perspective and multi-scale attention mechanisms, the authors provide a more reliable method for measuring and monitoring road distress. Future work will focus on optimizing performance for edge devices and exploring multi-view fusion to combine the strengths of both dashcam and top-down perspectives.

Reference: Xiao, X., et al. "TD-RD: A Top-Down Benchmark with Real-Time Framework for Road Damage Detection." (2024).

Summary updated

Abstract

Object detection has witnessed remarkable advancements over the past decade, largely driven by breakthroughs in deep learning and the proliferation of large scale datasets. However, the domain of road damage detection remains relatively under explored, despite its critical significance for applications such as infrastructure maintenance and road safety. This paper addresses this gap by introducing a novel top down benchmark that offers a complementary perspective to existing datasets, specifically tailored for road damage detection. Our proposed Top Down Road Damage Detection Dataset (TDRD) includes three primary categories of road damage cracks, potholes, and patches captured from a top down viewpoint. The dataset consists of 7,088 high resolution images, encompassing 12,882 annotated instances of road damage. Additionally, we present a novel real time object detection framework, TDYOLOV10, designed to handle the unique challenges posed by the TDRD dataset. Comparative studies with state of the art models demonstrate competitive baseline results. By releasing TDRD, we aim to accelerate research in this crucial area. A sample of the dataset will be made publicly available upon the paper's acceptance.
BibTeX
@inproceedings{xiao2025td,
  author = {Xi Xiao and Zhengji Li and Wentao Wang and Jiacheng Xie and Houjie Lin and Swalpa Kumar Roy and Tianyang Wang and Min Xu},
  title = {{TD-RD: A Top-Down Benchmark with Real-Time Framework for Road Damage Detection}},
  booktitle = {IEEE International Conference on Acoustics, Speech and Signal Processing},
  year = {2025},
  url = {https://openreview.net/forum?id=vygpHQu3Hk},
}
Figure