Diverse Road Damage Dataset (DRDD)
收藏资源简介:
Diverse Road Damage Dataset (DRDD) 是由阿曼技术与应用科学大学等机构创建的一个道路损坏检测数据集,旨在解决现有数据集中缺乏多样化道路损坏类型的问题。该数据集包含1500张图像,分辨率为1920×1440像素,涵盖了5种不同的损坏类型,并在不同天气条件和时间段下采集。数据集的创建过程涉及12名具有专业背景的标注人员,经过多轮质量评估。DRDD的应用领域主要集中在道路基础设施维护和损坏检测,旨在通过提供多样化的损坏场景,提升道路损坏检测模型的鲁棒性和准确性。
Diverse Road Damage Dataset (DRDD) was developed by institutions including the University of Technology and Applied Sciences of Oman and other relevant organizations, aiming to address the shortage of diverse road damage types in existing datasets. This dataset contains 1500 images with a resolution of 1920x1440 pixels, covering 5 distinct types of road damage, and was collected under various weather conditions and different time periods. The development process involved 12 annotators with professional backgrounds, and the dataset underwent multiple rounds of quality assessment. The main application fields of DRDD focus on road infrastructure maintenance and damage detection, with the goal of enhancing the robustness and accuracy of road damage detection models by providing diverse damage scenarios.
数据集概述
数据集名称
ERDD: Efficient Road Damage Detection
数据集简介
该数据集用于道路损坏检测,采用深度学习框架进行高效检测。该方法在CSPNeXtPAFPN颈部引入了Attention4D模块,以改进多尺度特征细化,从而更好地检测各种类型的道路损坏。该方法在检测大型道路裂缝方面表现出色,同时保持了整体检测的竞争力。
数据集内容
- 数据集类型: 道路损坏检测数据集
- 数据集状态: 即将发布
数据集性能评估
- 评估指标: AP (平均精度), AP₅₀, AP₇₅, APₛ (小目标), APₘ (中目标), APₗ (大目标), AR (平均召回率), ARₛ, ARₘ, ARₗ
- 评估结果:
- YOLOV8: AP=0.122, AP₅₀=0.299, AP₇₅=0.082, APₛ=0.000, APₘ=0.083, APₗ=0.127, AR=0.448, ARₛ=0.000, ARₘ=0.234, ARₗ=0.454
- YOLOV7: AP=0.255, AP₅₀=0.498, AP₇₅=0.233, APₛ=0.000, APₘ=0.127, APₗ=0.263, AR=0.547, ARₛ=0.000, ARₘ=0.351, ARₗ=0.553
- YOLOV6: AP=0.110, AP₅₀=0.263, AP₇₅=0.095, APₛ=0.000, APₘ=0.108, APₗ=0.114, AR=0.560, ARₛ=0.000, ARₘ=0.460, ARₗ=0.572
- PPYOLOE: AP=0.112, AP₅₀=0.463, AP₇₅=0.062, APₛ=0.000, APₘ=0.079, APₗ=0.117, AR=0.322, ARₛ=0.000, ARₘ=0.388, ARₗ=0.325
- RTMDET: AP=0.268, AP₅₀=0.527, AP₇₅=0.229, APₛ=0.000, APₘ=0.123, APₗ=0.280, AR=0.517, ARₛ=0.000, ARₘ=0.373, ARₗ=0.623
- YOLOX: AP=0.200, AP₅₀=0.377, AP₇₅=0.188, APₛ=0.000, APₘ=0.006, APₗ=0.204, AR=0.288, ARₛ=0.000, ARₘ=0.033, ARₗ=0.386
- Ours: AP=0.446, AP₅₀=0.687, AP₇₅=0.451, APₛ=0.000, APₘ=0.113, APₗ=0.458, AR=0.675, ARₛ=0.000, ARₘ=0.277, ARₗ=0.690
数据集使用
- 训练: 使用
tools/train.py进行训练 - 测试: 使用
tools/test.py进行测试 - 演示: 使用
demo/image_demo.py进行图像演示
引用
如果该数据集对您的研究有帮助,请考虑引用以下文献: bibtex @article{saqib2024road, title={Road Damage Detection Using Attention4D Blocks}, author={Saqib, Muhammad and Author2, Name and Author3, Name}, journal={arXiv preprint arXiv:xxxx.xxxx}, year={2024} }

- 1RDD4D: 4D Attention-Guided Road Damage Detection And Classification阿曼技术与应用科学大学、澳大利亚联邦科学与工业研究组织、澳大利亚国立大学、印度理工学院帕特纳分校、FutureDataMinds · 2025年



