HighRPD in YOLO format
HighRPD, a road pavement distress dataset of drone image tiles, in YOLO format.
Original dataset: https://doi.org/10.17632/sywswj7djj.1. Dataset paper: https://doi.org/10.1016/j.dib.2025.111377. Created by Jin He, Liting Gong, Chuan Xu, Pin Wang, Yiyong Zhang, Ou Zheng, Guanghe Su, Yufeng Yang, Jialin Hu and Yuchen Sun.
| View | Aerial, drone, 50 m above the ground |
| Place | One road segment in Shanxi Province, China |
| Capture | 8192 x 5460 drone photos cut into 640 x 640 tiles |
| Image size | 640 x 640 pixels |
| Images | 11,696 (train 8,335, val 1,563, test 1,798) |
| Boxes | 22,016 |
| Configuration | data.yaml |
| Number | Class | Meaning | Train | Val | Test | Total |
|---|---|---|---|---|---|---|
| 0 | line | Linear crack | 8,734 | 1,687 | 1,944 | 12,365 |
| 1 | block | Block crack (dense cracks) | 6,182 | 928 | 1,129 | 8,239 |
| 2 | pit | Pothole | 993 | 214 | 205 | 1,412 |
Use with Ultralytics
from huggingface_hub import snapshot_download
from ultralytics import YOLO
folder = snapshot_download("ashmeera/highrpd-yolo", repo_type="dataset")
YOLO("yolo26n.pt").train(data=f"{folder}/data.yaml")
Each data.yaml points to its images/train, images/val and images/test folders relative to its own location. Each image has a label file of the same name in the matching labels folder, one line per box: class number, x centre, y centre, width and height, all as fractions of the image size.
Split
HighRPD ships without a split. All tiles of one photo go into the same split, 70/15/15 by photo, with a fixed random seed (42). File names point to 578 source photos; the dataset paper reports 546.
Changes from the original
- Class names added from the dataset paper: 0 line, 1 block, 2 pit.
- New train, val and test splits, as described above.
- Boxes that extend past the image edge are cut at the edge. This affects 17 boxes, which cross the edge of their tile.
- Label lines are written with six decimals.
Limitations
- Models trained on it see one road segment only, as the dataset paper reports.
Licence
Shared under CC BY 4.0. The Mendeley Data record lists CC BY 4.0. The original creators provide the data as it is, with no warranty. Please credit the creators and cite the works below.
Citation
@article{He_2025,
title = {HighRPD: A high-altitude drone dataset of road pavement distress},
volume = {59},
journal = {Data in Brief},
publisher = {Elsevier BV},
author = {He, Jin and Gong, Liting and Xu, Chuan and Wang, Pin and Zhang, Yiyong and Zheng, Ou and Su, Guanghe and Yang, Yufeng and Hu, Jialin and Sun, Yuchen},
year = {2025},
pages = {111377},
doi = {10.1016/j.dib.2025.111377}
}
@misc{He_2024_data,
title = {HighRPD},
publisher = {Mendeley Data},
author = {He, Jin and Gong, Liting and Xu, Chuan and Wang, Pin and Zhang, Yiyong and Zheng, Ou and Su, Guanghe and Yang, Yufeng and Hu, Jialin and Sun, Yuchen},
year = {2024},
doi = {10.17632/sywswj7djj.1}
}
- Downloads last month
- 62

