HRP4K: High-Resolution Road Pothole Detection Dataset

Unofficial redistribution of the HRP4K road pothole-detection dataset (V1.00, Zenodo), under the original CC BY 4.0 license, with a documented upstream train-split completeness gap.

Dataset Description

• Homepage: https://github.com/hanshenChen/HRP4K
• Paper: https://doi.org/10.1038/s41597-026-07317-w

Disclaimer

This repository is not an official release of the HRP4K dataset.

HRP4K was created by Hanshen Chen, Zhoulin Tu, Yu Zhao, and Jianfeng Ye, who retain all copyright and intellectual property rights (to the extent applicable under the dataset's CC BY 4.0 license — see License below). This repository does not claim ownership of any images, annotations, or metadata.

This repository exists for two purposes:

1. To document and work around a known completeness gap in the official V1.00 archive's train split (see Changes from the Official Release below) so downstream users don't hit silent file-not-found errors or orphan annotations.
2. To provide a more reliable download source, as the original Zenodo hosting may be slow or difficult to access.

This redistribution is sourced directly from the official Zenodo release (doi:10.5281/zenodo.17522874), which already ships both YOLO .txt labels and COCO JSON annotations.

---

Dataset Overview

HRP4K is a perspective-view road pothole-detection benchmark: high-resolution road images captured for automated road-condition assessment, annotated for a single pothole class. It targets the same real-world problem as RDD2022's pothole class, but is pothole-specific and captured at much higher image resolution.

The paper and Zenodo metadata describe the dataset as 6,003 images / 7,217 pothole instances. This repository's own direct measurement of the official archive found otherwise for the train split — see below.

---

Changes from the Official Release

Unlike most redistributions in this project, the change here is not a format conversion (the official release already ships both YOLO .txt labels and COCO JSON) — it is a documented data-completeness filter:

• Verified against Zenodo's own file listing. The downloaded HRP4K.zip is byte-for-byte the exact size published in the Zenodo record (8,659,158,643 bytes) — this is not a corrupted or partial download.
• train.json (and the matching YOLO .txt labels) reference 4,203 unique images (contiguous ids 0–4202), but the archive itself only actually contains 2,286 of the referenced train image files (contiguous ids 0–2285). The remaining 1,917 train images are simply absent from the V1.00 zip. valid (900/900) and test (900/900) are complete — this gap is specific to train.
• This repository filters every split down to images that actually exist, dropping orphan annotations for the missing train images rather than carrying through broken references. No boxes were resized, moved, or reinterpreted for the images that do exist.
• Real, verified counts after filtering: train 2,286 images / 2,790 boxes, valid 900 images / 1,037 boxes, test 900 images / 921 boxes (4,086 images / 4,748 boxes total — smaller than the paper's headline 6,003 images / 7,217 instances, which appears to describe the full intended release rather than what the V1.00 archive actually ships).
• No image pixel content was modified. No splits were renamed or reshuffled.

---

Dataset Structure

<repo>/
├── README.md
├── hrp4k_banner.jpg
└── data/
    ├── data.yaml
    ├── images/
    │   ├── train/   (*.jpg + metadata.jsonl)
    │   ├── valid/   (*.jpg + metadata.jsonl)
    │   └── test/    (*.jpg + metadata.jsonl)
    └── labels/
        ├── train/   (*.txt, mirrors images)
        ├── valid/
        └── test/

where:

• data/images/<split>/ contains the high-resolution RGB road images, plus a metadata.jsonl (file_name and objects.bbox as absolute-pixel COCO [x, y, w, h] with objects.categories) that drives the Hugging Face dataset viewer.
• data/labels/<split>/ contains one YOLO-format .txt annotation file per image (class x_center y_center width height, normalized), mirroring the image layout.
• data/data.yaml is the Ultralytics dataset configuration file (class names, split paths, relative to data/).
• Splits (as verified in this repository, after filtering to images present on disk): train 2,286 images / 2,790 boxes · valid 900 images / 1,037 boxes · test 900 images / 921 boxes (4,086 images / 4,748 boxes total).

Classes (1)

pothole — single-class detection; images can contain from zero to 28 annotated potholes (mean 1.16 per image).

---

Dataset Sources

Original Paper

A high-resolution perspective-view road image dataset for pothole detection

Hanshen Chen, Zhoulin Tu, Yu Zhao, Jianfeng Ye

Scientific Data, volume 13, article 961, 2026. DOI: 10.1038/s41597-026-07317-w

Official Resources

• Official Repository: https://github.com/hanshenChen/HRP4K
• Zenodo Record: https://doi.org/10.5281/zenodo.17522874

---

Attribution

All credit for collecting and annotating this dataset belongs entirely to the original HRP4K authors: Hanshen Chen, Zhoulin Tu, Yu Zhao, and Jianfeng Ye.

This repository only filters the official release down to images verifiably present in the archive, for improved usability. It does not modify, reinterpret, or take credit for the underlying imagery or annotations.

If you use this dataset in your research, please cite the original publication below.

---

License

HRP4K is released by its creators under Creative Commons Attribution 4.0 International (CC BY 4.0), as stated on the official Zenodo record.

Accordingly:

• Attribution to the original authors is required.
• Commercial and non-commercial use are both permitted.
• No share-alike obligation (derivative works are not required to use the same license).

This repository is distributed under the same CC BY 4.0 license.

---

Citation

If you use this dataset, please cite:

@article{chen2026hrp4k,
  title={A high-resolution perspective-view road image dataset for pothole detection},
  author={Chen, Hanshen and Tu, Zhoulin and Zhao, Yu and Ye, Jianfeng},
  journal={Scientific Data},
  volume={13},
  pages={961},
  year={2026},
  doi={10.1038/s41597-026-07317-w}
}

---

Acknowledgements

We sincerely thank Hanshen Chen, Zhoulin Tu, Yu Zhao, and Jianfeng Ye for creating and publicly releasing this valuable road-infrastructure benchmark.