HRP4K

This repository contains the official code for the paper: "A high-resolution perspective-view road image dataset for pothole detection". The dataset is publicly available on Zenodo.

📖 Overview

HRP4K is a high-resolution, perspective-view pothole detection dataset designed to advance automated infrastructure monitoring and computer-vision-based road-surface analysis.
The dataset provides 6,003 4K-resolution images containing 7,217 annotated pothole instances captured from real-world driving scenes across 1,100 km of urban and rural roads in Hangzhou, Huzhou, and Jiaxing, China. Each pothole is annotated with a bounding box in both YOLO and COCO formats, enabling seamless integration with major deep-learning pipelines.

The dataset described in our paper will be released soon.

---

Dataset Highlights

• 📸 High-resolution imagery (4K) captured using mirrorless vehicle-mounted cameras (Sony Alpha A7IV and Alpha 9III)
• 🤖 Human-in-the-loop annotation pipeline combining AI-assisted pre-labeling and expert verification
• 🔒 Privacy-preserving anonymization for faces, license plates, and traffic signs
• 📂 Standardized data formats: YOLO .txt and COCO .json

---

🗂️ Dataset Structure

HRP4K/
├── train/
│ ├── images/
│ └── labels/
│ └── annotations/
├── valid/
│ ├── images/
│ └── labels/
│ └── annotations/
├── test/
│ ├── images/
│ └── labels/
│ └── annotations/

📝 1. Frame Extraction extract frames from the recorded 4K videos at 3 frames per second:
python extract_frames.py

🧩 2. Privacy Anonymization

Step1:Automatic Masking (YOLOv11) Automatically detects and masks faces and license plates using a YOLOv11-based detector：
python auto_privacy_detection_anonymization.py
Step2: LabelMe Manual annotation.
Step3: Manual Anonymization：manually anonymizing traffic signs, faces, and license plates：
python manual_plate_sign_face_anonymization.py

🧠 3. Model-Assisted Pre-Annotation

Semi-automated pre-annotation using YOLOv11 predictions to assist human labeling:
python pre-annotation.py

🧠 4. Format Conversion

Convert annotations between LabelMe, YOLO, and COCO formats. For specific conversion scripts, please refer to the following resources:
https://github.com/rooneysh/Labelme2YOLO; https://github.com/Tony607/labelme2coco