Pothole-Finetuned-YOLOv8

A fine-tuned YOLOv8 model for pothole detection in road images. This model is optimized for detecting potholes with high precision and is production-ready.

Model Details

• Architecture: YOLOv8 Medium
• Task: Object Detection
• Framework: PyTorch
• Input Size: 640x640
• Number of Classes: 1 (Pothole)
• Confidence Threshold: 0.28
• IOU Threshold: 0.5

Training Configuration

• Epochs: 40
• Batch Size: 64
• Optimizer: Auto (SGD/Adam)
• Patience: 200
• Augmentation: Enabled
• Mixed Precision (AMP): Enabled
• Training Device: GPU/CPU

Performance

This model was trained on a comprehensive pothole detection dataset with data augmentation and regularization techniques applied.

Usage

Installation

pip install ultralytics opencv-python pillow

Quick Start

from ultralytics import YOLO
from PIL import Image

# Load model from Hugging Face
model = YOLO('hf://your-username/Pothole-Finetuned-YOLOv8')

# Inference on image
results = model.predict(source='image.jpg', conf=0.28)

# Print results
for r in results:
    print(r.boxes)

Advanced Usage

from ultralytics import YOLO
import cv2

# Load model
model = YOLO('hf://your-username/Pothole-Finetuned-YOLOv8')

# Inference on video
results = model.predict(source='video.mp4', conf=0.28, save=True)

# Custom confidence threshold
results = model.predict(source='image.jpg', conf=0.35)

# Get results in different formats
for result in results:
    boxes = result.boxes.xyxy  # Box coordinates
    confidences = result.boxes.conf  # Confidence scores
    class_ids = result.boxes.cls  # Class IDs

Batch Inference

from ultralytics import YOLO

model = YOLO('hf://your-username/Pothole-Finetuned-YOLOv8')

# Batch inference
results = model.predict(source='image_folder/', batch=32, conf=0.28)

# Process results
for result in results:
    for box in result.boxes:
        print(f"Detection: {box.xyxy}, Confidence: {box.conf}")

Annotation/Training on Custom Data

You can fine-tune this model on your own pothole dataset:

from ultralytics import YOLO

# Load pre-trained model
model = YOLO('hf://your-username/Pothole-Finetuned-YOLOv8')

# Train on custom dataset
results = model.train(
    data='path/to/dataset/data.yaml',
    epochs=50,
    imgsz=640,
    batch=32,
    device=0,  # GPU device number
    patience=50
)

Dataset Format

Your dataset should follow YOLO format:

dataset/
├── images/
│   ├── train/
│   │   └── *.jpg
│   ├── val/
│   │   └── *.jpg
│   └── test/
│       └── *.jpg
├── labels/
│   ├── train/
│   │   └── *.txt
│   ├── val/
│   │   └── *.txt
│   └── test/
│       └── *.txt
└── data.yaml

data.yaml Example

path: /path/to/dataset
train: images/train
val: images/val
test: images/test

nc: 1  # number of classes
names: ['pothole']  # class names

Model Information

• Base Model: YOLOv8m
• Training Dataset: Comprehensive pothole detection dataset
• Output: Bounding box coordinates with confidence scores

Hardware Requirements

• Minimum RAM: 4GB
• Minimum VRAM: 2GB (GPU optional)
• Recommended: GPU with 6GB+ VRAM for faster inference

License

This model and its weights are provided as-is for research and commercial use.

Support & Feedback

For issues, questions, or feedback about this model, please create an issue in the repository.

Citation

If you use this model in your research, please cite:

@software{pothole_yolov8_2024,
  title={Pothole Detection YOLOv8 Fine-tuned Model},
  author={Your Name},
  year={2024},
  publisher={Hugging Face},
  howpublished={\url{https://huggingface.co/your-username/Pothole-Finetuned-YOLOv8}}
}

---

Last Updated: January 2026