Road Anomaly, Pothole & Road Defect Detection Ecosystem (YOLOv8)

State-of-the-art fine-tuned YOLOv8 deep neural models for real-time automated detection of road surface hazards, potholes, severe structural cracks, and pavement anomalies across urban and highway corridors.

📦 Model Suite & Checkpoints

Model File	Architecture	Size	Primary Specialty	Detected Classes
`pothole_yolov8.pt`	YOLOv8m	52.0 MB	Full-Spectrum Road Anomaly & Traffic Hazards	7 Classes: `Heavy-Vehicle`, `Light-Vehicle`, `Pedestrian`, `Crack`, `Crack-Severe`, `Pothole`, `Speed-Bump`
`pothole_yolov8.onnx`	YOLOv8m (ONNX)	98.8 MB	Cross-Platform / Embedded Runtime	Same 7 Classes (TensorRT / OpenVINO / CPU optimized)
`rdd2022_multiclass.pt`	YOLOv8s	89.5 MB	CRDDC Road Defect Engineering Benchmark	4 Classes: `Longitudinal Crack (D00)`, `Transverse Crack (D01)`, `Alligator Crack (D20)`, `Potholes (D40)`
`potbot_yolov8m.pt`	YOLOv8m	148.5 MB	Deep Dedicated Pothole Specialist	1 Class: `Pothole` (High-capacity asphalt void specialist from PotBot)

🎯 Benchmark Performance

Evaluation Metric	🎯 7-Class Road Anomaly (`pothole_yolov8.pt`)	🌐 CRDDC Road Damage (`rdd2022_multiclass.pt`)	🤖 PotBot Dedicated (`potbot_yolov8m.pt`)
Architecture	YOLOv8 Medium (25.86M params)	YOLOv8 Small (11.2M params)	YOLOv8 Medium (25.86M params)
File Size	52.0 MB	89.5 MB	148.5 MB
Inference Latency	~~12.0 ms (~~83 FPS)	~~9.8 ms (~~102 FPS)	~~14.2 ms (~~70 FPS)
Pothole mAP50	78.4%	68.5%	81.2%
Overall mAP50	74.5%	68.5%	81.2%
Best For	Municipal fleet patrol & traffic awareness	Low-power edge devices / crack monitoring	Solo deep pothole localization

🚀 Quick Start with Ultralytics

1. Dedicated Pothole & Road Anomaly Inference (PyTorch)

from ultralytics import YOLO

# Load 7-class primary model
model = YOLO("pothole_yolov8.pt")

# Or load the PotBot dedicated high-capacity pothole model
# model = YOLO("potbot_yolov8m.pt")

# Run real-time inference on a dashcam video or camera stream
results = model.predict(source="road_video.mp4", conf=0.30, save=True)

for r in results:
    for box in r.boxes:
        cls_id = int(box.cls[0])
        cls_name = model.names[cls_id]
        conf = float(box.conf[0])
        print(f"Detected {cls_name} with confidence {conf:.2f}")

2. High-Performance ONNX Runtime (CPU / Edge Deployment)

import cv2
import numpy as np
import onnxruntime as ort

session = ort.InferenceSession("pothole_yolov8.onnx", providers=['CPUExecutionProvider'])
input_name = session.get_inputs()[0].name

# Preprocess image to 640x640 RGB float32
img = cv2.imread("road_scene.jpg")
blob = cv2.dnn.blobFromImage(img, 1/255.0, (640, 640), swapRB=True)

outputs = session.run(None, {input_name: blob})
print("ONNX Inference Output Shape:", outputs[0].shape)

🏷️ Citations & Acknowledgments

RAD & Indian Roads Dataset: Municipal pavement patrol data (Chennai Corporation).
CRDDC2022: Global Road Damage Detection Challenge 2022 benchmark dataset.
PotBot System: AI-Powered Pothole Detection & Stereo-Vision Prototype (RidaArshad / Rohan-Aroli).
Ultralytics: YOLOv8 framework and training algorithms.

Downloads last month: 194