Pothole Detection (ONNX Model) Author: Darshan Modi This model is a lightweight and efficient object‑detection system designed to identify potholes and road‑surface defects in real‑world environments. The dataset includes diverse road textures, lighting conditions, camera angles, and weather variations, making the model suitable for transportation safety, automated road inspection, and smart‑city infrastructure applications.
🧠 Model Overview This ONNX model is optimized for real‑time inference on both edge devices and cloud environments. It is built using the Ultralytics YOLO26n architecture and exported to ONNX format for maximum compatibility and deployment flexibility. Key Features
- Detects potholes across varied road conditions
- Lightweight and fast inference
- Runs on Raspberry Pi, Jetson Nano, cloud servers, and desktop GPUs
- Ideal for smart‑city analytics, maintenance automation, and road‑safety systems
📊 Performance Metrics Evaluated on a held‑out validation set from the same dataset:
- Precision: 80.9%
- Recall: 66.0%
- mAP50: 73.9%
- mAP50‑95: 48.2% These metrics demonstrate strong precision and reliable detection performance suitable for real‑world deployment.
📦 Intended Use This model is designed for:
- Automated road‑condition monitoring
- Smart‑city infrastructure systems
- Transportation safety analytics
- Municipal maintenance planning
- Real‑time pothole detection on edge devices
- Integration into mobile, web, or embedded inspection tools
⚙️ Technical Details
- Format: ONNX
- Input Size: 640×640
- Architecture: YOLO26n (Ultralytics)
- Classes: Pothole
- Training Epochs: 100
- NMS: Enabled
👤 Author Created by: Darshan Modi Focused on building practical AI systems for real‑world deployment, including road‑safety applications, edge‑device inference, and lightweight computer‑vision solutions.