YOLOv5 Distress Detection

A deep learning project for automated detection of pavement road distresses using computer vision techniques, developed in response to the IEEE 2020 Global Road Detection Challenge.

Overview

This project implements object detection models to identify and classify various types of road distresses from images. The solution uses state-of-the-art deep learning frameworks and models to achieve accurate detection results.

Key Features

• Object Detection Models: YOLOv5, Faster R-CNN
• Frameworks: PyTorch, TensorFlow
• Target Application: Road distress detection and classification
• Research Paper: View published results

Project Structure

├── data_preprocessing/          # Data preprocessing utilities
├── notebooks/                   # Jupyter notebooks for analysis and training
│   ├── modeling/               # Model training notebooks
│   ├── adhoc/                  # Ad-hoc analysis notebooks
│   └── titanmu/                # TITANMU model experiments
└── README.md                   # This file

Getting Started

Prerequisites

• Python 3.7+
• CUDA-compatible GPU (recommended: NVIDIA RTX 3090)
• Docker (for containerized training)

Hardware Requirements

Recommended Hardware: NVIDIA RTX 3090

Driver and CUDA Requirements:

• NVIDIA Driver: 450+ (required for Ampere architecture)
• CUDA: 11.0+ (compatible with driver 450+)
• cuDNN: 8.0+ (compatible with CUDA 11.0+)

For detailed setup instructions, see: RTX 3090 Deep Learning Setup Guide

Docker Setup

This project uses Docker for consistent training environments. The notebooks reference YOLOv5 Docker images for model training.

Running with Docker

1. Pull the YOLOv5 Docker image:

   docker pull ultralytics/yolov5:latest

2. Run the container with GPU support:

   docker run --gpus all -it --rm -v $(pwd):/workspace ultralytics/yolov5:latest

3. For interactive development:

   docker run --gpus all -it --rm -v $(pwd):/workspace -p 8888:8888 ultralytics/yolov5:latest jupyter lab --ip=0.0.0.0 --port=8888 --allow-root

4. Access Jupyter Lab: Open your browser and navigate to http://localhost:8888

Docker Compose (Alternative)

Create a docker-compose.yml file for easier management:

version: '3.8'
services:
  yolov5:
    image: ultralytics/yolov5:latest
    container_name: yolov5-distress-detection
    runtime: nvidia
    volumes:
      - .:/workspace
    ports:
      - "8888:8888"
    command: jupyter lab --ip=0.0.0.0 --port=8888 --allow-root

Then run:

docker-compose up

Data Pipeline

The project includes a structured data processing pipeline:

1. XML to TXT Conversion: XML_to_TXT_Annotation_Conversion_Pipeline.ipynb

   • Converts XML annotation files to TXT format for YOLOv5 compatibility

2. Image Augmentation: A01 - Load and Augment an Image.ipynb

   • Defines and applies data augmentation techniques to training images

Usage

1. Data Preparation: Run the XML to TXT conversion notebook to prepare your annotations
2. Model Training: Use the notebooks in the modeling/ directory to train YOLOv5 models
3. Evaluation: Evaluate model performance using the prediction notebooks

Research

This project is based on research published in the paper available at: arXiv:2202.13285

Contributing

Please refer to the project objectives in objectives.md for current goals and development priorities.