Bridge Inspection Part Detection

日本語版READMEはこちら

This project contains machine learning models and scripts for detecting and classifying bridge components in inspection images.

Project Overview

This repository includes:

• Fine-tuning scripts for object detection models
• Validation and evaluation scripts
• Visualization tools for detection results
• Training and validation automation scripts
• Automatic monitoring and execution system

Key Components

Training Scripts

• 1st_finetuning_*.py - Fine-tuning scripts with various configurations
• Training supports different fine-tuning types: prompt, linear_prob, full

Validation Scripts

• 2nd_validation_*.py - Model validation and evaluation scripts
• Generates performance metrics and detection results

Visualization Tools

• plot_ap_results_*.py - Scripts for plotting Average Precision (AP) results
• visualize*.py - Tools for visualizing detection outputs
• show_output*.py - Scripts for displaying model outputs

Automation Scripts

• run_*.sh - Shell scripts for automated training and validation workflows
• auto_run_after_completion.sh - Automated execution after training completion
• auto_git_pull_monitor.sh - Automatic git repository monitoring and execution system
• auto_run.sh - Execution script triggered by the monitoring system

Automatic Monitoring System

Auto Git Pull Monitor

The auto_git_pull_monitor.sh script provides automatic monitoring of the git repository and execution of training scripts when changes are detected.

Features

• Continuous Monitoring: Checks for git repository changes every 5 minutes
• Automatic Git Pull: Pulls latest changes from the master branch
• Smart Execution: Executes specified scripts only when changes are detected
• Process Safety: Skips execution if training processes are already running
• Comprehensive Logging: Logs all activities with timestamps

Usage

# Start monitoring with default script
./auto_git_pull_monitor.sh

# Start monitoring with specific script
./auto_git_pull_monitor.sh run_training_validation_5cat.sh

# Available scripts for execution:
# - run_training_validation_all_item.sh
# - run_training_validation_5cat.sh
# - run_training_validation_without_cable_components.sh
# - run_all_item_training_validation_full_data.sh

Configuration

• Check Interval: 5 minutes (300 seconds)
• Log Files:
  • ./logs/git_auto_pull.log - Main monitoring log
  • ./logs/git_changes.log - Git changes log

How It Works

1. Repository Monitoring: Continuously monitors git repository for changes
2. Change Detection: Compares commit hashes before and after git pull
3. Smart Execution: Only executes scripts when actual changes are detected
4. Process Management: Prevents conflicts by checking for running training processes
5. Automatic Recovery: Continues monitoring even after script execution

Process Safety

• Checks for running training processes (1st_finetuning*.py, 2nd_validation*.py, run_*.sh)
• Skips git pull and execution if training is already in progress
• Waits 30 seconds after detecting changes before execution
• Double-checks for new training processes before final execution

Categories Detected

The model can detect 5 main categories of bridge components:

1. Cable Component (ケーブル部材)
2. Cable Support Structure (ケーブル付属構造物)
3. Cable Anchorage (ケーブル定着部)
4. Cable Rods and Bolts (ケーブル関連ロッド、ボルト類)
5. Other Cable Accessories (その他ケーブル関連部品)

Directory Structure

• 1st_OUTPUT/ - Training outputs
• 2nd_RESULT/ - Validation results
• evaluation_results/ - Evaluation metrics
• GLIP/ - GLIP model related files
• logs/ - Training and validation logs
• plot/ - Generated plots and visualizations
• visualized_output_images*/ - Visualization outputs

Usage

Training

Run training scripts using the provided shell scripts:

./run_training_validation_5cat.sh
./run_training_validation_all_item.sh

Evaluation

Evaluate models using validation scripts:

python 2nd_validation_*.py

Visualization

Generate plots and visualizations:

python plot_ap_results_5cat.py
python visualize.py

Requirements

• Python 3.x
• PyTorch
• OpenCV
• Matplotlib
• NumPy
• Other dependencies as specified in individual scripts

Notes

This project is designed for bridge inspection automation and focuses on cable-related component detection.

Workflow Integration

Automatic Training Pipeline

1. Code Updates: Push changes to the git repository
2. Auto Detection: The monitor detects changes within 5 minutes
3. Auto Execution: Specified training/validation scripts are executed automatically
4. Result Generation: Training outputs and validation results are generated
5. Continuous Monitoring: System continues monitoring for next changes

Manual Override

Even with automatic monitoring active, you can still run scripts manually:

./run_training_validation_5cat.sh
./run_training_validation_all_item.sh