Pothole Detection Model

Challenge participants: Build robust pothole detection model solutions.

Table of Contents

• Quick Start
• Introduction
• Requirements
• Project Structure
• Detailed Tutorial
• Submission Guidelines
• License

Quick Start

Download code baseline

# 1. Install dependencies (Do not add any other libraries to requirments.txt file)
pip install -r requirements.txt

# 2. Start developing your solution. Follow the tutorial below to implement your AI model

# 3. Run the demo to test your implementation
python demo.py

# 4. Verify Your Submission
python -m my_ai_lib.predict_submission

Note: You need to implement your pothole detection model following the tutorial guide.

Introduction

Pothole detection is a critical application in smart city infrastructure and road maintenance. This AI-powered solution enables automated identification of road surface defects, particularly potholes, using computer vision and deep learning techniques. The system can process road images to detect and locate potholes, contributing to safer driving conditions and more efficient road maintenance planning.

AIOZ Pothole Detection Challenge

In this challenge, participants will be provided with:

• Model Access: Pothole Detection Model

  • Code baseline to develop solutions.
  • Predefined libraries and tools in requirements.txt (Do not add any other libraries to requirments.txt file).

• Dataset Access:Pothole Detection Data

  • Training dataset to train your AI models.
  • Testing dataset to predict labels for generating the submission file.

Goal: Build and improve solutions based on provided resources while fostering creativity in model design and training strategies.

Requirements

System Requirements

• Python 3.10+
• CUDA-compatible GPU (recommended)

Dependencies

Install all required packages in requirements.txt (Do not add any other libraries to requirements.txt):

pip install -r requirements.txt

Project Structure

Your AI library should follow this structure:

repository/
├── my_ai_lib/                   # Your AI library
│   ├── __init__.py              # Required: Library initialization
│   ├── run.py                   # Required: Main workflow function
│   ├── predict_submission.py    # Required: Submission function
│   └── [your_modules]/          # Your custom modules
├── models/                      # Model weights directory
├── demo.py                      # Demo script
├── requirements.txt             # Dependencies
└── README.md                    # Documentation

Description

• The input and output of the AI model are defined in my_ai_lib/run.py.
• The AI model is designed to process a single input item.
• In my_ai_lib/predict_submission.py, we process all items in the test dataset and save the results to result.csv in the required submission format.
• Please refer to the instructions below for more details.

Key Components

Component	Description	Status
my_ai_lib/	Core AI library directory	Required
my_ai_lib/__init__.py	Library initialization	Required
my_ai_lib/run.py	Main AI workflow	Required
my_ai_lib/predict_submission.py	Submission function	Required
demo.py	Demo and testing script	Required

Detailed Tutorial

Step 1: Initialize Your AI Library

1.1 Define my_ai_lib/__init__.py

from .run import run

This file exposes the run() function from run.py as an attribute of the my_ai_lib, allowing it to be called via my_ai_lib.run()

1.2 Define Input/Output Objects in my_ai_lib/run.py

Create your custom input and output classes:

from pathlib import Path
from typing import Any, Union, Literal
from aioz_ainode_adapter.schemas import InputObject, OutputObject, FileObject

class MyInput(InputObject):
    input_image: str
    example_param: Any

class MyOutput(OutputObject):
    text: str
    output_image: FileObject

Step 2: Understanding AIOZ Schema Objects

The aioz_ainode_adapter library defines 3 core object types based on pydantic.BaseModel:

InputObject

Define the format for input when the AIOZ-AI-Node system sends it to your AI library.

Default Parameters:

Parameter	Type	Description
device	Choice	Device for your model: ["cuda", "cpu", "gpu"]
model_storage_directory	String	Directory containing model weights

Important : Always use model_storage_directory for model weight paths, as AIOZ-AI-Node will specify this location.

OutputObject

Define the format for output when your AI library sends to the AIOZ-AI-Node system.

FileObject

Define the format for the file, if your output has a file. This object has two fields:

Field	Type	Description
data	Choice	File data: io.BufferedReader, Path, or URL
name	String	File name

Example FileObject creation:

output_file = FileObject(data=open("file/path.csv", "rb"), name="output.csv")

Note:

• Input files must be local file paths or URLs
• Output files must be FileObject instances

Step 3: Implement the Main Workflow

3.1 Define Your AI Task Function

def do_ai_task(
        input_image: Union[str, Path],
        example_param: Any,
        model_storage_directory: Union[str, Path],
        device: Literal["cpu", "cuda", "gpu"] = "cpu",
        *args, **kwargs) -> Any:
    """
    Define AI task: load model, pre-process, post-process, etc.
    
    Args:
        input_image: Path to input image
        example_param: Example parameter
        model_storage_directory: Directory containing model weights
        device: Computing device
        
    Returns:
        Result
    """
    # Define your AI task workflow. Below is an example
    text = "This is the AI task result"
    output_image = open("wiki/aioz.png", "rb")  # io.BufferedReader
    return text, output_image

3.2 Implement the Required run() Function

def run(input_obj: InputObject) -> OutputObject:
    """
    Main entry point for your AI library.
    
    Args:
        input_obj: Input object containing all parameters
        
    Returns:
        OutputObject: Results of AI processing
    """
    # Validate and parse input
    my_input = MyInput.model_validate(input_obj.model_dump())
    print(f"Input: {my_input}")

    # Execute AI task
    text, output_image = do_ai_task(
        input_image=my_input.input_image,
        example_param=my_input.example_param,
        model_storage_directory=my_input.model_storage_directory,
        device=my_input.device
    )

    # Create output object
    output_file = FileObject(data=output_image, name="output_image.png")
    output_obj = MyOutput(text=text, output_image=output_file)
    return output_obj

Critical: The run() function name is mandatory and cannot be changed. The do_ai_task() function can be renamed and customized.

Step 4: Create Demo Script

Create demo.py to test your implementation:

import my_ai_lib
from aioz_ainode_adapter.schemas import InputObject

def main():
    """Demo function to test your AI library."""
    input_obj = InputObject(
        input_image="wiki/aioz.png", 
        example_param="example"
    )
    output_obj = my_ai_lib.run(input_obj)
    print(f"Output: {output_obj}")

if __name__ == '__main__':
    main()

The my_ai_lib.run() function receives an InputObject and returns an OutputObject.

Run this command to test your implementation:

python demo.py

Expected console output:

Input: type='InputObj' device='cuda' model_storage_directory='models' input_image='wiki/aioz.png' example_param='example'
Output: type='OutputObj' text='This is the AI task result' output_image=FileObject(type='FileObj', data=<_io.BufferedReader name='wiki/aioz.png'>, name='output_image.png')

Step 5: Add Model Weights

Place your trained model files in the models/ directory:

models/
├── model.pth          # Your trained model
├── config.json        # Model configuration
└── etc. 

Step 6: Create Prediction Script (For Private Submissions)

Implement the predict_submission() function in my_ai_lib/predict_submission.py:

Requirements:

• Function accepting test data folder path (string).
• Load your trained model.
• Process test dataset.
• Generate predictions.
• Save results as ./result.csv

Implementation Template:

from aioz_ainode_adapter.schemas import InputObject
import my_ai_lib

def predict_submission(test_data_folder: str):
    """
    Generate predictions for challenge submission.
    
    Args:
        test_data_folder: Path to test data directory
    """
    # Load your model
    # Process test data
    # Generate predictions
    # Save to ./result.csv

    # Example:
    # Find test data in folder (contains image files - can use os.walk() to find all image files)
    # Loop with each image:
    # - You define InputObject
    #     input_obj = InputObject(
    #     input_image=str(img_path),
    #     example_param=str(example),
    # )
    # - Output (label is predicted from your model): output = my_ai_lib.run(input_obj)
    # - Write the predicted label (output.text) to result.csv (format: id, predict_label) 
    pass

def main():
    """Main function for testing submission."""
    predict_submission("path/to/test/data")


if __name__ == '__main__':
    main()

Important: The result.csv must match the challenge's sample submission format.

Verify Your Submission:

python -m my_ai_lib.predict_submission

Submission Guidelines

Submission format

Submit a csv file with two columns:

• id: The unique identifier (image name) for each image from test set(e.g., 0001.jpg)
• predict_label: A string representing the predicted results in the image, this string has format: class confidence_score x_center y_center box_width box_height:
  • class: object category (1 : pothole).
  • confidence_score: the confidence score of the prediction (range 0-1).
  • x_center, y_center: normalized coordinates of the bounding box center.
  • box_width, box_height: normalized width and height of the bounding box.

Note: x_center and box_width are normalized by image_width; y_center and box_height are normalized by image_height.

Example:

• An object

id, predict_label
0001.jpg, 1 0.927 0.191 0.808 0.208 0.103

• Two objects

id, predict_label
0002.jpg, 1 0.927 0.191 0.808 0.208 0.103 1 0.555 0.123 0.138 0.338 0.443

License

This repository is licensed under the MIT License.