Graffiti Detection: Object Detection with TensorFlow

This project identifies graffiti styles and general objects using TensorFlow and a custom-trained object detection model. It supports two approaches: running inference via Jupyter notebooks or as standalone Python scripts. Below are the details and usage instructions for the project.

Features

1. Graffiti Style Detection

   • Custom label map for graffiti styles:
     • Caracter
     • Tag
     • Bomba
     • Roller
     • Wildstyle
     • 3D
     • Moniker
     • S_Tren

2. Object Detection

   • Uses the Open Images V4 dataset to detect objects from the world.

3. Database and FTP Integration

   • Fetch images and metadata from an SQL database.
   • Upload processed inference results to an FTP server.

---

Installation

Prerequisites

Ensure the following are installed on your system:

• Python 3.7+
• TensorFlow (>=2.0)
• MySQL Connector for Python
• Required libraries from requirements.txt:

  pip install -r requirements.txt

---

Usage

Jupyter Notebooks

1. Inference_SQL.ipynb

• Purpose: Performs inference and integrates with a custom database.
• Steps:
  1. Run all setup cells to install dependencies (TensorFlow, MySQL connector, etc.).
  2. Update the parameter MUID in the notebook to a valid value obtained from your database.
  3. Ensure database connectivity is configured in the config.json file.

2. Inference_TF_faster_rcnn.ipynb

• Purpose: Inference using the Open Images V4 dataset.
• Steps:
  1. Run all cells to set up the environment.
  2. Ensure the required models are present in the inference_graph/saved_model directory.

Standalone Python Script

inference-sql-argparse.py

• Purpose: Perform inference via command-line arguments with SQL and FTP integration.
• Steps:
  1. Install requirements and dependencies.
  2. Run the script with the necessary arguments:

     python inference-sql-argparse.py -MUID <Your_MUID>

  3. Ensure the following files are configured:
     • config.json: Contains SQL and FTP credentials.
     • labelmap.pbtxt: Custom label map for graffiti styles.

---

Configuration

SQL Database

• The script connects to the database to fetch metadata and image URLs.
• Example connection code:

  cnx = mysql.connector.connect(
      user=config["SQL"]["username"],
      password=config["SQL"]["password"],
      host=config["SQL"]["hostname"],
      database=config["SQL"]["database"]
  )

• Ensure you update the config.json file with correct database credentials:

  {
      "SQL": {
          "username": "your_username",
          "password": "your_password",
          "hostname": "your_hostname",
          "database": "your_database"
      }
  }

FTP Server

• After inference, processed images and JSON results are uploaded to an FTP server.
• Example upload function:

  def DataUpload(local_dir, target_dir):
      ftp_server = ftplib.FTP(
          config["FTP"]["hostname"],
          config["FTP"]["username"],
          config["FTP"]["password"]
      )
      ftp_server.encoding = "utf-8"
      ftp_server.cwd('/media/exported_images')
      # Create target directory if it doesn't exist
      if directory_exists(target_dir, ftp_server) is False:
          ftp_server.mkd(target_dir)
      ftp_server.cwd(target_dir)
      # Upload files
      for filename in os.listdir(local_dir):
          with open(os.path.join(local_dir, filename), 'rb') as file:
              ftp_server.storbinary(f'STOR {filename}', file)
      ftp_server.quit()

• Update FTP credentials in the config.json file:

  {
      "FTP": {
          "hostname": "your_ftp_hostname",
          "username": "your_ftp_username",
          "password": "your_ftp_password"
      }
  }

---

Directory Structure

project_root/
├── config.json           # Configuration file for SQL and FTP
├── requirements.txt      # Dependencies
├── labelmap.pbtxt        # Label map for graffiti styles
├── inference_graph/      # TensorFlow saved models
│   └── saved_model/
├── notebooks/
│   ├── Inference_SQL.ipynb
│   └── Inference_TF_faster_rcnn.ipynb
├── scripts/
│   └── inference-sql-argparse.py
└── exported_images/      # Processed inference results

---

Sample Results

• Insert sample images and JSON outputs here to demonstrate the results.

---

License

This project is licensed under the MIT License. See the LICENSE file for details.

---

Contributions

Feel free to fork this repository and submit pull requests. For major changes, please open an issue first to discuss what you would like to change.

---

Contact

For inquiries or support, please contact the project maintainer.