Photo to Monet with CycleGAN 📷🎨

Overview

This repository details the process of developing and deploying a CycleGAN model that transforms photos into Monet-style paintings. The project is designed to showcase best practices in machine learning model lifecycle management, from development through deployment, using modern tools like Docker, Kubernetes, and GitHub Actions.

Project Setup

Prerequisites

• Docker installed on your machine.
• Access to a GPU for model training and inference tasks.

Downloading Necessary Resources

• Docker Image: Download and use the Kaggle Python GPU image for a compatible setup:

  docker pull gcr.io/kaggle-images/python

• Trained Model: The trained model is not included in the GitHub repository due to file size constraints. Download the trained model from Chuyang Zhang's Model.

• Notebook: The notebook used for training with in-depth illustration on CycleGAN's components under the hood can be found in this link Chuyang Zhang's Notebook.

Running the Notebook for Model Development and Training

• Clone the Repository: Clone this repository to your local machine to get started with the project.

• Set Up the Docker Environment:

  • Start the Kaggle Docker Image:

    docker run --name photo2monet -v $(pwd):/home/jupyter -w /home/jupyter -p 8888:8888 --gpus all -it gcr.io/kaggle-gpu-images/python

  • Build Custom Docker Image: For training on newer GPU architectures (sm_86 and up, such as RTX 3090), use the Dockerfile provided in the repository to build a custom image that incorporates the necessary support:

    docker build -t custom_photo2monet_image .

    This command builds a Docker image using the Dockerfile that adjusts the environment to support the newer GPU architectures.

• Access the Jupyter Notebook: Access the Jupyter Notebook through localhost:8888 to interact with the project directly.

• Experiment with different Hyperparameters to train the model

Setting Up the Web Application

To host the CycleGAN model for inference through a web application:

• Requirements

  • Ensure the Docker environment is correctly set up as detailed above.
  • Download the model.pth file from my Kaggle page or train your own model. Rename the model file to serving_model_v0.pth and place it in the webapp/ directory.

• Running the Web Application

  • Build the Docker Image for the Web Application: Use the Dockerfile within the webapp/ folder to build the image:

    docker build -t webapp-photo2monet ./webapp

  • Start the Web Application: Run the Docker container for the web application:

    docker run --name webapp-photo2monet -p 5000:5000 webapp-photo2monet

  • Access the Web Application: Open a web browser and go to http://localhost:5000 to interact with the web application and perform image transformations.

  

Future Development

• CI/CD Pipeline: Implementation of a CI/CD pipeline using GitHub Actions to automate the build and push of Docker images.
• Testing in Local Kubernetes Environment: Configuring and testing the deployment within a local Kubernetes environment to ensure scalability and reliability.
• Deployment on Cloud Services: Future deployment plans include utilizing free tier services from cloud providers such as Azure to demonstrate cloud deployment scenarios.