Skip to content

Home

Jump to bottom
Puranjay Savar Mattas edited this page Jan 23, 2024 · 3 revisions

Photorealistic AI Model for Face Generation using DCGANS

Please note this project is still under development and the README & the code are still being updated.

Project Overview

In this project, I have defined and trained a DCGAN on a dataset of faces. My goal was to get a generator network to generate new images of faces that look as realistic as possible! At the end of the project, I was able to visualize the results of my trained Generator to see how it performs; my generated samples look like fairly realistic faces with small amounts of noise.

Due to resource constraints, we will be focusing on generating images at a lower resolution of 128x128 pixels. In this endeavor, we're utilizing the CelebFaces Attributes (CelebA) Dataset. This dataset is conveniently available on their website. Our chosen method for image synthesis is the Deep Convolutional Generative Adversarial Networks (DCGANs). Prior to training this DCGAN model, our dataset will undergo a rigorous preprocessing phase to ensure optimal results. The ultimate objective is to train our DCGAN to synthesize new, lifelike faces that evoke the essence of the celebrities in our training set, albeit in a fabricated manner.

Project Instructions - Training

  1. Clone the repository:

    git clone https://github.com/psavarmattas/PhotorealisticAI.git

  2. cd into the repository:

    cd ./PhotorealisticAI

  3. Download the dataset & unzip all the images in dataset/celebA_dataset

  4. Create a new environment with python 3.11 and activate it:

    conda create --name PhotoRealisticAI python=3.11 && conda activate PhotoRealisticAI

  5. Install the required packages:

    pip install -r requirements.txt` or `conda install --file requirements.txt

Note: If you are on MacOS please add the following dependencies in your environment for compatibility tensorflow-macos & tensorflow-metal. Also add them to the the requirements.txt so that the program checks them and does not miss any dependencies.

  1. Run the command: python main.py

Note: For any modifications or understanding of the program please go through the files as each and every function/file has it's own docstrings for detailed working.

Project Instructions - APIs

The backend code for the project is located in the backend folder. Please refer to the README in that folder for instructions on how to use the API.

Project Instructions - Web Dashboard

The frontend code for the project is located in the frontend folder. Please refer to the README in that folder for instructions on how to start the web dashboard.

Project Instructions - Full Program (Backend + Frontend)

  1. Navigate to the root of the project

    cd ./PhotorealisticAI

  2. Follow the above instructions to install the dependencies and create a new environment

  3. Run the command:

    python run.py

  4. Open your browser to http://localhost:4200/ and generate new faces!

The frontend code for the project is located in the frontend folder. Please refer to the README in that folder for instructions on how to start the web dashboard.

Project Structure

Please go through the project structure to understand the working of the project.

PhotoRealisticAI/
│
├── backend/
│   ├── utils/
│   │   ├── generator_loader.py
│   ├── generate_images_api.py
│   └── README.MD
│
├── ckpt/
│   ├── discriminator/
│   └── generator/
│
├── dataset/
│   └── celebA_dataset/
│
├── fig/
│
├── frontend/
│   ├── README.MD
│   ├── photorealistic-ai-frontend/
│   │   ├── node_modules/
│   │   ├── src/
│   │   │   ├── app/
│   │   │   │   ├── app.component.scss
│   │   │   │   ├── app.config.server.ts
│   │   │   │   ├── app.component.html
│   │   │   │   ├── app.component.spec.ts
│   │   │   │   ├── app.component.ts
│   │   │   │   ├── app.config.ts
│   │   │   │   ├── app.routes.ts
│   │   │   ├── assets/
│   │   ├── .editorconfig
│   │   ├── .gitignore
│   │   ├── angular.json
│   │   ├── package-lock.json
│   │   ├── package.json
│   │   ├── server.ts
│   │   ├── tsconfig.app.json
│   │   ├── tsconfig.json
│   │   ├── tsconfig.spec.json
│   │   ├── index.html
│   │   ├── main.server.ts
│   │   ├── main.ts
│   │   ├── styles.scss
│   │   └── favicon.ico
│
├── models/
│   ├── discriminator.py
│   ├── gan.py
│   ├── generator.py
│
├── out_metrics/
│   ├── logEpoch.txt
│   └── logFID.txt
│   ├── logMetric.txt
│   └── logVerbose.txt
│
├── plot/
│
├── utils/
│   ├── calculate_fid.py
│   ├── check_dependencies.py
│   ├── check_tf.py
│   ├── cooldown_gpu.py
│   ├── data_generation.py
│   ├── data_processing.py
│   ├── file_operations.py
│   ├── image_plotting.py
│   ├── metric_plotting.py
│   ├── model_operations.py
│   ├── run_servers.py
│   ├── train_load_finish.py
│   ├── train.py
│   └── visualization.py
│
├── .gitignore
├── LICENSE.MD
├── main.py
├── README.MD
├── requirements.txt
├── run.py
└── start_servers.sh

License

Please make sure you read the license before using the code.

Attribution for the dataset

The dataset is provided by title = Deep Learning Face Attributes in the Wild author = Liu, Ziwei and Luo, Ping and Wang, Xiaogang and Tang, Xiaoou, booktitle = Proceedings of International Conference on Computer Vision (ICCV), month = December, year = 2015

Clone this wiki locally