Face-GAN

This project is inspired by the paper "StarGAN: Unified Generative Adversarial Networks for Multi-Domain Image-to-Image Translation".

In this work, I aim to transfer facial features—such as expression, lighting, and facial details—from a secondary image to a primary image while preserving the originality of the primary image.

---

Dataset

To train the model, you'll need the CelebHQ-A dataset. Follow these steps:

1. Download the dataset by running the following command:

   bash download.sh celeba-hq-dataset

2. After this you will have a zip folder ("celeba_hq") inside "data" folder. Please extract it

3. The data set would be in this format:

    data
    ├── celeba_hq
    │

Training

1. To train the model run

   python train.py --data_path /path/to/data --output_path /path/to/output

2. For example run the script in this manner:

   python train.py --data_path ./data --output_path ./output

Testing

For testing check the "test.ipynb" file

Results

Below are the results showcasing the feature transfer:

• The first image is the primary input.
• The second image is the secondary input.
• The third image is the generated output.

Example Outputs:

Result 1:

Result 2:

Result 3: