The goal of this project is to learn latent representations of faces so we can predict what our children would look like, if we were to hook up with our Stack B colleagues.
- Implement a VAE
- Implement a CVAE
- Use a convolutional encoder
- Implement MNIST training & testing
- Implement CelebA training & testing
- Label our custom dataset
- Preprocess our custom dataset
- Train & evaluate a smiley CVAE on our dataset
- Walk in latent space to interpolate between samples (2D PCA projection?)
- Learn disentangled identities and smile levels