Objective:
- Train a Vector-Quantized Variational Autoencoder (VQ-VAE) on the skin lesion dataset to efficiently encode and decode high-dimensional image data while capturing meaningful latent representations.
- Train an Auto-Regressive Model of your choice to generate new, realistic images based on the learned latent space representations.
Dataset: ISIC dataset.
Visualization can be observed below:
Reconstructed images on Test Data after 2500 Epochs:
Generated images using PixelCNN:
