A PyTorch implementation of Variational Autoencoders and Denoising Diffusion Probabilistic Models applied to a custom “hotdog” image dataset. Demonstrates end-to-end generative modeling, from data loading through training, sampling, and retrieval.
- VAE baseline on MNIST to validate reconstruction & latent representations
- Latent Diffusion Model (LDM) built with Hugging Face Diffusers and a custom Stable Diffusion VAE wrapper
- Training & evaluation notebooks covering:
- Hypothesis formulation & loss analysis
- Sampling & visualizing generated images
- Nearest-neighbour retrieval in latent space
- Language: Python 3.8+
- Frameworks: PyTorch, Torchvision, Hugging Face Diffusers
- Data & Analysis: NumPy, Pandas, Matplotlib
- Notebook: Jupyter