Table of Contents
Latent diffusion is a sophisticated technique employed in machine learning, particularly in the context of generative models. It plays a pivotal role in the controlled generation of diverse and high-quality data by traversing a continuous latent space. In latent diffusion, data is generated by perturbing an initial latent representation through a series of steps while introducing noise or perturbations at each stage. This gradual transformation allows the model to explore and learn the complex underlying structures within the data distribution.
Install all the libraries
pip install pytorch torchvision numpy albumentationsChange the arguments values like dataset,batch size,etc in ddpm.py file and call train function inside ddpm.py, save the weights and test using testing.py
- Implementing latent diffusion from scratch
- Writing sample function for testing
- Implementing EMA and class conditioning
- Training on landscape dataset
See the open issues for a full list of proposed features (and known issues).