This is my (Pranav Talluri) Part II Project and Dissertation at the University of Cambridge.

Completed So Far

• Implemented base UNET
• Built unconditional image generation for MNIST digits and MNIST fashion (black and white)
• Implemented ResUNET
• Built unconditional image generation for CIFAR-10 (colour)
• Implemeted FID score based evaluation
• Notes on unconditional DDPMs

TODO

• Test impact of attention blocks
• Test FID score evaluation with larger sample size to generate comparable results with original papers
• Learn how to use HPC
• Implement methods for image masking for conditional image generation
• Notes on conditional DDPMs
• Implement EMA
• The original paper does not sweep over all hyperparameters (learning rate, batch size, ema decay factor)

To Learn

• How to perform training and sampling with conditional DDPM
• Interpolation?

Reading List

• Free form image inpainting with gated convolution - https://arxiv.org/abs/1806.03589
• Classifier free diffusion guidance - https://arxiv.org/abs/2207.12598
• Generative modelling with inverse heat dissapation - https://arxiv.org/abs/2206.13397