Denoising Diffusion Probabilistic Model

Original paper: https://arxiv.org/abs/2006.11239

TODO

• Training and generating samples of vanilla DDPM
• Add dataloader for arbitary image dataset
• Training on larger size images
• Trying out Classifier-free guidance

Training on CIFAR 100 dataset

python train.py

Generated samples during training

Each training epoch will render samples twice, each time a set of 64 images are sampled.

python render_samples.py

Training log