Noise Conditional Score Network

This is a baseline 'reproduction' of the work Generative Modeling by Estimating Gradients of theData Distribution by Song et al.

Here I use the MNIST dataset with a modified UNET network. The UNET decoder uses Conditional Instance Normalization ++ from the paper to decode based on the input noise. This is not added to the encoder due to technical reasons.

Dataset

No noise added	Noise added

Samples generated by the trained diffusion model

MNIST	CIFAR10

@misc{qbeer,
  author       = {Alex Olar},
  title        = {Noise Conditional Score Network},
  howpublished = {GitHub repository},
  month        = {February},
  year         = {2023},
  url          = {https://github.com/qbeer/noisy_diffusion}
}