A simple reproduction of score-based model NCSN (Noise Conditional Score Networks)
Original paper: https://proceedings.neurips.cc/paper_files/paper/2019/file/3001ef257407d5a371a96dcd947c7d93-Paper.pdf
Official repo: https://github.com/ermongroup/ncsn
Usage: python train.py
Implementation details: Use naive U-Net composed of convolutions and IN instead of the refineNet archetecture from original paper, which turns out that the convergence will be slower than official release
Some samples on MNIST after 50 epochs:
Loss trend:
