The code removes distortion synthetically introduced to the MNIST dataset.
The code supplements the results in the paper "Global Guarantees for Blind Demodulation using Generative Prior" by Paul Hand and Babhru Joshi

-----------------------------------------------------------
Dependicies: PyTorch and Juypter Notebook

-----------------------------------------------------------
Folders: 
- "Data" contains the MNIST dataset
- "weights" contain the pre-trained weights for the distortion and MINIST (signal) generator. See the paper for the description of the architecture.
- "Distortion_used" contains the distortion images used in generating the simulation results. See the paper for the how these distortaion was generated.

-----------------------------------------------------------
Juypter Notebook:
"Generative_blind_demodulation.ipnb" takes an image from MNIST dataset and and image from the "Distortion_used" folder to produce a modulated image. The code then solves the l2 empirical risk minimization program to recover the MINST image from the modulated image.