Mathis Matthieu, Benjamin Pipaud, Lucas Saban. MC course @Ensae

Ising Model denoising using hyperparameter estimation and MCMC techniques.

With systematic scan on the left and randomized scan on the right:

pip install -r requirements.txt

Then :

main.py --findsigma True --alpha 0.0 #etc...

The arguments available are :

• alpha : The alpha parameter of the Ising Model, default value is 0
• beta: The beta parameter of the Ising Model, default value is 1.3
• sigma : Variance of the gaussian noise, default value is 179 (for 8 bit images)
• findsigma : If set to True, the denoising will be done without being given the value of sigma, default value False
• g : If set to true a gif will be produced. Default value to True.
• b : Number of burn in steps. Default to 40
• ns: Number of sampling steps. Default to 5
• imp: Path of the raw image. Default to 'data/input/test_img.jpeg'