search

CEA-COSMIC / pysap-mri

MRI external plugin for Python Sparse data Analysis Package
Other
43 stars 18 forks source link

Auto threshold #161

Closed paquiteau closed 1 year ago

paquiteau commented 1 year ago

This PR implements auto-thresholding tuning for Proximal operator.

The main two function wavelet_noise_estimation and wavelet_threshold_estimation are operator independent, to let the user access them (and later fine tune the threshold if required).

The estimation of threshold is done in 2 Step:

  1. Estimate the noise variance on specific ranges (on every subband, on every level, or using a single global estimate )
  2. Compute a threshold for a specific ranges using one of the 3 following method:
    • SURE
    • Hybrid-SURE
    • Universal threshold. In total this make 9 auto-tuned configuration available.

I also provide an example reconstruction script.

I don't have much time to write good unittests for now, and IMO the test module should undergo the same refactoring I did for ModOpt.

chaithyagr commented 1 year ago

@paquiteau can you please ensure that the example works fine? I made a few changes, I still see that the weights are all 0 though in this example.