SBD-iPALM is a MATLAB package for sparse blind deconvolution (SBD) using the iPALM method, motivated by studies in blind deconvolution as a nonconvex optimization problem, and by applications in Scanning Tunneling Microscopy (STM).
The iPALM method is an accelerated first-order method with optimal convergence guarantees to a stationary point when solving nonconvex optimization problems. We combine this with the l1-reweighting method to produce sparse, robust and interpretable activation maps.
Please contact Yenson Lau for any requests / feedback. We thank Sky Chueng, John Shin and Abhay Pasupathy for the simulated kernel data Data_N_50_nDef_1_thop_-0.200.mat used in genconvdata.m and example.m.
2018-06-20
-
Added
PDRegularizerfor regularizers of the form g(Dx)- comes with a Chambolle-Pock primal-dual method to solve for prox updates
-
A difference operator
imgdifffor image deblurring: plug into D above. Improvements in progress. -
Numerous minor adjustments to clean up the code, which will continue to roll in the next few updates.
See UPDATES.md for complete list.
-
An ADMM iterator may be added in the future.
-
Reimplement wrappers of SBD, CDL, etc. as classes? To add various associated methods such as synthesis.