This repository is built based on the orignal "Double DIP" implementation by Gandelsman et al. (http://www.wisdom.weizmann.ac.il/~vision/DoubleDIP/).
To account for the partial-band transfer of features on different slices
when they superimpose, an additional filtering is added to the images
generated by the DIPs. You can choose to use single-layer filters by
setting blur_type='filter', or a 3-layer shallow U-Net by setting
blur-type='net'.
You may use zp_separation.py as a starting point.
Suggested paramas:
(blur) gamma_excl = 0.4
(net) gamma_excl = 0.04 - 0.1
learning_rate = 1e-3
gamma_reg = 0.5
reg_threshold = 100 iters