zakobian/ct_framework_

CT framework

Example call:

python train.py --setup=3 --dataperc=75 --epochs=100 --lr=1e-5 --eps=1e-5 --alg=ADR --iterates=100 --valid=10 --batch-size=20 --detectors=128 --gpu=1 --noise=0.03 --load=False --seed=10 --wclip=False

Description

This framework was created in order to help compare learned and variational approaches to CT reconstruction in a systematic way. The implementation is based on python libraries odl and pyTorch. The list of implemented algorithms include:

• FBP (Filtered back-projection)
• TV (Total Variation)
• ADR (Adversarial Regularizer): https://arxiv.org/abs/1805.11572
• LG (Learned gradient descent): https://arxiv.org/abs/1704.04058
• LPD (Learned primal dual): https://arxiv.org/abs/1707.06474
• FL (Fully learned): https://nature.com/articles/nature25988.pdf
• FBP+U (FBP with a U-Net denoiser): https://arxiv.org/abs/1505.04597

In order to add your own algorithms to the list, create a new file in the Algorithms folder in the form name.py and use BaseAlg.py as the template.

ct_framework_'s Issues

Bug?

Hi!
Nice work

I think in this line should be
delf = config.fwd_op_mod(f[:,1:2])

CT_framework_/Algorithms/LPD.py

Line 42 in e906e57

delf = config.fwd_op_mod(f[:,0:1])

Structure of each dataset

Hi👋, I want to use this framework to do image reconstruction, but I am having trouble loading the dataset. Please, can you tell me What is the file structure of each dataset? Thank you!