iwuqing/scope

[IEEE TCI] Self-Supervised Coordinate Projection Network for Sparse-View Computed Tomography

scope's Issues

Replace Sigmoid with ReLU as the activate function of output layer might perform better

In my experiments, I found that using ReLU as activate the function of output layer might get better results, especially in more soft tissue scenes. With Sigmoid, the result shows more noise like Salt and Pepper. I guess it might be related to the grad explosion or disappearance in backward process.

About the fanbeam

Hi,

Thanks for your great work! And I'm trying to reproduce your work, I've been able to successfully replicate the part of parallel beam but it appears that there is no code available for fanbeam reconstruction. I'm quite interested in this aspect, and I'm wondering if you could provide the code for fanbeam reconstruction. I attempted to implement it on my own, but the results have been unsatisfactory. I'm not sure if there's a bug or some other issue causing the problem.

Best Regards

Can the author provide the code for 3d svct recon?

Thanks for your impressive work. I run your code on my 2d CT dataset and it achieves the comparable performance with my re-implementation results. However, this code cannot handle the 3d SVCT reconstrcution problem, achieving inferior perfomance to our re-implementation results in the same settings. Can you provide the code for 3d svct reconsturction? Thank you so much.
ps: pyc file is very easy to decompile.