search

cabouman / mbircone

BSD 3-Clause "New" or "Revised" License
11 stars 9 forks source link

Remove wght recon #54

Closed dyang37 closed 2 years ago

dyang37 commented 2 years ago

This PR is consisted of two parts.

  1. upgrade numpy version to 1.22, and tensorflow & keras versions to 2.8 in requirements files.
  2. Remove spatially variant recon weight calculation and associated functions (see issue #4 ). This is a partial fix for the memory issue (issue #51 ). (I also talked to Soumendu and he thinks it's okay to remove spatially variant recon weights).

Testing:

  1. Tested version compatibility by running clean_install_all.sh, and ran demo_mace3D_fast.py on Gilbreth. Everything works, and the output of nvidia-smi indicates that the new tensorflow version is able to use GPU on Gilbreth.
  2. Compared memory usage of 3D shepp Logan reconstruction (increase num_det_channels, num_det_rows, num_views by 2x and ran demo_3D_shepp_logan.py ) before and after removing wghtRecon functionality and variables. For a reconstruction size of 230x115x115, the memory usage drops from 820M to 808M, and results are visually the same as before.
cabouman commented 2 years ago

Does this change any functionality in the code?

dyang37 commented 2 years ago

Does this change any functionality in the code?

No it does not.