mTCT is the best way to config any geometry into CT projection operator of nerual network, such as CNN-based network, Tranformer-based network or NeRF.
Firstly, run the geometry editing script with MATLAB, where you can get projection data file(etc. projVecReal.m) containing a matrix of N*12. Then you can use pytorch CUDA extension to compile our operators. finally, Link the location of geometric files in the YAML script and use compiled operators to project the embedded network, as shown in this script.