CGLS_CUDA

This is a GPU implementation of the Conjugate Gradient Least Squares (CGLS) algorithm for 2D data sets. It takes projection data and an initial reconstruction as input, and returns the reconstruction after a specified number of iterations.

The internal state of the CGLS algorithm is reset every time astra_mex_algorithm(‘iterate’) is called. This implies that running CGLS for N iterations and then running it for another N iterations may yield different results from running it 2N iterations at once.

Supported geometries: parallel, parallel_vec, fanflat, fanflat_vec.

Configuration options

name	type	description
cfg.ProjectionDataId	required	The astra_mex_data2d ID of the projection data
cfg.ReconstructionDataId	required	The astra_mex_data2d ID of the reconstruction data. The content of this when starting CGLS is used as the initial reconstruction.
cfg.option.SinogramMaskId	optional	If specified, the astra_mex_data2d ID of a projection-data-sized volume to be used as a mask.
cfg.option.ReconstructionMaskId	optional	If specified, the astra_mex_data2d ID of a volume-data-sized volume to be used as a mask.
cfg.option.GPUindex	optional	Specifies which GPU to use. Default = 0.
cfg.option.DetectorSuperSampling	optional	For the forward projection, DetectorSuperSampling rays will be used. This should only be used if your detector pixels are larger than the voxels in the reconstruction volume. Defaults to 1.
cfg.option.PixelSuperSampling	optional	For the backward projection, PixelSuperSampling^2 rays will be used. This should only be used if your voxels in the reconstruction volume are larger than the detector pixels. Defaults to 1.

Example

import astra
import matplotlib.pyplot as plt
import numpy

# create geometries and projector
proj_geom = astra.create_proj_geom('parallel', 1.0, 256, numpy.linspace(0, numpy.pi, 180, endpoint=False))
vol_geom = astra.create_vol_geom(256,256)
proj_id = astra.create_projector('cuda', proj_geom, vol_geom)

# generate phantom image
V_exact_id, V_exact = astra.data2d.shepp_logan(vol_geom)

# create forward projection
sinogram_id, sinogram = astra.create_sino(V_exact, proj_id)

# reconstruct
recon_id = astra.data2d.create('-vol', vol_geom, 0)
cfg = astra.astra_dict('CGLS_CUDA')
cfg['ProjectorId'] = proj_id
cfg['ProjectionDataId'] = sinogram_id
cfg['ReconstructionDataId'] = recon_id
cgls_id = astra.algorithm.create(cfg)
astra.algorithm.run(cgls_id, 100)
V = astra.data2d.get(recon_id)
plt.gray()
plt.imshow(V)
plt.show()

# garbage disposal
astra.data2d.delete([sinogram_id, recon_id, V_exact_id])
astra.projector.delete(proj_id)
astra.algorithm.delete(cgls_id)

%% create phantom
V_exact = phantom(256);

%% create geometries
proj_geom = astra_create_proj_geom('parallel', 1.0, 256, linspace2(0,pi,180));
vol_geom = astra_create_vol_geom(256,256);

%% create forward projection
[sinogram_id, sinogram] = astra_create_sino_cuda(V_exact, proj_geom, vol_geom);

%% reconstruct
recon_id = astra_mex_data2d('create', '-vol', vol_geom, 0);
cfg = astra_struct('CGLS_CUDA');
cfg.ProjectionDataId = sinogram_id;
cfg.ReconstructionDataId = recon_id;
cgls_id = astra_mex_algorithm('create', cfg);
astra_mex_algorithm('iterate', cgls_id, 100);
V = astra_mex_data2d('get', recon_id);
imshow(V, []);

%% garbage disposal
astra_mex_data2d('delete', sinogram_id, recon_id);
astra_mex_algorithm('delete', cgls_id);

Extra features

CGLS_CUDA supports astra.algorithm.get_res_norm() / astra_mex_algorithm(‘get_res_norm’) to get the 2-norm of the difference between the projection data and the projection of the reconstruction. (The square root of the sum of squares of differences.)