completing setup, cleanup

.github/workflows/tomobar_conda_upload.yml

@@ -1,6 +1,6 @@
 name: tomobar_conda_upload
 
-on: [push, pull_request, pull_request_target]
+on: [push]
 
 jobs:
   build-linux:
@@ -20,8 +20,6 @@ jobs:
     - name: Install dependencies with Conda
       run: |
         $CONDA/bin/conda env update --file conda-recipe/environment/environment.yml --name base
-        conda config --set anaconda_upload no
-        export VERSION=`date +%Y.%m` 
 
     - name: Decrypt a secret
       run: ./.scripts/decrypt_secret.sh

.scripts/conda_upload.sh

@@ -8,14 +8,21 @@ CONDA_TOKEN=$(cat $HOME/.secrets/my_secret.json)
 mkdir ~/conda-bld
 conda config --set anaconda_upload no
 export CONDA_BLD_PATH=~/conda-bld
-export VERSION=`date +%Y.%m`
-conda install --yes anaconda-client
-conda build . --numpy=1.20 --python=3.9
-conda build . --numpy=1.21 --python=3.9
-conda build . --numpy=1.22 --python=3.9
-conda build . --numpy=1.20 --python=3.10
-conda build . --numpy=1.21 --python=3.10
-conda build . --numpy=1.22 --python=3.10
+#export VERSION=`date +%Y.%m`
+#conda install --yes anaconda-client
+#conda build . --numpy=1.20 --python=3.9
+#conda build . --numpy=1.21 --python=3.9
+#conda build . --numpy=1.22 --python=3.9
+#conda build . --numpy=1.20 --python=3.10
+#conda build . --numpy=1.21 --python=3.10
+#conda build . --numpy=1.22 --python=3.10
+
+for python_ver in 3.9 3.10; do   
+    for numpy_ver in 1.21 1.22 1.23 1.24; do
+        export VERSION=`date +%Y.%m`"-py"$python_ver"_np"$numpy_ver
+        conda build . --numpy $numpy_ver --python $python_ver
+    done
+done
 
 # upload packages to conda
 find $CONDA_BLD_PATH/$OS -name *.tar.bz2 | while read file

Demos/Python/DemoFISTA_artifacts3D.py

@@ -255,33 +255,6 @@
 plt.title('3D Huber Rec, sagittal')
 plt.show()
 
-# adding RING minimisation component (better model for data with rings - different from GH!)
-#data.update({'huber_threshold' : None})
-_data_.update({'ring_weights_threshold' : 2.0})
-_data_.update({'ring_tuple_halfsizes' : (9,7,9)})
-
-# Run FISTA reconstrucion algorithm with 3D regularisation and a better ring model
-RecFISTA_HuberRING_TV = Rectools.FISTA(_data_, _algorithm_, _regularisation_)
-
-Qtools = QualityTools(phantom_tm, RecFISTA_HuberRING_TV)
-RMSE_FISTA_HUBER_RING_TV = Qtools.rmse()
-print("RMSE for FISTA-OS-Huber-Ring-TV is {}".format(RMSE_FISTA_HUBER_RING_TV))
-
-sliceSel = int(0.5*N_size)
-max_val = 1
-plt.figure() 
-plt.subplot(131)
-plt.imshow(RecFISTA_HuberRING_TV[sliceSel,:,:],vmin=0, vmax=max_val)
-plt.title('3D HuberRing Rec, axial')
-
-plt.subplot(132)
-plt.imshow(RecFISTA_HuberRING_TV[:,sliceSel,:],vmin=0, vmax=max_val)
-plt.title('3D HuberRing Rec, coronal')
-
-plt.subplot(133)
-plt.imshow(RecFISTA_HuberRING_TV[:,:,sliceSel],vmin=0, vmax=max_val)
-plt.title('3D HuberRing Rec, sagittal')
-plt.show()
 #%%
 print ("%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%")
 print ("Reconstructing with FISTA-OS-SWLS-TV using tomobar")

Readme.md

@@ -63,7 +63,6 @@ conda install -c savu-dep tomobar
 export VERSION=`date +%Y.%m` (unix) / set VERSION=2020.10 (Windows)
 conda build conda-recipe/ --numpy 1.15 --python 3.7
 conda install -c file://${CONDA_PREFIX}/conda-bld/ tomobar --force-reinstall
-conda install tomobar --use-local --force-reinstall # if Python2
 ```
 
 ### Python development environment

conda-recipe/bld.bat

@@ -3,5 +3,5 @@ ECHO VERSION Not Defined.
 exit 1
 )
 
-"%PYTHON%" -m pip install -e .
+"%PYTHON%" -m pip install .
 if errorlevel 1 exit 1

conda-recipe/build.sh

@@ -1,3 +1,3 @@
 #!/bin/bash
 
-python -m pip install -e .
+python -m pip install .

conda-recipe/conda_build_config.yaml

@@ -1,4 +1,4 @@
 python:
   - 3.10
-numpy:  
-  - 1.22
+numpy:
+  - 1.24

conda-recipe/meta.yaml

@@ -5,12 +5,22 @@ package:
   version: {{ environ['VERSION'] }}
 
 source:
-  path: ..
+  path: ../
 
 build:
   preserve_egg_dir: False
   number: 0
 
+test:
+  requires:
+    - pip
+    - pytest
+  source_files:
+    - tests/*
+  commands:
+    - pip check
+    - pytest
+
 requirements:
   host:
     - python
@@ -24,15 +34,7 @@ requirements:
     - pillow
     - scikit-image
     - pytest
-# test:
-#   requires:
-#     - pip
-#     - pytest
-#   source_files:
-#     - tests/*
-#   commands:
-#     - pip check
-#     - pytest
+
 
 about:
   home: https://github.com/dkazanc/ToMoBAR
@@ -41,4 +43,4 @@ about:
     A collection of direct and iterative methods for parallel beam tomography
 extra:
   recipe-maintainers:
-    - dkazanc
+    - dkazanc

pyproject.toml

@@ -3,7 +3,7 @@ requires = ["setuptools>=42", "wheel", "setuptools-git-versioning"]
 build-backend = "setuptools.build_meta"
 
 [tool.setuptools]
-packages = ["tomobar"]
+packages = ["tomobar", "tomobar.supp"]
 
 [tool.setuptools-git-versioning]
 enabled = true
@@ -45,7 +45,6 @@ dev = [
   "h5py"
 ]
 
-
 [project.urls]
 "Repository" = "https://github.com/dkazanc/ToMoBAR"
 

tomobar/methodsIR.py

@@ -102,15 +102,9 @@ def dict_check(self, _data_, _algorithm_, _regularisation_):
     # Students#t data model to supress artifacts
     if 'studentst_threshold' not in _data_:
         _data_['studentst_threshold'] = None
-    # threshold to produce additional weights to supress ring artifacts
-    if 'ring_weights_threshold' not in _data_:
-        _data_['ring_weights_threshold'] = None
     # defines the strength of Huber penalty to supress artifacts 1 = Huber, > 1 more strength
     if 'ring_huber_power' not in _data_:
         _data_['ring_huber_power'] = 1.75
-    # a tuple for half window sizes as [detector, angles, number of projections]
-    if 'ring_tuple_halfsizes' not in _data_:
-        _data_['ring_tuple_halfsizes'] = (9,7,9)
     # Group-Huber data model to supress full rings of the same intensity
     if 'ringGH_lambda' not in _data_:
         _data_['ringGH_lambda'] = None
@@ -278,9 +272,7 @@ class RecToolsIR:
             --OS_number # the number of subsets, if None or 1 - classical (full data) algorithm
             --huber_threshold # threshold for Huber function to apply to data model (supress outliers)
             --studentst_threshold # threshold for Students't function to apply to data model (supress outliers)
-            --ring_weights_threshold # threshold to produce additional weights to supress ring artifacts
             --ring_huber_power # defines the strength of Huber penalty to supress artifacts 1 = Huber, > 1 more penalising
-            --ring_tuple_halfsizes # a tuple for half window sizes as [detector, angles, num of projections]
             --ringGH_lambda # a parameter for Group-Huber data model to supress full rings of the same intensity
             --ringGH_accelerate # Group-Huber data model acceleration factor (use carefully to avoid divergence, 50 default)
             --beta_SWLS # a regularisation parameter for stripe-weighted LS model (given as a vector size of DetectorsDimH)
@@ -482,12 +474,6 @@ def FISTA(self, _data_, _algorithm_, _regularisation_):
                 else:
                     r = r_x - np.multiply(L_const_inv,vec)
 
-            if ((_data_['OS_number'] != 1) and (_data_['ring_weights_threshold'] is not None) and (iter_no > 0)):
-                # Ordered subset approach for a better ring model
-                res_full = self.Atools.forwproj(X_t) - _data_['projection_norm_data']
-                rings_weights = RING_WEIGHTS(np.float32(res_full), _data_['ring_tuple_halfsizes'][0], _data_['ring_tuple_halfsizes'][1], _data_['ring_tuple_halfsizes'][2])
-                ring_function_weight = np.ones(np.shape(res_full))
-                ring_function_weight[(np.where(np.abs(rings_weights) > _data_['ring_weights_threshold']))] = np.divide(_data_['ring_weights_threshold'], np.abs(rings_weights[(np.where(np.abs(rings_weights) > _data_['ring_weights_threshold']))])**_data_['ring_huber_power'])
             # loop over subsets (OS)
             for sub_ind in range(_data_['OS_number']):
                 X_old = X
@@ -519,8 +505,6 @@ def FISTA(self, _data_, _algorithm_, _regularisation_):
                             # ring removal part for Group-Huber (GH) fidelity (2D)
                             if ((_data_['ringGH_lambda'] is not None) and (iter_no > 0)):
                                 res[:,0:None] = res[:,0:None] + _data_['ringGH_accelerate']*r_x[:,0]
-                            if ((_data_['ring_weights_threshold'] is not None) and (iter_no > 0)):
-                                res = np.multiply(ring_function_weight[indVec,:],res)
                         else: # 3D
                             if (self.datafidelity == 'LS'):
                                 # 3D Least-squares (LS) data fidelity - OS (linear)
@@ -543,8 +527,6 @@ def FISTA(self, _data_, _algorithm_, _regularisation_):
                             if ((_data_['ringGH_lambda'] is not None) and (iter_no > 0)):
                                 for ang_index in range(len(indVec)):
                                     res[:,ang_index,:] = res[:,ang_index,:] + _data_['ringGH_accelerate']*r_x
-                            if ((_data_['ring_weights_threshold'] is not None) and (iter_no > 0)):
-                                res = np.multiply(ring_function_weight[:,indVec,:],res)
                 else: # CLASSICAL all-data approach
                         if (self.datafidelity == 'LS'):
                             # full residual for LS fidelity
@@ -566,12 +548,6 @@ def FISTA(self, _data_, _algorithm_, _regularisation_):
                                     res[:,ang_index,:] = res[:,ang_index,:] + _data_['ringGH_accelerate']*r_x
                                     vec = res.sum(axis = 1)
                                     r = r_x - np.multiply(L_const_inv,vec)
-                        if ((_data_['ring_weights_threshold'] is not None) and (iter_no > 0)):
-                            # Approach for a better ring model
-                            rings_weights = RING_WEIGHTS(np.float32(res), _data_['ring_tuple_halfsizes'][0], _data_['ring_tuple_halfsizes'][1], _data_['ring_tuple_halfsizes'][2])
-                            ring_function_weight = np.ones(np.shape(res))
-                            ring_function_weight[(np.where(np.abs(rings_weights) > _data_['ring_weights_threshold']))] = np.divide(_data_['ring_weights_threshold'], np.abs(rings_weights[(np.where(np.abs(rings_weights) > _data_['ring_weights_threshold']))])**_data_['ring_huber_power'])
-                            res = np.multiply(ring_function_weight,res)
                         if (self.datafidelity == 'SWLS'):
                             res = self.Atools.forwproj(X_t) - _data_['projection_norm_data']
                             if (self.geom == '2D'):
@@ -681,7 +657,7 @@ def ADMM_Atb(b):
             if (_algorithm_['nonnegativity'] == 'ENABLE'):
                 X[X < 0.0] = 0.0
             # z-update with relaxation
-            zold = z.copy();
+            zold = z.copy()
             x_hat = _algorithm_['ADMM_relax_par']*X + (1.0 - _algorithm_['ADMM_relax_par'])*zold;
             if (self.geom == '2D'):
                 x_prox_reg = (x_hat + u).reshape([self.ObjSize, self.ObjSize])