Merge pull request #1034 from mantidproject/cil_20210701

Merge CIL recon

conda/meta.yaml

@@ -19,14 +19,14 @@ requirements:
     - pip
     - setuptools=49.6.0
     - sphinx=3.3.1
-    - numpy=1.20.*
+    - numpy=1.18.*
   run:
     - python=3.8.*
     - pip
     - astropy=4.2
     - scipy=1.5.3
     - scikit-image=0.18.*
-    - numpy=1.20.*
+    - numpy=1.18.*
     - tomopy=1.9.*=cuda*
     - cudatoolkit=9.2*
     - cupy
@@ -35,6 +35,10 @@ requirements:
     - h5py=3.1.0
     - sarepy=2020.07
     - psutil=5.8.*
+    - cil=21.2.*
+    - cil-astra=21.2.*
+    - ccpi-regulariser=20.9
+    - tomophantom=1.4.*
 
 build:
   number: 1

docs/release_notes/next.rst

@@ -17,6 +17,7 @@ New features
 - #1011 : Pixel size should allow setting decimal places
 - #1013 : NeXus Loading Window
 - #904 : Default to using image center for ring removal COR
+- #1025 : Add the Total Variation (TV) with Primal-Dual Hybrid Gradient (PDHG) from CIL
 
 Fixes
 -----
@@ -46,4 +47,4 @@ Developer Changes
 Dependency updates
 ------------------
 
-- pyqtgraph 0.12, scikit-image 0.18, tomopy 1.9, numpy 1.20
+- pyqtgraph 0.12, scikit-image 0.18, tomopy 1.9, numpy 1.18

environment-dev.yml

@@ -4,6 +4,7 @@ channels:
   - dtasev
   - astra-toolbox/label/dev
   - conda-forge
+  - ccpi
   - defaults
 dependencies:
   - mantidimaging

environment.yml

@@ -4,6 +4,7 @@ channels:
   - dtasev
   - astra-toolbox/label/dev
   - conda-forge
+  - ccpi
   - defaults
 # Dependencies that can be installed with conda should be in conda/meta.yaml
 dependencies:

mantidimaging/core/io/loader/img_loader.py

@@ -4,10 +4,11 @@
 This module handles the loading of FIT, FITS, TIF, TIFF
 """
 import os
-from typing import Tuple, Optional, List, Callable, Union
+from typing import Tuple, Optional, List, Callable, Union, TYPE_CHECKING
 
 import numpy as np
-import numpy.typing as npt
+if TYPE_CHECKING:
+    import numpy.typing as npt
 
 from mantidimaging.core.data import Images
 from mantidimaging.core.io.utility import get_file_names, get_prefix
@@ -25,7 +26,7 @@ def execute(load_func: Callable[[str], np.ndarray],
             dark_before_path: Optional[str],
             dark_after_path: Optional[str],
             img_format: str,
-            dtype: npt.DTypeLike,
+            dtype: 'npt.DTypeLike',
             indices: Union[List[int], Indices, None],
             progress: Optional[Progress] = None) -> Dataset:
     """
@@ -88,7 +89,7 @@ def __init__(self,
                  load_func: Callable[[str], np.ndarray],
                  img_format: str,
                  img_shape: Tuple[int, ...],
-                 data_dtype: npt.DTypeLike,
+                 data_dtype: 'npt.DTypeLike',
                  indices: Union[List[int], Indices, None],
                  progress: Optional[Progress] = None):
         self.load_func = load_func

mantidimaging/core/io/loader/loader.py

@@ -4,10 +4,11 @@
 from dataclasses import dataclass
 from logging import getLogger, Logger
 from pathlib import Path
-from typing import Tuple, List, Optional, Union
+from typing import Tuple, List, Optional, Union, TYPE_CHECKING
 
 import numpy as np
-import numpy.typing as npt
+if TYPE_CHECKING:
+    import numpy.typing as npt
 
 from mantidimaging.core.data import Images
 from mantidimaging.core.data.dataset import Dataset
@@ -88,7 +89,7 @@ class FileInformation:
 def read_in_file_information(input_path: str,
                              in_prefix: str = '',
                              in_format: str = DEFAULT_IO_FILE_FORMAT,
-                             data_dtype: npt.DTypeLike = np.float32) -> FileInformation:
+                             data_dtype: 'npt.DTypeLike' = np.float32) -> FileInformation:
     input_file_names = get_file_names(input_path, in_format, in_prefix)
     dataset = load(input_path,
                    in_prefix=in_prefix,
@@ -110,7 +111,7 @@ def load_log(log_file: str) -> IMATLogFile:
         return IMATLogFile(f.readlines(), log_file)
 
 
-def load_p(parameters: ImageParameters, dtype: npt.DTypeLike, progress: Progress) -> Images:
+def load_p(parameters: ImageParameters, dtype: 'npt.DTypeLike', progress: Progress) -> Images:
     return load(input_path=parameters.input_path,
                 in_prefix=parameters.prefix,
                 in_format=parameters.format,
@@ -134,7 +135,7 @@ def load(input_path: Optional[str] = None,
          input_path_dark_after: Optional[str] = None,
          in_prefix: str = '',
          in_format: str = DEFAULT_IO_FILE_FORMAT,
-         dtype: npt.DTypeLike = np.float32,
+         dtype: 'npt.DTypeLike' = np.float32,
          file_names: Optional[List[str]] = None,
          indices: Optional[Union[List[int], Indices]] = None,
          progress: Optional[Progress] = None) -> Dataset:

mantidimaging/core/io/loader/stack_loader.py

@@ -1,9 +1,10 @@
 # Copyright (C) 2021 ISIS Rutherford Appleton Laboratory UKRI
 # SPDX - License - Identifier: GPL-3.0-or-later
-from typing import Optional, Callable, Union, List, Tuple
+from typing import Optional, Callable, Union, List, Tuple, TYPE_CHECKING
 
 import numpy as np
-import numpy.typing as npt
+if TYPE_CHECKING:
+    import numpy.typing as npt
 
 from mantidimaging.core.data import Images
 from mantidimaging.core.parallel import utility as pu
@@ -36,7 +37,7 @@ def do_stack_load_seq(data: np.ndarray, new_data: np.ndarray, img_shape: Tuple[i
 
 def execute(load_func: Callable[[str], np.ndarray],
             file_name: str,
-            dtype: npt.DTypeLike,
+            dtype: 'npt.DTypeLike',
             name: str,
             indices: Union[List[int], Indices, None] = None,
             progress: Optional[Progress] = None) -> Images:

mantidimaging/core/parallel/utility.py

@@ -8,10 +8,11 @@
 from logging import getLogger
 from multiprocessing import Array
 from multiprocessing.pool import Pool
-from typing import List, Tuple, Type, Union
+from typing import List, Tuple, Type, Union, TYPE_CHECKING
 
 import numpy as np
-import numpy.typing as npt
+if TYPE_CHECKING:
+    import numpy.typing as npt
 
 from mantidimaging.core.utility.memory_usage import system_free_memory
 from mantidimaging.core.utility.progress_reporting import Progress
@@ -27,7 +28,7 @@ def enough_memory(shape, dtype):
     return full_size_KB(shape=shape, axis=0, dtype=dtype) < system_free_memory().kb()
 
 
-def create_array(shape: Tuple[int, ...], dtype: npt.DTypeLike = np.float32) -> np.ndarray:
+def create_array(shape: Tuple[int, ...], dtype: 'npt.DTypeLike' = np.float32) -> np.ndarray:
     """
     Create an array, either in a memory file (if name provided), or purely in memory (if name is None)
 
@@ -44,7 +45,7 @@ def create_array(shape: Tuple[int, ...], dtype: npt.DTypeLike = np.float32) -> n
     return _create_shared_array(shape, dtype)
 
 
-def _create_shared_array(shape, dtype: npt.DTypeLike = np.float32) -> np.ndarray:
+def _create_shared_array(shape, dtype: 'npt.DTypeLike' = np.float32) -> np.ndarray:
     ctype: SimpleCType = ctypes.c_float  # default to numpy float32 / C type float
     if dtype == np.uint8 or dtype == 'uint8':
         ctype = ctypes.c_uint8

mantidimaging/core/reconstruct/__init__.py

@@ -1,14 +1,16 @@
 # Copyright (C) 2021 ISIS Rutherford Appleton Laboratory UKRI
 # SPDX - License - Identifier: GPL-3.0-or-later
 
-from typing import Union
-
 from .astra_recon import AstraRecon
 from .tomopy_recon import TomopyRecon
+from .cil_recon import CILRecon
+from .base_recon import BaseRecon
 
 
-def get_reconstructor_for(algorithm: str) -> Union[AstraRecon, TomopyRecon]:
+def get_reconstructor_for(algorithm: str) -> BaseRecon:
     if algorithm == "gridrec":
         return TomopyRecon()
+    if algorithm.startswith("CIL"):
+        return CILRecon()
     else:
         return AstraRecon()

mantidimaging/core/reconstruct/astra_recon.py

@@ -104,7 +104,7 @@ def single_sino(sino: np.ndarray, cor: ScalarCoR, proj_angles: ProjectionAngles,
                     recon_params: ReconstructionParameters) -> np.ndarray:
         assert sino.ndim == 2, "Sinogram must be a 2D image"
 
-        sino = BaseRecon.sino_recon_prep(sino)
+        sino = BaseRecon.negative_log(sino)
         image_width = sino.shape[1]
         vectors = vec_geom_init2d(proj_angles, 1.0, cor.to_vec(image_width).value)
         vol_geom = astra.create_vol_geom((image_width, image_width))

mantidimaging/core/reconstruct/base_recon.py

@@ -16,8 +16,12 @@
 
 class BaseRecon:
     @staticmethod
-    def sino_recon_prep(sino: np.ndarray):
-        return -np.log(np.maximum(sino, MIN_PIXEL_VALUE))
+    def find_cor(images: Images, slice_idx: int, start_cor: float, recon_params: ReconstructionParameters) -> float:
+        raise NotImplementedError("Base class call")
+
+    @staticmethod
+    def negative_log(data: np.ndarray) -> np.ndarray:
+        return -np.log(np.maximum(data, MIN_PIXEL_VALUE))
 
     @staticmethod
     def single_sino(sino: np.ndarray, cor: ScalarCoR, proj_angles: ProjectionAngles,
@@ -53,4 +57,4 @@ def full(images: Images,
 
     @staticmethod
     def allowed_filters() -> List[str]:
-        raise NotImplementedError("Base class call")
+        return []