Shepp logan phantom

direct/cli/predict.py

@@ -16,14 +16,16 @@ def register_parser(parser: argparse._SubParsersAction):
         Examples:
         ---------
         Run on single machine:
-            $ direct predict <data_root> <output_directory> --cfg <cfg_path_or_url> --checkpoint <checkpoint_path_or_url>
-                            --num-gpus <num_gpus> [--other-flag-args <other_flags>]
+            $ direct predict <output_directory> --cfg <cfg_path_or_url> --checkpoint <checkpoint_path_or_url> \
+                            --num-gpus <num_gpus> [--data-root <data_root>] [--other-flag-args <other_flags>]
 
         Run on multiple machines:
-            (machine0)$ direct predict <data_root> <output_directory> --cfg <cfg_path_or_url> --checkpoint
-                            <checkpoint_path_or_url> --machine-rank 0 --num-machines 2 --dist-url <URL> [--other-flag-args <other_flags>]
-            (machine1)$ direct predict <data_root> <output_directory> --cfg <cfg_path_or_url> --checkpoint
-                            <checkpoint_path_or_url> --machine-rank 1 --num-machines 2 --dist-url <URL> [--other-flag-args <other_flags>]
+            (machine0)$ direct predict <output_directory> --cfg <cfg_path_or_url> --checkpoint \
+                            <checkpoint_path_or_url> --machine-rank 0 --num-machines 2 [--data-root <data_root>] \
+                                [--dist-url <URL> --other-flag-args <other_flags>]
+            (machine1)$ direct predict <output_directory> --cfg <cfg_path_or_url> --checkpoint \
+                            <checkpoint_path_or_url> --machine-rank 1 --num-machines 2 [--data-root <data_root>] \
+                                [--dist-url <URL> --other-flag-args <other_flags>]
         """
     common_parser = Args(add_help=False)
     predict_parser = parser.add_parser(
@@ -33,8 +35,13 @@ def register_parser(parser: argparse._SubParsersAction):
         epilog=epilog,
         formatter_class=argparse.RawDescriptionHelpFormatter,
     )
-    predict_parser.add_argument("data_root", type=pathlib.Path, help="Path to the inference data directory.")
     predict_parser.add_argument("output_directory", type=pathlib.Path, help="Path to the output directory.")
+    predict_parser.add_argument(
+        "--data-root",
+        type=pathlib.Path,
+        help="Path to the inference data directory (if required by inference dataset).",
+        required=False,
+    )
     predict_parser.add_argument(
         "--cfg",
         dest="cfg_file",

direct/cli/train.py

@@ -15,10 +15,10 @@ def register_parser(parser: argparse._SubParsersAction):
         Examples:
         ---------
         Run on single machine:
-            $ direct train training_set validation_set experiment_dir --num-gpus 8 --cfg cfg.yaml
+            $ direct train experiment_dir --num-gpus 8 --cfg cfg.yaml [--training-root training_set --validation-root validation_set]
         Run on multiple machines:
-            (machine0)$ direct train training_set validation_set experiment_dir --machine-rank 0 --num-machines 2 --dist-url <URL> [--other-flags]
-            (machine1)$ direct train training_set validation_set experiment_dir --machine-rank 1 --num-machines 2 --dist-url <URL> [--other-flags]
+            (machine0)$ direct train experiment_dir --machine-rank 0 --num-machines 2 --dist-url <URL> [--training-root training_set --validation-root validation_set] [--other-flags]
+            (machine1)$ direct train experiment_dir --machine-rank 1 --num-machines 2 --dist-url <URL> [--training-root training_set --validation-root validation_set] [--other-flags]
         """
     common_parser = Args(add_help=False)
     train_parser = parser.add_parser(
@@ -28,14 +28,15 @@ def register_parser(parser: argparse._SubParsersAction):
         epilog=epilog,
         formatter_class=argparse.RawDescriptionHelpFormatter,
     )
-
-    train_parser.add_argument("training_root", type=pathlib.Path, help="Path to the training data.")
-    train_parser.add_argument("validation_root", type=pathlib.Path, help="Path to the validation data.")
     train_parser.add_argument(
         "experiment_dir",
         type=pathlib.Path,
         help="Path to the experiment directory.",
     )
+    train_parser.add_argument("--training-root", type=pathlib.Path, help="Path to the training data.", required=False)
+    train_parser.add_argument(
+        "--validation-root", type=pathlib.Path, help="Path to the validation data.", required=False
+    )
     train_parser.add_argument(
         "--cfg",
         dest="cfg_file",

direct/cli/utils.py

@@ -1,9 +1,11 @@
 # coding=utf-8
 # Copyright (c) DIRECT Contributors
+
 import argparse
 import pathlib
 import sys
 
+from direct.types import FileOrUrl, PathOrString
 from direct.utils.io import check_is_valid_url
 
 
@@ -14,12 +16,14 @@ def is_file(path):
     raise argparse.ArgumentTypeError(f"{path} is not a valid file or url.")
 
 
-def file_or_url(path):
+def file_or_url(path: PathOrString) -> FileOrUrl:
     if check_is_valid_url(path):
-        return path
+        return FileOrUrl(path)
+
     path = pathlib.Path(path)
     if path.is_file():
-        return path
+        return FileOrUrl(path)
+
     raise argparse.ArgumentTypeError(f"{path} is not a valid file or url.")
 
 

direct/common/subsample.py

@@ -126,8 +126,11 @@ def mask_func(self, shape, return_acs=False, seed=None):
         The mask selects a subset of columns from the input k-space data. If the k-space data has N
         columns, the mask picks out:
 
-            #.  :math:`N_{\text{low\_freqs}} = (N \times \text{center_fraction})`  columns in the center corresponding to low-frequencies.  # pylint: disable=line-too-long
-            #.  The other columns are selected uniformly at random with a probability equal to: :math:`\text{prob} = (N / \text{acceleration} - N_{\text{low\_freqs}}) / (N - N_{\text{low\_freqs}})`. This ensures that the expected number of columns selected is equal to (N / acceleration)  # pylint: disable=line-too-long
+            #.  :math:`N_{\text{low freqs}} = (N \times \text{center_fraction})`  columns in the center corresponding
+                to low-frequencies.
+            #.  The other columns are selected uniformly at random with a probability equal to:
+                :math:`\text{prob} = (N / \text{acceleration} - N_{\text{low freqs}}) / (N - N_{\text{low freqs}})`.
+                This ensures that the expected number of columns selected is equal to (N / acceleration).
 
         It is possible to use multiple center_fractions and accelerations, in which case one possible
         (center_fraction, acceleration) is chosen uniformly at random each time the MaskFunc object is
@@ -204,15 +207,18 @@ def mask_func(self, shape, return_acs=False, seed=None):
         FastMRIEquispacedMaskFunc creates a sub-sampling mask of a given shape. The mask selects a subset of columns
         from the input k-space data. If the k-space data has N columns, the mask picks out:
 
-            #.  :math:`N_{\text{low\_freqs}} = (N \times \text{center_fraction})` columns in the center corresponding to low-frequencies.  # pylint: disable=line-too-long
-            #.  The other columns are selected with equal spacing at a proportion that reaches the desired acceleration rate taking into consideration the number of low frequencies. This ensures that the expected number of columns selected is equal to :math:`\frac{N}{\text{acceleration}}`.  # pylint: disable=line-too-long
+            #.  :math:`N_{\text{low freqs}} = (N \times \text{center_fraction})` columns in the center corresponding
+                to low-frequencies.
+            #.  The other columns are selected with equal spacing at a proportion that reaches the desired acceleration
+                rate taking into consideration the number of low frequencies. This ensures that the expected number of
+                columns selected is equal to :math:`\frac{N}{\text{acceleration}}`.
 
         It is possible to use multiple center_fractions and accelerations, in which case one possible
         (center_fraction, acceleration) is chosen uniformly at random each time the EquispacedMaskFunc object is called.
 
-        Note that this function may not give equispaced samples (documented in https://github.com/facebookresearch/fastMRI/issues/54),
-        which will require modifications to standard GRAPPA approaches. Nonetheless, this aspect of the function has
-        been preserved to match the public multicoil data.
+        Note that this function may not give equispaced samples (documented in
+        https://github.com/facebookresearch/fastMRI/issues/54), which will require modifications to standard GRAPPA
+        approaches. Nonetheless, this aspect of the function has been preserved to match the public multicoil data.
 
         Parameters
         ----------
@@ -374,7 +380,9 @@ class CIRCUSMaskFunc(BaseMaskFunc):
     References
     ----------
 
-    .. [1] Liu J, Saloner D. Accelerated MRI with CIRcular Cartesian UnderSampling (CIRCUS): a variable density Cartesian sampling strategy for compressed sensing and parallel imaging. Quant Imaging Med Surg. 2014 Feb;4(1):57-67. doi: 10.3978/j.issn.2223-4292.2014.02.01. PMID: 24649436; PMCID: PMC3947985.
+    .. [1] Liu J, Saloner D. Accelerated MRI with CIRcular Cartesian UnderSampling (CIRCUS): a variable density
+        Cartesian sampling strategy for compressed sensing and parallel imaging. Quant Imaging Med Surg.
+        2014 Feb;4(1):57-67. doi: 10.3978/j.issn.2223-4292.2014.02.01. PMID: 24649436; PMCID: PMC3947985.
     """
 
     def __init__(