Merge branch 'dev' into compose-resampling

Signed-off-by: Wenqi Li <[email protected]>

Author: wyli

.github/workflows/cron.yml

@@ -50,7 +50,7 @@ jobs:
         python -c 'import torch; print(torch.rand(5, 3, device=torch.device("cuda:0")))'
         BUILD_MONAI=1 ./runtests.sh --build --coverage --unittests --disttests  # unit tests with coverage report
         BUILD_MONAI=1 ./runtests.sh --build --coverage --net  # integration tests with coverage report
-        coverage xml
+        coverage xml --ignore-errors
         if pgrep python; then pkill python; fi
     - name: Upload coverage
       uses: codecov/codecov-action@v1
@@ -93,7 +93,7 @@ jobs:
         python -c 'import torch; print(torch.rand(5, 3, device=torch.device("cuda:0")))'
         BUILD_MONAI=1 ./runtests.sh --build --coverage --unittests --disttests  # unit tests with coverage report
         BUILD_MONAI=1 ./runtests.sh --build --coverage --net  # integration tests with coverage report
-        coverage xml
+        coverage xml --ignore-errors
         if pgrep python; then pkill python; fi
     - name: Upload coverage
       uses: codecov/codecov-action@v1
@@ -192,7 +192,7 @@ jobs:
         ngc --version
         BUILD_MONAI=1 ./runtests.sh --build --coverage --pytype --unittests --disttests  # unit tests with pytype checks, coverage report
         BUILD_MONAI=1 ./runtests.sh --build --coverage --net  # integration tests with coverage report
-        coverage xml
+        coverage xml --ignore-errors
         if pgrep python; then pkill python; fi
     - name: Upload coverage
       uses: codecov/codecov-action@v1

.github/workflows/pythonapp-gpu.yml

@@ -142,7 +142,7 @@ jobs:
           # test the clang-format tool downloading once
           coverage run -m tests.clang_format_utils
         fi
-        coverage xml
+        coverage xml --ignore-errors
         if pgrep python; then pkill python; fi
       shell: bash
     - name: Upload coverage

.github/workflows/setupapp.yml

@@ -61,7 +61,7 @@ jobs:
         python -c 'import torch; print(torch.rand(5, 3, device=torch.device("cuda:0")))'
         BUILD_MONAI=1 ./runtests.sh --build --coverage --unittests --disttests  # unit tests with coverage report
         BUILD_MONAI=1 ./runtests.sh --build --coverage --net  # integration tests with coverage report
-        coverage xml
+        coverage xml --ignore-errors
         if pgrep python; then pkill python; fi
       shell: bash
     - name: Upload coverage
@@ -105,7 +105,7 @@ jobs:
         python -m pip list
         python -c 'import torch; print(torch.__version__); print(torch.rand(5,3))'
         BUILD_MONAI=1 ./runtests.sh --build --quick --unittests --disttests
-        coverage xml
+        coverage xml --ignore-errors
     - name: Upload coverage
       uses: codecov/codecov-action@v1
       with:

monai/engines/evaluator.py

@@ -161,7 +161,7 @@ def get_stats(self, *vars):
         stats = {
             ESKeys.RANK: self.state.rank,
             ESKeys.BEST_VALIDATION_EPOCH: self.state.best_metric_epoch,
-            ESKeys.BEST_VALIDATION_METRTC: self.state.best_metric,
+            ESKeys.BEST_VALIDATION_METRIC: self.state.best_metric,
         }
         for k in vars:
             stats[k] = getattr(self.state, k, None)

monai/fl/utils/exchange_object.py

@@ -40,6 +40,7 @@ def __init__(
         self.metrics = metrics
         self.weight_type = weight_type
         self.statistics = statistics
+        self._summary: Dict = {}
 
     @property
     def metrics(self):
@@ -81,5 +82,26 @@ def is_valid_weights(self):
             return False
         return True
 
+    def _add_to_summary(self, key, value):
+        if value:
+            if isinstance(value, dict):
+                self._summary[key] = len(value)
+            elif isinstance(value, WeightType):
+                self._summary[key] = value
+            else:
+                self._summary[key] = type(value)
+
     def summary(self):
-        return dir(self)
+        self._summary.update(self)
+        for k, v in zip(
+            ["weights", "optim", "metrics", "weight_type", "statistics"],
+            [self.weights, self.optim, self.metrics, self.weight_type, self.statistics],
+        ):
+            self._add_to_summary(k, v)
+        return self._summary
+
+    def __repr__(self):
+        return str(self.summary())
+
+    def __str__(self):
+        return str(self.summary())

monai/transforms/spatial/array.py

@@ -201,31 +201,33 @@ def __call__(self, img: NdarrayOrTensor) -> NdarrayOrTensor:
 
 class EnsureChannelFirst(Transform):
     """
-    Automatically adjust or add the channel dimension of input data to ensure `channel_first` shape.
+    Adjust or add the channel dimension of input data to ensure `channel_first` shape.
 
     This extracts the `original_channel_dim` info from provided meta_data dictionary or MetaTensor input. This value
     should state which dimension is the channel dimension so that it can be moved forward, or contain "no_channel" to
     state no dimension is the channel and so a 1-size first dimension is to be added.
 
     Args:
         strict_check: whether to raise an error when the meta information is insufficient.
-        add_channel_default: If True and the input image `img` is not a MetaTensor and `meta_dict` is not given,
-            or `img` is a MetaTensor but doesn't specify channel dimension, use the value is `no_channel`
+        channel_dim: If the input image `img` is not a MetaTensor or `meta_dict` is not given,
+            this argument can be used to specify the original channel dimension (integer) of the input array.
+            If the input array doesn't have a channel dim, this value should be ``'no_channel'`` (default).
+            If this is set to `None`, this class relies on `img` or `meta_dict` to provide the channel dimension.
     """
 
     backend = [TransformBackends.TORCH, TransformBackends.NUMPY]
 
-    def __init__(self, strict_check: bool = True, add_channel_default=True):
+    def __init__(self, strict_check: bool = True, channel_dim: Union[None, str, int] = "no_channel"):
         self.strict_check = strict_check
-        self.add_channel_default = add_channel_default
+        self.input_channel_dim = channel_dim
 
     def __call__(self, img: torch.Tensor, meta_dict: Optional[Mapping] = None) -> torch.Tensor:
         """
         Apply the transform to `img`.
         """
         if not isinstance(img, MetaTensor) and not isinstance(meta_dict, Mapping):
-            if not self.add_channel_default:
-                msg = "Metadata not available and default add channel is disabled, EnsureChannelFirst is not in use."
+            if self.input_channel_dim is None:
+                msg = "Metadata not available and channel_dim=None, EnsureChannelFirst is not in use."
                 if self.strict_check:
                     raise ValueError(msg)
                 warnings.warn(msg)
@@ -236,16 +238,18 @@ def __call__(self, img: torch.Tensor, meta_dict: Optional[Mapping] = None) -> to
         if isinstance(img, MetaTensor):
             meta_dict = img.meta
 
-        channel_dim = meta_dict.get("original_channel_dim")  # type: ignore
+        channel_dim = meta_dict.get("original_channel_dim", None) if isinstance(meta_dict, Mapping) else None
 
         if channel_dim is None:
-            if not self.add_channel_default:
-                msg = "Unknown original_channel_dim in the MetaTensor meta dict or `meta_dict`."
+            if self.input_channel_dim is None:
+                msg = "Unknown original_channel_dim in the MetaTensor meta dict or `meta_dict` or `channel_dim`."
                 if self.strict_check:
                     raise ValueError(msg)
                 warnings.warn(msg)
                 return img
-            meta_dict["original_channel_dim"] = channel_dim = "no_channel"  # type: ignore
+            channel_dim = self.input_channel_dim
+            if isinstance(meta_dict, dict):
+                meta_dict["original_channel_dim"] = self.input_channel_dim
 
         if channel_dim == "no_channel":
             result = img[None]

monai/transforms/spatial/dictionary.py

@@ -301,19 +301,21 @@ def __init__(
         meta_key_postfix: str = DEFAULT_POST_FIX,
         strict_check: bool = True,
         allow_missing_keys: bool = False,
-        add_channel_default=True,
+        channel_dim="no_channel",
     ) -> None:
         """
         Args:
             keys: keys of the corresponding items to be transformed.
                 See also: :py:class:`monai.transforms.compose.MapTransform`
             strict_check: whether to raise an error when the meta information is insufficient.
             allow_missing_keys: don't raise exception if key is missing.
-            add_channel_default: If True and the input image `img` is not a MetaTensor and `meta_dict` is not given,
-                or `img` is a MetaTensor but doesn't specify channel dimension, use the value is `no_channel`
+            channel_dim: If the input image `img` is not a MetaTensor or `meta_dict` is not given,
+                this argument can be used to specify the original channel dimension (integer) of the input array.
+                If the input array doesn't have a channel dim, this value should be ``'no_channel'`` (default).
+                If this is set to `None`, this class relies on `img` or `meta_dict` to provide the channel dimension.
         """
         super().__init__(keys, allow_missing_keys)
-        self.adjuster = EnsureChannelFirst(strict_check=strict_check, add_channel_default=add_channel_default)
+        self.adjuster = EnsureChannelFirst(strict_check=strict_check, channel_dim=channel_dim)
         self.meta_keys = ensure_tuple_rep(meta_keys, len(self.keys))
         self.meta_key_postfix = ensure_tuple_rep(meta_key_postfix, len(self.keys))
 

monai/transforms/utility/array.py

@@ -586,6 +586,9 @@ def create_grid(
     """
     compute a `spatial_size` mesh.
 
+        - when ``homogeneous=True``, the output shape is (N+1, dim_size_1, dim_size_2, ..., dim_size_N)
+        - when ``homogeneous=False``, the output shape is (N, dim_size_1, dim_size_2, ..., dim_size_N)
+
     Args:
         spatial_size: spatial size of the grid.
         spacing: same len as ``spatial_size``, defaults to 1.0 (dense grid).

monai/transforms/utility/dictionary.py

@@ -36,12 +36,14 @@
     MetaKeys,
     Method,
     MetricReduction,
+    NdimageMode,
     NumpyPadMode,
     PostFix,
     ProbMapKeys,
     PytorchPadMode,
     SkipMode,
     SpaceKeys,
+    SplineMode,
     StrEnum,
     TraceKeys,
     TransformBackends,

monai/transforms/utils.py

@@ -19,10 +19,12 @@
     "StrEnum",
     "NumpyPadMode",
     "GridSampleMode",
+    "SplineMode",
     "InterpolateMode",
     "UpsampleMode",
     "BlendMode",
     "PytorchPadMode",
+    "NdimageMode",
     "GridSamplePadMode",
     "Average",
     "MetricReduction",
@@ -92,6 +94,22 @@ class NumpyPadMode(StrEnum):
     EMPTY = "empty"
 
 
+class NdimageMode(StrEnum):
+    """
+    The available options determine how the input array is extended beyond its boundaries when interpolating.
+    See also: https://docs.scipy.org/doc/scipy/reference/generated/scipy.ndimage.map_coordinates.html
+    """
+
+    REFLECT = "reflect"
+    GRID_MIRROR = "grid-mirror"
+    CONSTANT = "constant"
+    GRID_CONSTANT = "grid-constant"
+    NEAREST = "nearest"
+    MIRROR = "mirror"
+    GRID_WRAP = "grid-wrap"
+    WRAP = "wrap"
+
+
 class GridSampleMode(StrEnum):
     """
     See also: https://pytorch.org/docs/stable/generated/torch.nn.functional.grid_sample.html
@@ -110,6 +128,21 @@ class GridSampleMode(StrEnum):
     BICUBIC = "bicubic"
 
 
+class SplineMode(StrEnum):
+    """
+    Order of spline interpolation.
+
+    See also: https://docs.scipy.org/doc/scipy/reference/generated/scipy.ndimage.map_coordinates.html
+    """
+
+    ZERO = 0
+    ONE = 1
+    TWO = 2
+    THREE = 3
+    FOUR = 4
+    FIVE = 5
+
+
 class InterpolateMode(StrEnum):
     """
     See also: https://pytorch.org/docs/stable/generated/torch.nn.functional.interpolate.html
@@ -493,4 +526,4 @@ class EngineStatsKeys(StrEnum):
     TOTAL_EPOCHS = "total_epochs"
     TOTAL_ITERATIONS = "total_iterations"
     BEST_VALIDATION_EPOCH = "best_validation_epoch"
-    BEST_VALIDATION_METRTC = "best_validation_metric"
+    BEST_VALIDATION_METRIC = "best_validation_metric"

monai/utils/__init__.py

@@ -175,6 +175,13 @@ def convert_to_numpy(data, dtype: DtypeLike = None, wrap_sequence: bool = False)
     elif has_cp and isinstance(data, cp_ndarray):
         data = cp.asnumpy(data).astype(dtype, copy=False)
     elif isinstance(data, (np.ndarray, float, int, bool)):
+        # Convert into a contiguous array first if the current dtype's size is smaller than the target dtype's size.
+        # This help improve the performance because (convert to contiguous array) -> (convert dtype) is faster
+        # than (convert dtype) -> (convert to contiguous array) when src dtype (e.g., uint8) is smaller than
+        # target dtype(e.g., float32) and we are going to convert it to contiguous array anyway later in this
+        # method.
+        if isinstance(data, np.ndarray) and data.ndim > 0 and data.dtype.itemsize < np.dtype(dtype).itemsize:
+            data = np.ascontiguousarray(data)
         data = np.asarray(data, dtype=dtype)
     elif isinstance(data, list):
         list_ret = [convert_to_numpy(i, dtype=dtype) for i in data]

monai/utils/enums.py

@@ -136,6 +136,7 @@ def run_testsuit():
         "test_rand_zoom",
         "test_rand_zoomd",
         "test_randtorchvisiond",
+        "test_resample_backends",
         "test_resize",
         "test_resized",
         "test_resample_to_match",

monai/utils/type_conversion.py

@@ -11,7 +11,6 @@
 
 import json
 import os
-import subprocess
 import tempfile
 import unittest
 
@@ -20,7 +19,7 @@
 from monai.bundle import ConfigParser
 from monai.data import load_net_with_metadata
 from monai.networks import save_state
-from tests.utils import skip_if_windows
+from tests.utils import command_line_tests, skip_if_windows
 
 TEST_CASE_1 = [""]
 
@@ -49,7 +48,7 @@ def test_export(self, key_in_ckpt):
             cmd = ["coverage", "run", "-m", "monai.bundle", "ckpt_export", "network_def", "--filepath", ts_file]
             cmd += ["--meta_file", meta_file, "--config_file", f"['{config_file}','{def_args_file}']", "--ckpt_file"]
             cmd += [ckpt_file, "--key_in_ckpt", key_in_ckpt, "--args_file", def_args_file]
-            subprocess.check_call(cmd)
+            command_line_tests(cmd)
             self.assertTrue(os.path.exists(ts_file))
 
             _, metadata, extra_files = load_net_with_metadata(

tests/min_tests.py

@@ -11,7 +11,6 @@
 
 import json
 import os
-import subprocess
 import tempfile
 import unittest
 
@@ -21,7 +20,13 @@
 import monai.networks.nets as nets
 from monai.apps import check_hash
 from monai.bundle import ConfigParser, load
-from tests.utils import SkipIfBeforePyTorchVersion, skip_if_downloading_fails, skip_if_quick, skip_if_windows
+from tests.utils import (
+    SkipIfBeforePyTorchVersion,
+    command_line_tests,
+    skip_if_downloading_fails,
+    skip_if_quick,
+    skip_if_windows,
+)
 
 TEST_CASE_1 = [
     ["model.pt", "model.ts", "network.json", "test_output.pt", "test_input.pt"],
@@ -64,7 +69,7 @@ def test_download_bundle(self, bundle_files, bundle_name, repo, hash_val):
             with tempfile.TemporaryDirectory() as tempdir:
                 cmd = ["coverage", "run", "-m", "monai.bundle", "download", "--name", bundle_name, "--source", "github"]
                 cmd += ["--bundle_dir", tempdir, "--repo", repo, "--progress", "False"]
-                subprocess.check_call(cmd)
+                command_line_tests(cmd)
                 for file in bundle_files:
                     file_path = os.path.join(tempdir, bundle_name, file)
                     self.assertTrue(os.path.exists(file_path))
@@ -83,7 +88,7 @@ def test_url_download_bundle(self, bundle_files, bundle_name, url, hash_val):
                 parser.export_config_file(config=def_args, filepath=def_args_file)
                 cmd = ["coverage", "run", "-m", "monai.bundle", "download", "--args_file", def_args_file]
                 cmd += ["--url", url]
-                subprocess.check_call(cmd)
+                command_line_tests(cmd)
                 for file in bundle_files:
                     file_path = os.path.join(tempdir, bundle_name, file)
                     self.assertTrue(os.path.exists(file_path))

tests/test_bundle_ckpt_export.py

@@ -10,14 +10,13 @@
 # limitations under the License.
 
 import os
-import subprocess
 import tempfile
 import unittest
 
 import torch
 
 from monai.networks.nets import UNet
-from tests.utils import skip_if_windows
+from tests.utils import command_line_tests, skip_if_windows
 
 
 @skip_if_windows
@@ -30,7 +29,7 @@ def test_bundle(self):
             bundle_root = tempdir + "/test_bundle"
 
             cmd = ["coverage", "run", "-m", "monai.bundle", "init_bundle", bundle_root, tempdir + "/test.pt"]
-            subprocess.check_call(cmd)
+            command_line_tests(cmd)
 
             self.assertTrue(os.path.exists(bundle_root + "/configs/metadata.json"))
             self.assertTrue(os.path.exists(bundle_root + "/configs/inference.json"))