Merge branch 'dev' into optimize/vectorize-dice-metric

Author: aymuos15

.github/workflows/codeql-analysis.yml

@@ -67,8 +67,8 @@ jobs:
         rm -rf /opt/hostedtoolcache/{node,go,Ruby,Java*}
         ls -al /opt/hostedtoolcache
         rm -rf /usr/share/dotnet/
-        python -m pip install -U pip wheel
-        python -m pip install -r requirements-dev.txt
+        python -m pip install -U --no-build-isolation pip wheel wheel-stub
+        python -m pip install --no-build-isolation -r requirements-dev.txt
         BUILD_MONAI=1 ./runtests.sh --build
 
     - name: Perform CodeQL Analysis

.github/workflows/pythonapp-gpu.yml

@@ -1,14 +1,16 @@
 # Jenkinsfile.monai-premerge
 name: premerge-gpu
 
-# on:
+on:
 #   # quick tests for pull requests and the releasing branches
 #   push:
 #     branches:
 #       - main
 #       - releasing/*
 #   pull_request:
 #     types: [opened, synchronize, closed]
+  # Allows you to run this workflow manually from the Actions tab
+  workflow_dispatch:
 
 concurrency:
   # automatically cancel the previously triggered workflows when there's a newer version

docs/requirements.txt

@@ -21,7 +21,7 @@ sphinx-autodoc-typehints==1.11.1
 pandas
 einops
 transformers>=4.53.0
-mlflow>=2.12.2
+mlflow>=2.12.2,<3.13
 clearml>=1.10.0rc0
 tensorboardX
 imagecodecs; platform_system == "Linux" or platform_system == "Darwin"

monai/data/image_reader.py

@@ -1231,17 +1231,26 @@ class NumpyReader(ImageReader):
         npz_keys: if loading npz file, only load the specified keys, if None, load all the items.
             stack the loaded items together to construct a new first dimension.
         channel_dim: if not None, explicitly specify the channel dim, otherwise, treat the array as no channel.
+        allow_pickle: if True, allows loading pickled contents from NPY/NPZ files. Note that the default value of False
+            prevents the risk of remote code execution, set this to True only for loading known trusted data. If this
+            argument is False and pickled data is loaded, a ValueError will be raised.
         kwargs: additional args for `numpy.load` API except `allow_pickle`. more details about available args:
             https://numpy.org/doc/stable/reference/generated/numpy.load.html
-
     """
 
-    def __init__(self, npz_keys: KeysCollection | None = None, channel_dim: str | int | None = None, **kwargs):
+    def __init__(
+        self,
+        npz_keys: KeysCollection | None = None,
+        channel_dim: str | int | None = None,
+        allow_pickle: bool = False,
+        **kwargs,
+    ):
         super().__init__()
         if npz_keys is not None:
             npz_keys = ensure_tuple(npz_keys)
         self.npz_keys = npz_keys
         self.channel_dim = float("nan") if channel_dim == "no_channel" else channel_dim
+        self.allow_pickle = allow_pickle
         self.kwargs = kwargs
 
     def verify_suffix(self, filename: Sequence[PathLike] | PathLike) -> bool:
@@ -1267,14 +1276,25 @@ def read(self, data: Sequence[PathLike] | PathLike, **kwargs):
                 More details about available args:
                 https://numpy.org/doc/stable/reference/generated/numpy.load.html
 
+        Raises:
+            ValueError: when `self.allow_pickle` is False but loaded data contains pickled objects.
         """
         img_: list[Nifti1Image] = []
 
         filenames: Sequence[PathLike] = ensure_tuple(data)
         kwargs_ = self.kwargs.copy()
         kwargs_.update(kwargs)
         for name in filenames:
-            img = np.load(name, allow_pickle=True, **kwargs_)
+            try:
+                img = np.load(name, allow_pickle=self.allow_pickle, **kwargs_)
+            except ValueError as e:
+                # if a ValueError is raised, this is likely about pickle loading so raise an exception about this
+                raise ValueError(
+                    "MONAI default value for argument `allow_pickle` of `np.load` changed to `False`, "
+                    "explicitly pass `allow_pickle=True` as a constructor argument to NumpyReader "
+                    "to enable pickle loading."
+                ) from e
+
             if Path(name).name.endswith(".npz"):
                 # load expected items from NPZ file
                 npz_keys = list(img.keys()) if self.npz_keys is None else self.npz_keys

requirements-dev.txt

@@ -34,7 +34,7 @@ pandas
 requests
 einops
 transformers>=4.53.0
-mlflow>=2.12.2
+mlflow>=2.12.2,<3.13
 clearml>=1.10.0rc0
 matplotlib>=3.6.3
 tensorboardX

setup.cfg

@@ -71,7 +71,7 @@ all =
     pandas
     einops
     transformers>=4.53.0
-    mlflow>=2.12.2
+    mlflow>=2.12.2,<3.13
     clearml>=1.10.0rc0
     matplotlib>=3.6.3
     tensorboardX
@@ -135,7 +135,7 @@ einops =
 transformers =
     transformers>=4.36.0, <4.41.0; python_version <= '3.10'
 mlflow =
-    mlflow>=2.12.2
+    mlflow>=2.12.2,<3.13
 matplotlib =
     matplotlib>=3.6.3
 clearml =

tests/data/test_numpy_reader.py

@@ -81,7 +81,13 @@ def test_npy_pickle(self):
             np.save(filepath, test_data, allow_pickle=True)
 
             reader = NumpyReader()
+
+            with self.assertRaises(ValueError):
+                reader.get_data(reader.read(filepath))
+
+            reader = NumpyReader(allow_pickle=True)
             result = reader.get_data(reader.read(filepath))[0].item()
+
         np.testing.assert_allclose(result["test"].shape, test_data["test"].shape)
         np.testing.assert_allclose(result["test"], test_data["test"])
 
@@ -92,6 +98,11 @@ def test_kwargs(self):
             np.save(filepath, test_data, allow_pickle=True)
 
             reader = NumpyReader(mmap_mode="r")
+
+            with self.assertRaises(ValueError):
+                reader.get_data(reader.read(filepath, mmap_mode=None))
+
+            reader = NumpyReader(mmap_mode="r", allow_pickle=True)
             result = reader.get_data(reader.read(filepath, mmap_mode=None))[0].item()
         np.testing.assert_allclose(result["test"].shape, test_data["test"].shape)