Updated code due to dependencies' API changes

README.md

@@ -38,7 +38,8 @@ pip install monai-deploy-app-sdk
 
 ### Prerequisites
 
-- This SDK depends on [NVIDIA Holoscan SDK](https://pypi.org/project/holoscan/) for its core implementation as well as its CLI, hence inherits its prerequisites, e.g. Ubuntu 22.04 with glibc 2.35 on X86-64 and NVIDIA dGPU drivers version 535 or above.
+- This SDK depends on [NVIDIA Holoscan SDK](https://pypi.org/project/holoscan/) for its core implementation as well as its CLI, hence inherits its prerequisites, e.g. Ubuntu 22.04 with glibc 2.35 on X86-64 and NVIDIA dGPU drivers version 535 or above. Important to note that `holoscan` and `holoscan-cli` up to v4.2 are compatible.
+- Key runtime dependencies also include [nvidia-nvimgcodec](https://pypi.org/project/nvidia-nvimgcodec-cu12/) and its own dependencies for GPU accecelerated DICOM image decoding.
 - [CUDA 12.2](https://developer.nvidia.com/cuda-12-2-0-download-archive) or above is required along with a supported NVIDIA GPU with at least 8GB of video RAM.
 - If inference is not used in an example application and a GPU is not installed, at least [CUDA 12 runtime](https://pypi.org/project/nvidia-cuda-runtime-cu12/) is required, as this is one of the requirements of Holoscan SDK. In addition, the `LIB_LIBRARY_PATH` must be set to include the installed shared library, e.g. in a Python 3.10 env, ```export LD_LIBRARY_PATH=`pwd`/.venv/lib/python3.10/site-packages/nvidia/cuda_runtime/lib:$LD_LIBRARY_PATH```
 - Python: 3.10 to 3.13
@@ -80,20 +81,13 @@ Tutorials are provided to help getting started with the App SDK, to name but a f
 
 #### [2) Creating MedNIST Classifier app](https://monai.readthedocs.io/projects/monai-deploy-app-sdk/en/stable/getting_started/tutorials/mednist_app.html)
 
-YouTube Video (to be updated with the new version):
-
 - [MedNIST Classification Example](https://www.youtube.com/watch?v=WwjilJFHuU4)
 
-### [3) Creating a Segmentation app](https://monai.readthedocs.io/projects/monai-deploy-app-sdk/en/stable/getting_started/tutorials/segmentation_app.html)
-
-YouTube Video (demonstrating the previous version of the App SDK):
-
-- [Spleen Organ Segmentation - Jupyter Notebook Tutorial](https://www.youtube.com/watch?v=cqDVxzYt9lY)
-- [Spleen Organ Segmentation - Deep Dive](https://www.youtube.com/watch?v=nivgfD4pwWE)
+#### [3) Creating a Segmentation app](https://monai.readthedocs.io/projects/monai-deploy-app-sdk/en/stable/getting_started/tutorials/segmentation_app.html)
 
-### [4) Creating a Segmentation app including visualization with Clara Viz](https://monai.readthedocs.io/projects/monai-deploy-app-sdk/en/stable/getting_started/tutorials/segmentation_clara-viz_app.html)
+#### [4) Creating a Segmentation app including visualization with Clara Viz](https://monai.readthedocs.io/projects/monai-deploy-app-sdk/en/stable/getting_started/tutorials/segmentation_clara-viz_app.html)
 
-### [5) Creating a Segmentation app consuming a MONAI Bundle](https://monai.readthedocs.io/projects/monai-deploy-app-sdk/en/stable/getting_started/tutorials/monai_bundle_app.html)
+#### [5) Creating a Segmentation app consuming a MONAI Bundle](https://monai.readthedocs.io/projects/monai-deploy-app-sdk/en/stable/getting_started/tutorials/monai_bundle_app.html)
 
 ### [Examples](https://monai.readthedocs.io/projects/monai-deploy-app-sdk/en/stable/getting_started/examples.html)
 

monai/deploy/core/models/model.py

@@ -16,7 +16,7 @@
 from monai.deploy.exceptions import ItemNotExistsError, UnknownTypeError
 
 # Store all supported model types in the order they should be checked
-REGISTERED_MODELS = []
+REGISTERED_MODELS: List[type["Model"]] = []
 
 
 class Model:

monai/deploy/graphs/__init__.py

@@ -1 +1,29 @@
-from holoscan.graphs import *
+import importlib
+
+try:
+    _flow_graphs = importlib.import_module("holoscan.flow_graphs")
+except ModuleNotFoundError as e:
+    if e.name != "holoscan.flow_graphs":
+        raise
+    _graphs = importlib.import_module("holoscan.graphs")
+    FlowGraph = _graphs.FlowGraph
+    FragmentFlowGraph = _graphs.FragmentFlowGraph
+    OperatorFlowGraph = _graphs.OperatorFlowGraph
+
+    __all__ = ["FlowGraph", "FragmentFlowGraph", "OperatorFlowGraph"]
+else:
+    FlowGraphImpl = _flow_graphs.FlowGraphImpl
+    FragmentFlowGraph = _flow_graphs.FragmentFlowGraph
+    FragmentFlowGraphImpl = _flow_graphs.FragmentFlowGraphImpl
+    OperatorFlowGraph = _flow_graphs.OperatorFlowGraph
+    OperatorFlowGraphImpl = _flow_graphs.OperatorFlowGraphImpl
+    FlowGraph = FlowGraphImpl
+
+    __all__ = [
+        "FlowGraph",
+        "FlowGraphImpl",
+        "FragmentFlowGraph",
+        "FragmentFlowGraphImpl",
+        "OperatorFlowGraph",
+        "OperatorFlowGraphImpl",
+    ]

monai/deploy/operators/monet_bundle_inference_operator.py

@@ -9,14 +9,16 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-from typing import Any, Dict, Tuple, Union
+from collections.abc import Hashable, Mapping
+from typing import Any, Dict, Tuple, Union, cast
 
 from monai.deploy.core import Image
-from monai.deploy.operators.monai_bundle_inference_operator import MonaiBundleInferenceOperator, get_bundle_config
-from monai.deploy.utils.importutil import optional_import
-from monai.transforms import ConcatItemsd, ResampleToMatch
 from monai.deploy.core.models.torch_model import TorchScriptModel
 from monai.deploy.core.models.triton_model import TritonModel
+from monai.deploy.operators.monai_bundle_inference_operator import MonaiBundleInferenceOperator
+from monai.deploy.utils.importutil import optional_import
+from monai.transforms import ConcatItemsd, ResampleToMatch
+
 torch, _ = optional_import("torch", "1.10.2")
 MetaTensor, _ = optional_import("monai.data.meta_tensor", name="MetaTensor")
 __all__ = ["MONetBundleInferenceOperator"]
@@ -88,7 +90,9 @@ def predict(self, data: Any, *args, **kwargs) -> Union[Image, Any, Tuple[Any, ..
             for key in kwargs.keys():
                 if isinstance(kwargs[key], MetaTensor):
                     multimodal_data[key] = ResampleToMatch(mode="bilinear")(kwargs[key], img_dst=data)
-            data = ConcatItemsd(keys=list(multimodal_data.keys()), name="image")(multimodal_data)["image"]
+            data = ConcatItemsd(keys=list(multimodal_data.keys()), name="image")(
+                cast(Mapping[Hashable, Any], multimodal_data)
+            )["image"]
         if len(data.shape) == 4:
             data = data[None]
         prediction = self._nnunet_predictor(data)

monai/deploy/utils/importutil.py

@@ -9,6 +9,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
+import importlib.util
 import inspect
 import re
 import runpy
@@ -411,7 +412,44 @@ def dist_requires(project_name: str) -> List[str]:
     return []
 
 
-holoscan_init_content_txt = """
+def _holoscan_package_path() -> Path:
+    """Return the installed holoscan package directory without importing holoscan."""
+    spec = importlib.util.find_spec("holoscan")
+    if spec is None:
+        raise ModuleNotFoundError("Holoscan is not installed; cannot locate the holoscan package directory.")
+    if spec.submodule_search_locations:
+        return Path(spec.submodule_search_locations[0]).resolve()
+    if spec.origin is None:
+        raise ModuleNotFoundError("Holoscan package spec has no origin or submodule_search_locations.")
+    return Path(spec.origin).resolve().parent
+
+
+def _holoscan_graph_module_name(holoscan_pkg_path: Path) -> Optional[str]:
+    """Return the graph submodule name present in the installed Holoscan package."""
+    if (holoscan_pkg_path / "flow_graphs").is_dir():
+        return "flow_graphs"
+    if (holoscan_pkg_path / "graphs").is_dir():
+        return "graphs"
+    return None
+
+
+def _build_holoscan_extra_modules(holoscan_pkg_path: Path) -> list[str]:
+    extra_modules = [
+        "conditions",
+        "executors",
+        "logger",
+        "operators",
+        "resources",
+    ]
+    graph_module = _holoscan_graph_module_name(holoscan_pkg_path)
+    if graph_module is not None:
+        extra_modules.insert(2, graph_module)
+    return extra_modules
+
+
+def _build_holoscan_init_content(extra_modules: list[str]) -> str:
+    extra_modules_repr = ",\n    ".join(f"{name!r}" for name in extra_modules)
+    return f"""\
 # SPDX-FileCopyrightText: Copyright (c) 2022-2023 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 # SPDX-License-Identifier: Apache-2.0
 #
@@ -435,12 +473,7 @@ def dist_requires(project_name: str) -> List[str]:
 
 # Other modules are exposed to the public API but will only be lazily loaded
 _EXTRA_MODULES = [
-    "conditions",
-    "executors",
-    "graphs",
-    "logger",
-    "operators",
-    "resources",
+    {extra_modules_repr},
 ]
 __all__.extend(_EXTRA_MODULES)
 
@@ -456,12 +489,12 @@ def __getattr__(name):
     import sys
 
     if name in _EXTRA_MODULES:
-        module_name = f"{__name__}.{name}"
+        module_name = f"{{__name__}}.{{name}}"
         module = importlib.import_module(module_name)  # import
         sys.modules[module_name] = module  # cache
         return module
     else:
-        raise AttributeError(f"module {__name__} has no attribute {name}")
+        raise AttributeError(f"module {{__name__}} has no attribute {{name}}")
 
 """
 
@@ -470,11 +503,12 @@ def fix_holoscan_import():
     """Fix holoscan __init__ to enable lazy load for avoiding failure on loading low level libs."""
 
     try:
-        project_name = "holoscan"
-        holoscan_init_path = Path(dist_module_path(project_name)) / project_name / "__init__.py"
+        holoscan_pkg_path = _holoscan_package_path()
+        holoscan_init_path = holoscan_pkg_path / "__init__.py"
+        extra_modules = _build_holoscan_extra_modules(holoscan_pkg_path)
 
         with open(str(holoscan_init_path), "w") as f_w:
-            f_w.write(holoscan_init_content_txt)
+            f_w.write(_build_holoscan_init_content(extra_modules))
         return str(holoscan_init_path)
     except Exception as ex:
         return ex