Support SAM2 for python >=3.10

Signed-off-by: Sachidanand Alle <salle@nvidia.com>

monailabel/sam2/utils.py

@@ -0,0 +1,9 @@
+from monai.utils import optional_import
+
+
+def is_sam2_module_available():
+    try:
+        _, flag = optional_import("sam2")
+        return flag
+    except ImportError:
+        return False

plugins/cvat/README.md

@@ -57,8 +57,9 @@ chmod +x nuctl-$NUCLIO_VERSION-linux-amd64
 ln -sf $(pwd)/nuctl-$NUCLIO_VERSION-linux-amd64 /usr/local/bin/nuctl
 ```
 
-#### Deployment of Endoscopy Models
+#### Deployment of Endoscopy/SAM2 Models
 This step is to deploy MONAI Label plugin with endoscopic models using Nuclio tool.
+> **Prerequisite:** MONAI Label Server is up and running for _**endoscopy**_ app.
 
 ```bash
 # Run MONAI Label Server (Make sure this Host/IP is accessible inside a docker)
@@ -68,8 +69,13 @@ git clone https://github.com/Project-MONAI/MONAILabel.git
 
 # Deploy all endoscopy models
 ./plugins/cvat/deploy.sh endoscopy
+
 # Or to deploy specific function and model, e.g., tooltracking
 ./plugins/cvat/deploy.sh endoscopy tooltracking
+
+# Deploy SAM2 Interactor
+./plugins/cvat/deploy.sh sam2 interactor
+
 ```
 
 After model deployment, you can see the model names in the `Models` page of CVAT.

plugins/cvat/pathology/deepedit_nuclei.yaml

@@ -34,7 +34,7 @@ spec:
     directives:
       preCopy:
         - kind: ENV
-          value: MONAI_LABEL_SERVER=http://10.117.9.63:8000
+          value: MONAI_LABEL_SERVER=http://monailabel.com
         - kind: ENV
           value: MONAI_LABEL_MODEL=deepedit_nuclei
 

plugins/cvat/pathology/nuclick.yaml

@@ -36,7 +36,7 @@ spec:
     directives:
       preCopy:
         - kind: ENV
-          value: MONAI_LABEL_SERVER=http://10.117.9.63:8000
+          value: MONAI_LABEL_SERVER=http://monailabel.com
         - kind: ENV
           value: MONAI_LABEL_MODEL=nuclick
 

plugins/cvat/pathology/segmentation_nuclei.yaml

@@ -38,7 +38,7 @@ spec:
     directives:
       preCopy:
         - kind: ENV
-          value: MONAI_LABEL_SERVER=http://10.117.9.63:8000
+          value: MONAI_LABEL_SERVER=http://monailabel.com
         - kind: ENV
           value: MONAI_LABEL_MODEL=segmentation_nuclei
 

plugins/cvat/sam2/interactor.yaml

@@ -34,9 +34,9 @@ spec:
     directives:
       preCopy:
         - kind: ENV
-          value: MONAI_LABEL_SERVER=http://10.117.9.63:8000
+          value: MONAI_LABEL_SERVER=http://monailabel.com
         - kind: ENV
-          value: MONAI_LABEL_MODEL=sam2
+          value: MONAI_LABEL_MODEL=sam_2d
 
   triggers:
     myHttpTrigger:

plugins/cvat/sam2/tracker.yaml

@@ -30,9 +30,9 @@ spec:
     directives:
       preCopy:
         - kind: ENV
-          value: MONAI_LABEL_SERVER=http://10.117.9.63:8000
+          value: MONAI_LABEL_SERVER=http://monailabel.com
         - kind: ENV
-          value: MONAI_LABEL_MODEL=sam2
+          value: MONAI_LABEL_MODEL=sam_2d
 
   triggers:
     myHttpTrigger:

requirements.txt

@@ -44,7 +44,7 @@ urllib3==2.2.2
 scikit-learn
 scipy
 google-auth==2.29.0
-git+https://github.com/facebookresearch/sam2.git@c2ec8e14a185632b0a5d8b161928ceb50197eddc
+git+https://github.com/facebookresearch/sam2.git@c2ec8e14a185632b0a5d8b161928ceb50197eddc ; python_version >= '3.10'
 
 
 # scipy and scikit-learn latest packages are missing on python 3.8

sample-apps/README.md

@@ -54,6 +54,14 @@ The endoscopy template includes example models for interactive and automated too
 #### [MONAI Bundle](./monaibundle)
 The MONAI Bundle format provides a portable description of deep learning models. This template includes example models for interactive and automated segmentation using MONAI bundles defined in the MONAI Model Zoo. It can pull any bundle defined in the MONAI Model Zoo that is compatible and meets the requirements specified on the [MONAI Bundle Apps page](./monaibundle/).
 
+----
+> [**SAM2**](https://github.com/facebookresearch/sam2/)
+>
+> By default, SAM2 is available **_(python >= 3.10)_** for all the above Apps
+>  - **sam_2d**: for any organ or tissue and others.
+>  - **sam_3d**: in case of Radiology/MonaiBundle apps to support SAM2 propagation over multiple slices.
+
+
 ### Creating a Custom App
 Researchers may want to define and add their own models. Follow the steps below to add a new segmentation model:
 

sample-apps/endoscopy/main.py

@@ -30,7 +30,7 @@
 from monailabel.interfaces.tasks.scoring import ScoringMethod
 from monailabel.interfaces.tasks.strategy import Strategy
 from monailabel.interfaces.tasks.train import TrainTask
-from monailabel.sam2.infer import Sam2InferTask
+from monailabel.sam2.utils import is_sam2_module_available
 from monailabel.tasks.activelearning.random import Random
 from monailabel.utils.others.class_utils import get_class_names
 from monailabel.utils.others.generic import create_dataset_from_path, strtobool
@@ -131,8 +131,10 @@ def init_infers(self) -> Dict[str, InferTask]:
         #################################################
         # SAM
         #################################################
-        if self.sam:
-            infers["sam2"] = Sam2InferTask(
+        if is_sam2_module_available() and self.sam:
+            from monailabel.sam2.infer import Sam2InferTask
+
+            infers["sam_2d"] = Sam2InferTask(
                 model_dir=self.model_dir,
                 type=InferType.ANNOTATION,
                 dimension=2,

sample-apps/monaibundle/main.py

@@ -21,7 +21,7 @@
 from monailabel.interfaces.tasks.scoring import ScoringMethod
 from monailabel.interfaces.tasks.strategy import Strategy
 from monailabel.interfaces.tasks.train import TrainTask
-from monailabel.sam2.infer import Sam2InferTask
+from monailabel.sam2.utils import is_sam2_module_available
 from monailabel.tasks.activelearning.first import First
 from monailabel.tasks.activelearning.random import Random
 from monailabel.tasks.infer.bundle import BundleInferTask
@@ -80,7 +80,9 @@ def init_infers(self) -> Dict[str, InferTask]:
         #################################################
         # SAM
         #################################################
-        if self.sam:
+        if is_sam2_module_available() and self.sam:
+            from monailabel.sam2.infer import Sam2InferTask
+
             infers["sam_2d"] = Sam2InferTask(model_dir=self.model_dir, dimension=2)
             infers["sam_3d"] = Sam2InferTask(model_dir=self.model_dir, dimension=3)
         return infers

sample-apps/pathology/main.py

@@ -28,7 +28,7 @@
 from monailabel.interfaces.tasks.infer_v2 import InferTask, InferType
 from monailabel.interfaces.tasks.strategy import Strategy
 from monailabel.interfaces.tasks.train import TrainTask
-from monailabel.sam2.infer import Sam2InferTask
+from monailabel.sam2.utils import is_sam2_module_available
 from monailabel.tasks.infer.basic_infer import BasicInferTask
 from monailabel.transform.post import FindContoursd
 from monailabel.transform.writer import PolygonWriter
@@ -146,8 +146,10 @@ def init_infers(self) -> Dict[str, InferTask]:
         #################################################
         # SAM
         #################################################
-        if self.sam:
-            infers["sam2"] = Sam2InferTask(
+        if is_sam2_module_available() and self.sam:
+            from monailabel.sam2.infer import Sam2InferTask
+
+            infers["sam_2d"] = Sam2InferTask(
                 model_dir=self.model_dir,
                 type=InferType.ANNOTATION,
                 dimension=2,

sample-apps/radiology/main.py

@@ -27,7 +27,7 @@
 from monailabel.interfaces.tasks.scoring import ScoringMethod
 from monailabel.interfaces.tasks.strategy import Strategy
 from monailabel.interfaces.tasks.train import TrainTask
-from monailabel.sam2.infer import Sam2InferTask
+from monailabel.sam2.utils import is_sam2_module_available
 from monailabel.tasks.activelearning.first import First
 from monailabel.tasks.activelearning.random import Random
 
@@ -168,7 +168,9 @@ def init_infers(self) -> Dict[str, InferTask]:
         #################################################
         # SAM
         #################################################
-        if self.sam:
+        if is_sam2_module_available() and self.sam:
+            from monailabel.sam2.infer import Sam2InferTask
+
             infers["sam_2d"] = Sam2InferTask(model_dir=self.model_dir, dimension=2)
             infers["sam_3d"] = Sam2InferTask(model_dir=self.model_dir, dimension=3)
 

setup.cfg

@@ -70,7 +70,7 @@ install_requires =
     scikit-learn
     scipy
     google-auth>=2.29.0
-    git+https://github.com/facebookresearch/sam2.git@c2ec8e14a185632b0a5d8b161928ceb50197eddc
+    git+https://github.com/facebookresearch/sam2.git@c2ec8e14a185632b0a5d8b161928ceb50197eddc ; python_version >= '3.10'
 
 [flake8]
 select = B,C,E,F,N,P,T4,W,B9

tests/unit/endpoints/test_infer_v2.py

@@ -15,6 +15,8 @@
 
 import torch
 
+from monailabel.sam2.utils import is_sam2_module_available
+
 from .context import BasicBundleTestSuite, BasicDetectionBundleTestSuite, BasicEndpointV2TestSuite
 
 
@@ -43,7 +45,7 @@ def test_deepgrow_pipeline(self):
         time.sleep(1)
 
     def test_sam_2d(self):
-        if not torch.cuda.is_available():
+        if not is_sam2_module_available() or not torch.cuda.is_available():
             return
 
         model = "sam_2d"
@@ -55,7 +57,7 @@ def test_sam_2d(self):
         time.sleep(1)
 
     def test_sam_3d(self):
-        if not torch.cuda.is_available():
+        if not is_sam2_module_available or not torch.cuda.is_available():
             return
 
         model = "sam_3d"