feat: Remove opencv python (#128)

* build: Update dependencies to retain only opencv-python-headless

* build: Upgrade version, update changelog

* test: Skip draem test as it doesn't work with the current anomalib dependency

* refactor: Update gradcam integration code to support the new version

CHANGELOG.md

@@ -2,6 +2,12 @@
 # Changelog
 All notable changes to this project will be documented in this file.
 
+### [2.2.3]
+
+#### Updated
+
+- Keep only opencv-python-headless as a dependency for quadra to avoid conflicts with the non-headless version
+
 ### [2.2.2]
 
 #### Updated

pyproject.toml

@@ -1,6 +1,6 @@
 [tool.poetry]
 name = "quadra"
-version = "2.2.2"
+version = "2.2.3"
 description = "Deep Learning experiment orchestration library"
 authors = [
 	"Federico Belotti <[email protected]>",
@@ -60,7 +60,7 @@ python_dotenv = "~0.21"
 rich = "~13.2"
 scikit_learn = "~1.2"
 pydantic = "1.10.10"
-grad_cam = "1.4.6"
+grad-cam-orobix = "1.5.3.dev001"
 matplotlib = "~3.6"
 seaborn = "~0.12"
 albumentations = "~1.3"
@@ -72,7 +72,7 @@ h5py = "~3.8"
 timm = "0.9.12"
 
 segmentation_models_pytorch-orobix = "0.3.3.dev1"
-anomalib-orobix = "0.7.0.dev140"
+anomalib-orobix = "0.7.0.dev141"
 xxhash = "~3.2"
 torchinfo = "~1.8"
 typing_extensions = { version = "4.11.0", python = "<3.10" }

quadra/__init__.py

@@ -1,4 +1,4 @@
-__version__ = "2.2.2"
+__version__ = "2.2.3"
 
 
 def get_version():

quadra/modules/classification/base.py

@@ -138,13 +138,9 @@ def prepare_gradcam(self) -> None:
         if isinstance(self.model.features_extractor, timm.models.resnet.ResNet):
             target_layers = [cast(BaseNetworkBuilder, self.model).features_extractor.layer4[-1]]
 
-            # Get model current device
-            device = next(self.model.parameters()).device
-
             self.cam = GradCAM(
                 model=self.model,
                 target_layers=target_layers,
-                use_cuda=device.type == "cuda",
             )
             # Activating gradients
             for p in self.model.features_extractor.layer4[-1].parameters():
@@ -262,7 +258,7 @@ def __init__(
                 self.gradcam = False
             else:
                 target_layers = [model[0].features_extractor.layer4[-1]]
-                self.cam = GradCAM(model=model, target_layers=target_layers, use_cuda=torch.cuda.is_available())
+                self.cam = GradCAM(model=model, target_layers=target_layers)
 
     def forward(self, x):
         return self.model(x)

quadra/tasks/classification.py

@@ -1138,7 +1138,6 @@ def prepare_gradcam(self) -> None:
             self.cam = GradCAM(
                 model=self.deployment_model.model,
                 target_layers=target_layers,
-                use_cuda=(self.device != "cpu"),
             )
             for p in self.deployment_model.model.features_extractor.layer4[-1].parameters():
                 p.requires_grad = True

quadra/utils/models.py

@@ -131,7 +131,6 @@ def get_feature(
             cam = GradCAM(
                 model=feature_extractor,
                 target_layers=target_layers,
-                use_cuda=torch.cuda.is_available(),
             )
             for p in feature_extractor.features_extractor.layer4[-1].parameters():
                 p.requires_grad = True

tests/tasks/test_anomaly.py

@@ -286,6 +286,10 @@ def test_fastflow(tmp_path: Path, base_anomaly_dataset: base_anomaly_dataset, ta
     shutil.rmtree(tmp_path)
 
 
+@pytest.mark.skip(
+    reason="This test requires anomalib with imgaug installed which we don't want to include in the dependencies as it "
+    "requires the non headless version of opencv-python"
+)
 @pytest.mark.slow
 @pytest.mark.usefixtures("mock_training")
 @pytest.mark.parametrize("task", ["classification", "segmentation"])