[Anomaly Task] 🐞 Fix inference when model backbone changes (#1242)

* Add check for change in model backbone

* Add load model to train

* Limit padim to only resnet18

* Fix comment

external/anomaly/configs/base/padim/configuration.py

@@ -38,10 +38,14 @@ class LearningParameters(BaseAnomalyConfig.LearningParameters):
         header = string_attribute("Learning Parameters")
         description = header
 
+        # Editable is set to false as WideResNet50 is very large for
+        # onnx's protobuf (2gb) limit. This ends up crashing the export.
         backbone = selectable(
             default_value=ModelBackbone.RESNET18,
             header="Model Backbone",
             description="Pre-trained backbone used for feature extraction",
+            editable=False,
+            visible_in_ui=False,
         )
 
     learning_parameters = add_parameter_group(LearningParameters)

external/anomaly/configs/classification/padim/configuration.yaml

@@ -35,7 +35,7 @@ learning_parameters:
     auto_hpo_value: null
     default_value: resnet18
     description: Pre-trained backbone used for feature extraction
-    editable: true
+    editable: false
     enum_name: ModelBackbone
     header: Model Backbone
     options:
@@ -48,7 +48,7 @@ learning_parameters:
       rules: []
       type: UI_RULES
     value: resnet18
-    visible_in_ui: true
+    visible_in_ui: false
     warning: null
   description: Learning Parameters
   header: Learning Parameters

external/anomaly/configs/detection/padim/configuration.yaml

@@ -35,7 +35,7 @@ learning_parameters:
     auto_hpo_value: null
     default_value: resnet18
     description: Pre-trained backbone used for feature extraction
-    editable: true
+    editable: false
     enum_name: ModelBackbone
     header: Model Backbone
     options:
@@ -48,7 +48,7 @@ learning_parameters:
       rules: []
       type: UI_RULES
     value: resnet18
-    visible_in_ui: true
+    visible_in_ui: false
     warning: null
   description: Learning Parameters
   header: Learning Parameters

external/anomaly/configs/segmentation/padim/configuration.yaml

@@ -35,7 +35,7 @@ learning_parameters:
     auto_hpo_value: null
     default_value: resnet18
     description: Pre-trained backbone used for feature extraction
-    editable: true
+    editable: false
     enum_name: ModelBackbone
     header: Model Backbone
     options:
@@ -48,7 +48,7 @@ learning_parameters:
       rules: []
       type: UI_RULES
     value: resnet18
-    visible_in_ui: true
+    visible_in_ui: false
     warning: null
   description: Learning Parameters
   header: Learning Parameters

external/anomaly/tasks/inference.py

@@ -120,8 +120,8 @@ def load_model(self, ote_model: Optional[ModelEntity]) -> AnomalyModule:
             AnomalyModule: Anomalib
                 classification or segmentation model with/without weights.
         """
-        model = get_model(config=self.config)
         if ote_model is None:
+            model = get_model(config=self.config)
             logger.info(
                 "No trained model in project yet. Created new model with '%s'",
                 self.model_name,
@@ -130,10 +130,16 @@ def load_model(self, ote_model: Optional[ModelEntity]) -> AnomalyModule:
             buffer = io.BytesIO(ote_model.get_data("weights.pth"))
             model_data = torch.load(buffer, map_location=torch.device("cpu"))
 
+            if model_data["config"]["model"]["backbone"] != self.config["model"]["backbone"]:
+                logger.warning(
+                    "Backbone of the model in the Task Environment is different from the one in the template. "
+                    f"creating model with backbone={model_data['config']['model']['backbone']}"
+                )
+                self.config["model"]["backbone"] = model_data["config"]["model"]["backbone"]
             try:
+                model = get_model(config=self.config)
                 model.load_state_dict(model_data["model"])
                 logger.info("Loaded model weights from Task Environment")
-
             except BaseException as exception:
                 raise ValueError("Could not load the saved model. The model file structure is invalid.") from exception
 

external/anomaly/tasks/train.py

@@ -14,11 +14,14 @@
 # See the License for the specific language governing permissions
 # and limitations under the License.
 
+import io
 from typing import Optional
 
+import torch
 from adapters.anomalib.callbacks import ProgressCallback
 from adapters.anomalib.data import OTEAnomalyDataModule
 from adapters.anomalib.logger import get_logger
+from anomalib.models import AnomalyModule, get_model
 from anomalib.utils.callbacks import (
     MetricsConfigurationCallback,
     MinMaxNormalizationCallback,
@@ -83,3 +86,42 @@ def train(
         self.save_model(output_model)
 
         logger.info("Training completed.")
+
+    def load_model(self, ote_model: Optional[ModelEntity]) -> AnomalyModule:
+        """Create and Load Anomalib Module from OTE Model.
+
+        This method checks if the task environment has a saved OTE Model,
+        and creates one. If the OTE model already exists, it returns the
+        the model with the saved weights.
+
+        Args:
+            ote_model (Optional[ModelEntity]): OTE Model from the
+                task environment.
+
+        Returns:
+            AnomalyModule: Anomalib
+                classification or segmentation model with/without weights.
+        """
+        model = get_model(config=self.config)
+        if ote_model is None:
+            logger.info(
+                "No trained model in project yet. Created new model with '%s'",
+                self.model_name,
+            )
+        else:
+            buffer = io.BytesIO(ote_model.get_data("weights.pth"))
+            model_data = torch.load(buffer, map_location=torch.device("cpu"))
+
+            try:
+                if model_data["config"]["model"]["backbone"] == self.config["model"]["backbone"]:
+                    model.load_state_dict(model_data["model"])
+                    logger.info("Loaded model weights from Task Environment")
+                else:
+                    logger.info(
+                        "Model backbone does not match. Created new model with '%s'",
+                        self.model_name,
+                    )
+            except BaseException as exception:
+                raise ValueError("Could not load the saved model. The model file structure is invalid.") from exception
+
+        return model