add anomaly threshold as parameter, related to #20

src/dopt_sensor_anomalies/_interface.py

@@ -30,11 +30,13 @@ def sensor_anomalies_detection(
     user_img_path: str,
     pixels_per_metric_X: float,
     pixels_per_metric_Y: float,
+    anomaly_threshold: float,
 ) -> int:
     res = detection.pipeline(
         user_img_path=user_img_path,
         pixels_per_metric_X=pixels_per_metric_X,
         pixels_per_metric_Y=pixels_per_metric_Y,
+        anomaly_threshold=anomaly_threshold,
     )
     if res.status.code != 0:
         _print_error_state(res.status, out_stream=sys.stderr)

src/dopt_sensor_anomalies/constants.py

@@ -9,7 +9,7 @@ THRESHOLD_BW: Final[int] = 63
 BACKBONE: Final[str] = "wide_resnet50_2"
 LAYERS: Final[tuple[str, ...]] = ("layer1", "layer2", "layer3")
 RATIO: Final[float] = 0.01
-ANOMALY_THRESHOLD: Final[float] = 0.14
+ANOMALY_THRESHOLD_DEFAULT: Final[float] = 0.14
 
 NUM_VALID_ELECTRODES: Final[int] = 6
 HEATMAP_FILENAME_SUFFIX: Final[str] = "_Heatmap"

src/dopt_sensor_anomalies/detection.py

@@ -182,6 +182,7 @@ def measure_length(
 def infer_image(
     image: npt.NDArray[np.uint8],
     model: Patchcore,
+    anomaly_threshold: float,
 ) -> t.InferenceResult:
     torch_device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
     model.to(torch_device)
@@ -200,7 +201,7 @@ def infer_image(
         output = model(input_tensor)
 
     anomaly_score = output.pred_score.item()
-    anomaly_label = bool(1 if anomaly_score >= const.ANOMALY_THRESHOLD else 0)
+    anomaly_label = bool(1 if anomaly_score >= anomaly_threshold else 0)
     anomaly_map = output.anomaly_map.squeeze().cpu().numpy()
 
     img_np = np.array(pil_image)
@@ -219,6 +220,7 @@ def anomaly_detection(
     detection_models: t.DetectionModels,
     data_csv: t.CsvData,
     sensor_images: t.SensorImages,
+    anomaly_threshold: float,
 ) -> None:
     file_stem = img_path.stem
     folder_path = img_path.parent
@@ -233,7 +235,7 @@ def anomaly_detection(
         checkpoint = torch.load(detection_models[side])
         model.load_state_dict(checkpoint["model_state_dict"])
 
-        result = infer_image(image, model)
+        result = infer_image(image, model, anomaly_threshold)
         data_csv.extend([int(result.anomaly_label)])
 
         ax = axes[i]
@@ -265,6 +267,7 @@ def pipeline(
     user_img_path: str,
     pixels_per_metric_X: float,
     pixels_per_metric_Y: float,
+    anomaly_threshold: float,
 ) -> None:
     file_path = Path(user_img_path)
     if not file_path.exists():
@@ -283,4 +286,5 @@ def pipeline(
         detection_models=DETECTION_MODELS,
         data_csv=data_csv,
         sensor_images=sensor_images,
+        anomaly_threshold=anomaly_threshold,
     )

src/dopt_sensor_anomalies/detection.pyi

@@ -86,6 +86,7 @@ def measure_length(
 def infer_image(
     image: npt.NDArray[np.uint8],
     model: Patchcore,
+    anomaly_threshold: float,
 ) -> t.InferenceResult:
     """evaluate one image
 
@@ -113,6 +114,7 @@ def anomaly_detection(
     detection_models: t.DetectionModels,
     data_csv: t.CsvData,
     sensor_images: t.SensorImages,
+    anomaly_threshold: float,
 ) -> None:
     """load the model, call function for anomaly detection and store the results
 
@@ -134,6 +136,7 @@ def pipeline(
     user_img_path: str,
     pixels_per_metric_X: float,
     pixels_per_metric_Y: float,
+    anomaly_threshold: float,
 ) -> None:
     """full pipeline defined by the agreed requirements
     wrapped as result pattern, handle errors on higher abstraction level