introduction gradcam++ pour efficientnetv2m

app/model.py

@@ -8,7 +8,7 @@ from PIL import Image
 from keras.applications.efficientnet_v2 import preprocess_input as effnet_preprocess
 from keras.applications.resnet_v2 import preprocess_input as resnet_preprocess
 import io
-from tf_keras_vis.gradcam import Gradcam
+from tf_keras_vis.gradcam import Gradcam,GradcamPlusPlus
 from tf_keras_vis.utils import normalize
 
 import numpy as np
@@ -38,6 +38,7 @@ class ModelStruct(TypedDict):
     preprocess_input: Callable[[np.ndarray], Any]
     target_size: tuple[int, int]
     last_conv_layer:str
+    gradcam_type:str
 
 
 def load_models() -> list[ModelStruct]:
@@ -55,7 +56,8 @@ def load_models() -> list[ModelStruct]:
         "gradcam_model":model1_for_gradcam,
         "preprocess_input": effnet_preprocess,
         "target_size": (480, 480),
-        "last_conv_layer":"top_activation"
+        "last_conv_layer":"top_activation",
+        "gradcam_type":"gradcam++"
 
     },
     {
@@ -64,7 +66,10 @@ def load_models() -> list[ModelStruct]:
         "model":model2,
         "preprocess_input":resnet_preprocess,
         "target_size":(224, 224),
-        "last_conv_layer":"conv5_block3_out"
+        "last_conv_layer":"conv5_block3_out",
+        "gradcam_type":"gradcam"
+        #"gradcam_type":"gradcam++"
+
 
     }
    
@@ -127,7 +132,7 @@ def compute_saliency_map(model, image_array, class_index=None):
     return saliency_map
 
 
-def compute_gradcam(model, image_array, class_index=None, layer_name=None):
+def compute_gradcam(model, image_array, class_index=None, layer_name=None,gradcam_type="gradcam"):
     """
     Calcule la carte Grad-CAM pour une image et un modèle Keras.
 
@@ -140,13 +145,16 @@ def compute_gradcam(model, image_array, class_index=None, layer_name=None):
     Returns:
         gradcam_map: np.array (H, W), normalisée entre 0 et 1.
     """
+    logging.info(f"Lancement calcul de la gradcam avec le type {gradcam_type}")
 
     if image_array.ndim == 3:
         input_tensor = np.expand_dims(image_array, axis=0)
     else:
         input_tensor = image_array
-
-    gradcam = Gradcam(model, clone=False)
+    if gradcam_type=="gradcam++":
+        gradcam = GradcamPlusPlus(model, clone=False)
+    else: 
+        gradcam = Gradcam(model, clone=False)
 
     def loss(output):
         if class_index is None:
@@ -363,7 +371,7 @@ def predict_with_model(config, image_bytes: bytes,show_heatmap=False):
             logger.info(f"🛠️ Dernière couche utilisée: {config['last_conv_layer']}")
 
             # Calcul de la heatmap
-            heatmap = compute_gradcam(config["gradcam_model"], raw_input, class_index=predicted_class_index, layer_name=config["last_conv_layer"])
+            heatmap = compute_gradcam(config["gradcam_model"], raw_input, class_index=predicted_class_index, layer_name=config["last_conv_layer"],gradcam_type=config["gradcam_type"])
 
             elapsed_time = time.time() - start_time
             logger.info(f"✅ Heatmap générée en {elapsed_time:.2f} secondes")