Update app.py

app.py

@@ -154,23 +154,51 @@ def _update_progress(progress, value, desc=""):
         progress(value / 100.0, desc=desc)
 
 # ---- PREDICT SINGLE ----
+from tensorflow.keras.preprocessing import image
+
 def predict_single(img_input, weights=(0.45, 0.25, 0.3), normalize=True):
-    # Chargement et pré-traitement (avec logs)
+    print("🔍 DEBUG GRADIO - Début de la prédiction")
+
+    # Chargement et pré-traitement avec Keras (comme en local)
     if isinstance(img_input, str):
-        pil_img = Image.open(img_input).convert("RGB")
+        # Chargement depuis un chemin
+        img_raw_x = image.load_img(img_input, target_size=(299, 299), color_mode="rgb")
+        img_raw_r = image.load_img(img_input, target_size=(224, 224), color_mode="rgb")
+        img_raw_d = image.load_img(img_input, target_size=(224, 224), color_mode="rgb")
     else:
-        pil_img = img_input.convert("RGB")
-
-    # Pré-traitement Xception
-    img_x = np.expand_dims(preprocess_xception(np.array(pil_img.resize((299, 299)))), axis=0)
+        # Cas d'upload via interface Gradio (img_input est déjà une image)
+        # Conversion en array puis rechargement avec Keras pour uniformiser
+        temp_path = "temp_debug_image.jpg"
+        img_input.save(temp_path)  # Sauvegarde temporaire
+        img_raw_x = image.load_img(temp_path, target_size=(299, 299), color_mode="rgb")
+        img_raw_r = image.load_img(temp_path, target_size=(224, 224), color_mode="rgb")
+        img_raw_d = image.load_img(temp_path, target_size=(224, 224), color_mode="rgb")
+        import os
+        os.remove(temp_path)  # Nettoyage
+
+    print(f"📸 Images loaded:")
+    print(f"   Xception (299x299): {np.array(img_raw_x).shape}")
+    print(f"   ResNet (224x224): {np.array(img_raw_r).shape}")
+    print(f"   DenseNet (224x224): {np.array(img_raw_d).shape}")
+
+    # Conversion en arrays
+    array_x = image.img_to_array(img_raw_x)
+    array_r = image.img_to_array(img_raw_r)
+    array_d = image.img_to_array(img_raw_d)
+
+    print(f"🔧 Arrays avant preprocessing:")
+    print(f"   X shape: {array_x.shape}, dtype: {array_x.dtype}, range: [{array_x.min()}, {array_x.max()}]")
+    print(f"   R shape: {array_r.shape}, dtype: {array_r.dtype}, range: [{array_r.min()}, {array_r.max()}]")
+    print(f"   D shape: {array_d.shape}, dtype: {array_d.dtype}, range: [{array_d.min()}, {array_d.max()}]")
+
+    # Pré-traitement
+    img_x = np.expand_dims(preprocess_xception(array_x), axis=0)
     print("Xception range (Gradio):", np.min(img_x), np.max(img_x))
 
-    # Pré-traitement ResNet
-    img_r = np.expand_dims(preprocess_resnet(np.array(pil_img.resize((224, 224)))), axis=0)
+    img_r = np.expand_dims(preprocess_resnet(array_r), axis=0)
     print("ResNet range (Gradio):", np.min(img_r), np.max(img_r))
 
-    # Pré-traitement DenseNet
-    img_d = np.expand_dims(preprocess_densenet(np.array(pil_img.resize((224, 224)))), axis=0)
+    img_d = np.expand_dims(preprocess_densenet(array_d), axis=0)
     print("DenseNet range (Gradio):", np.min(img_d), np.max(img_d))
 
     # Prédictions
@@ -180,11 +208,13 @@ def predict_single(img_input, weights=(0.45, 0.25, 0.3), normalize=True):
         print("\n--- Xception (Gradio) ---")
         for i, (class_name, prob) in enumerate(zip(CLASS_NAMES, preds['xception'])):
             print(f"{class_name}: {prob*100:.2f}%")
+
     if model_resnet50 is not None:
         preds['resnet50'] = model_resnet50.predict(img_r, verbose=0)[0]
         print("\n--- ResNet (Gradio) ---")
         for i, (class_name, prob) in enumerate(zip(CLASS_NAMES, preds['resnet50'])):
             print(f"{class_name}: {prob*100:.2f}%")
+
     if model_densenet is not None:
         preds['densenet201'] = model_densenet.predict(img_d, verbose=0)[0]
         print("\n--- DenseNet (Gradio) ---")
@@ -220,7 +250,6 @@ def predict_single(img_input, weights=(0.45, 0.25, 0.3), normalize=True):
         for i, (class_name, prob) in enumerate(zip(CLASS_NAMES, ensemble)):
             print(f"{class_name}: {prob*100:.2f}%")
         print("Ensemble sum final (Gradio):", np.sum(ensemble))
-
         # Afficher l'impact de la normalisation
         print("\n--- Impact de la normalisation ---")
         for i, (class_name, prob_before, prob_after) in enumerate(zip(CLASS_NAMES, ensemble_before_norm, ensemble)):
@@ -231,6 +260,7 @@ def predict_single(img_input, weights=(0.45, 0.25, 0.3), normalize=True):
 
 
 
+
 # ---- Helpers Grad-CAM ----
 LAST_CONV_LAYERS = {
     "xception": "block14_sepconv2_act",