Update app.py

app.py

@@ -9,6 +9,7 @@ import gradio as gr
 import io
 import base64
 import tensorflow as tf
+import keras  # <--- nuevo import para TFSMLayer
 import zipfile
 import os
 
@@ -28,7 +29,9 @@ extract_dir = "saved_model"
 if not os.path.exists(extract_dir):
     with zipfile.ZipFile(zip_path, 'r') as zip_ref:
         zip_ref.extractall(extract_dir)
-model_isic = tf.keras.models.load_model(extract_dir)
+
+# Cargar modelo con TFSMLayer (solo para inferencia)
+model_isic = keras.layers.TFSMLayer(extract_dir, call_endpoint='serving_default')
 
 # 🔹 Clases y niveles de riesgo
 CLASSES = [
@@ -53,7 +56,7 @@ def preprocess_image_isic(image: Image.Image):
     if img_array.shape[-1] == 4:  # eliminar canal alpha si existe
         img_array = img_array[..., :3]
     img_array = np.expand_dims(img_array, axis=0)  # batch dimension
-    return img_array
+    return img_array.astype(np.float32)
 
 def analizar_lesion_combined(img):
     # Convertir imagen para Fastai
@@ -73,9 +76,10 @@ def analizar_lesion_combined(img):
     prob_malignant = float(probs_fast_mal[1])  # índice 1 = maligno
     pred_fast_type, _, probs_fast_type = model_norm2000.predict(img_fastai)
 
-    # Modelo TensorFlow ISIC
+    # Modelo TensorFlow ISIC (usando TFSMLayer)
     x_isic = preprocess_image_isic(img)
-    preds_isic = model_isic.predict(x_isic)[0]  # vector probabilidades
+    preds_isic_tensor = model_isic(x_isic)  # devuelve tensor
+    preds_isic = preds_isic_tensor.numpy()[0]  # convertir a numpy y quitar batch dim
     pred_idx_isic = int(np.argmax(preds_isic))
     pred_class_isic = CLASSES[pred_idx_isic]
     confidence_isic = preds_isic[pred_idx_isic]
@@ -139,4 +143,3 @@ demo = gr.Interface(
 
 if __name__ == "__main__":
     demo.launch()
-