Update app.py

app.py

@@ -2,6 +2,7 @@ import gradio as gr
 import numpy as np
 from tensorflow.keras.models import load_model
 from tensorflow.keras.preprocessing.image import img_to_array
+from tensorflow.keras.applications.xception import preprocess_input
 from PIL import Image
 from huggingface_hub import hf_hub_download
 
@@ -11,20 +12,16 @@ model = load_model(model_path)
 
 # Inference function
 def predict(image):
-    # === Short-Term Workaround for Inference ===
-    image = image.convert("RGB")                      # Ensure 3 color channels
-    image = image.resize((299, 299))                  # Resize to match model input
-    image = img_to_array(image).astype("float32")     # Convert to array + ensure float32
-    image = np.expand_dims(image, axis=0)             # Add batch dimension
-    image /= 255.0                                     # Normalize to [0, 1] like during training
-
-    # Model prediction
-    prob = model.predict(image)[0][0]
+    image = image.resize((299, 299))  # Resize to match model input
+    image = img_to_array(image)       # Convert to numpy array
+    image = np.expand_dims(image, axis=0)  # Add batch dimension
+    image = preprocess_input(image)   # Apply Xception preprocessing (important fix!)
 
-    # 0 = Fake, 1 = Real
-    label = "Fake" if prob < 0.3 else "Real"
-    confidence = round(float(1 - prob if label == "Fake" else prob), 3)
+    prob = model.predict(image)[0][0]
 
+    # Based on training: label 0 = Fake, label 1 = Real
+    label = "Real" if prob > 0.5 else "Fake"
+    confidence = round(float(prob if prob > 0.5 else 1 - prob), 3)
 
     return f"{label} ({confidence * 100:.1f}%)"
 
@@ -38,4 +35,3 @@ iface = gr.Interface(
 
 iface.launch()
 
-