Create app.py

app.py

@@ -0,0 +1,98 @@
+import gradio as gr
+import tensorflow as tf
+import numpy as np
+from PIL import Image
+import os
+import io
+
+# --- Configuration ---
+PHOTO_SIZE = 224
+MODEL_FILE_NAME = 'vgg_model50.h5' # Make sure this matches your uploaded model file
+CLASS_NAMES = ["Non-Autstic", "Autstic"] # Match indices 0 and 1 from your notebook
+
+# --- Load the Keras model ---
+# Ensure the model file is in the same directory in the Space repository
+model_path = os.path.join(os.path.dirname(__file__), MODEL_FILE_NAME)
+if not os.path.exists(model_path):
+    # Basic error handling if model is missing
+    raise FileNotFoundError(f"Model file '{MODEL_FILE_NAME}' not found. Please upload it to the Space.")
+
+try:
+    model = tf.keras.models.load_model(model_path)
+    print("Model loaded successfully.")
+except Exception as e:
+    print(f"Error loading model: {e}")
+    # You might want more robust error handling or display in the Gradio interface
+    model = None
+
+def preprocess_image(pil_image):
+    """
+    Preprocesses the PIL image object for the VGG16 model.
+    Resizes to (PHOTO_SIZE, PHOTO_SIZE), normalizes to [0, 1].
+    """
+    try:
+        # Gradio provides a PIL image object directly
+        img = pil_image.convert('RGB') # Ensure image is RGB
+        img = img.resize((PHOTO_SIZE, PHOTO_SIZE))
+        np_image = np.array(img).astype('float32') / 255.0
+        # Expand dimensions to create batch size of 1
+        np_image = np.expand_dims(np_image, axis=0)
+        print(f"Image preprocessed successfully. Shape: {np_image.shape}")
+        return np_image
+    except Exception as e:
+        print(f"Error preprocessing image: {e}")
+        return None
+
+def predict_autism(image_input):
+    """
+    Takes a PIL Image input from Gradio, preprocesses, predicts, and returns the class name.
+    """
+    if model is None:
+        return "Error: Model not loaded." # Or raise an error
+
+    print(f"Received image of type: {type(image_input)}") # Should be PIL Image
+
+    # Preprocess the image (Gradio image input provides PIL image)
+    processed_image = preprocess_image(image_input)
+    if processed_image is None:
+        return "Error: Image preprocessing failed."
+
+    # Make prediction
+    print("Making prediction...")
+    prediction = model.predict(processed_image)
+    predicted_class_index = np.argmax(prediction, axis=1)[0]
+    predicted_class_name = CLASS_NAMES[predicted_class_index]
+    confidence = float(np.max(prediction)) # Get the confidence score
+
+    print(f"Prediction result index: {predicted_class_index}, Class: {predicted_class_name}, Confidence: {confidence:.4f}")
+
+    # Return prediction as a dictionary (Gradio handles JSON conversion for API)
+    # The key 'label' often works well with Gradio output components
+    # Or return just the string if using a simple Textbox output
+    # return {CLASS_NAMES[0]: float(1-confidence), CLASS_NAMES[1]: confidence} # Example for Label output
+    return predicted_class_name # Simpler for Textbox output
+
+# --- Create Gradio Interface ---
+# Input: Image Upload
+# Output: Textbox to display the predicted class
+# Allow flagging for feedback (optional but good practice)
+# Add title and description
+# Provide example images if available in your Space repo (e.g., in an 'examples' folder)
+# examples_folder = os.path.join(os.path.dirname(__file__), "examples")
+# example_images = [os.path.join(examples_folder, img) for img in os.listdir(examples_folder)] if os.path.exists(examples_folder) else None
+
+iface = gr.Interface(
+    fn=predict_autism,
+    inputs=gr.Image(type="pil", label="Upload Image"), # Input is PIL format
+    outputs=gr.Textbox(label="Prediction Result"), # Output is simple text
+    # outputs=gr.Label(num_top_classes=2), # Alternative: Label output shows confidences
+    title="Autism Classification from Facial Images (VGG16)",
+    description="Upload a facial image to classify as Autistic or Non-Autistic using a VGG16 model.",
+    allow_flagging="never",
+    # examples=example_images # Uncomment if you add example images
+)
+
+# --- Launch the Gradio app ---
+# When run on Hugging Face Spaces, it automatically uses the Space's URL
+if __name__ == "__main__":
+    iface.launch() # share=True is not needed on Spaces