Upload app.py with huggingface_hub

app.py

@@ -0,0 +1,52 @@
+
+import gradio as gr
+import tensorflow as tf
+import numpy as np
+from datasets import load_dataset
+from PIL import Image
+
+# Define the preprocessing function
+def preprocess_inference_image(image):
+    # Ensure image is a tensor
+    image = tf.convert_to_tensor(image, dtype=tf.float32)
+    # Convert image to RGB if it has 4 channels (RGBA)
+    if tf.shape(image)[-1] == 4:
+        image = image[..., :3]  # Select first 3 channels (RGB)
+    # Normalize to [0,1]
+    image = tf.image.convert_image_dtype(image, tf.float32)
+    # Resize to MobileNetV3 input size
+    image = tf.image.resize(image, [224, 224])
+    # Add batch dimension
+    image = tf.expand_dims(image, axis=0)
+    return image
+
+# Load the trained model
+model = tf.keras.models.load_model('maize_disease_model.keras')
+
+# Load the dataset to get label names
+ds = load_dataset("aquib1011/maize-leaf-disease", cache_dir=None)
+label_names = ds['train'].features['label'].names
+
+def predict_maize_disease(image):
+    # Convert PIL Image to numpy array
+    image = np.array(image)
+    # Apply preprocessing to the input image
+    processed_image = preprocess_inference_image(image)
+
+    # Make a prediction
+    predictions = model.predict(processed_image)
+
+    # Return the results as a dictionary for Gradio's Label component
+    return {label_names[i]: float(predictions[0][i]) for i in range(len(label_names))}
+
+# Create the Gradio interface
+iface = gr.Interface(
+    fn=predict_maize_disease,
+    inputs=gr.Image(type="pil"),
+    outputs=gr.Label(),
+    title="Maize Leaf Disease Classifier",
+    description="Upload an image of a maize leaf to get a prediction of the disease."
+)
+
+if __name__ == "__main__":
+    iface.launch()