Update app.py

app.py

@@ -1,70 +1,22 @@
 from tensorflow.keras.models import load_model
-
-# importing the preprocessing steps for the model architecture which i used for transfer learning
-from tensorflow.keras.applications.xception import preprocess_input
-
+import gradio as gr
+from PIL import Image
 import numpy as np
-from tensorflow.keras.preprocessing.image import load_img, img_to_array
-import streamlit as st
-import cv2
-
-
-# import tensorflow as tf
-# print(tf.__version__)
-# print(np.__version__)
-# print(st.__version__)
-# print(cv2.__version__)
-
-
-st.write('# Cat and Dog Classifier')
-st.markdown(
-    '''
-        This app uses transfer learning on  the Xception model to predict images of cats and dogs.
-        It achieved an accuracy of approx. 99 percent on the validation set.
-        
-        *View on [Github](https://github.com/eskayML/cat-and-dogs-classification)*
-        
-        > ### Enter an image of either a cat or a dog for the model to predict.
-    '''
-)
-
-# image_path = 'sample_images/hang-niu-Tn8DLxwuDMA-unsplash.jpg'
 
 model = load_model('Pikachu_and_Raichu.h5')
 
+classnames = ['Pikachu', 'Raichu']
 
-def test_image(object_image):
-    # Convert the file to an opencv image.
-    file_bytes = np.asarray(bytearray(object_image.read()), dtype=np.uint8)
-    opencv_image = cv2.imdecode(file_bytes, 1)
-    opencv_image = cv2.resize(opencv_image, (200, 200))
-    opencv_image.shape = (1, 200, 200, 3)
-    opencv_image = preprocess_input(opencv_image)
-    predictions = model.predict(opencv_image)
-
-    if predictions[0, 0] >= 0.5:
-        result = 'DOG'
-        confidence = predictions[0, 0] * 100
-    else:
-        result = 'CAT'
-        confidence = 100 - (predictions[0, 0] * 100)
-
-    return result, round(confidence, 2)
-    # it returns the predicted label and the precision i.e the confidence score
-
-
-object_image = st.file_uploader("Upload an image...", type=[
-                                'png', 'jpg', 'webp', 'jpeg'])
-submit = st.button('Predict')
-
-if submit:
-    if object_image is not None:
-        output = test_image(object_image)
-
-        # Displaying the image
-        st.image(object_image, channels="BGR")
-        st.markdown(f"""## This is an image of a: {output[0]} """)
-        st.write(f'# Confidence: ${ output[1]}$ %')
+def predict(img):
+    img=img.reshape(-1,224,224,3)
+    """images_list = []
+    images_list.append(np.array(img))
+    x = np.asarray(images_list)"""
+    prediction = model.predict(img)[0]
+    return {classnames[i]: float(prediction[i]) for i in range(len(classnames))}
 
+image = gr.inputs.Image(shape=(298, 384))
+label = gr.outputs.Label(num_top_classes=2)
 
-# print(f'The image was predicted as a {test_image(image_path)}')
+gr.Interface(fn=predict, inputs=image,  title="Garbage Classifier",
+    description="This is a Garbage Classification Model Trained using Dataset 11 by Sud.Deployed to Hugging Faces using Gradio.",outputs=label,interpretation='default').launch(debug=True,enable_queue=True)