Testing a different configuration

app.py

@@ -1,74 +1,72 @@
 import os
 import gradio as gr
 import tensorflow as tf
-from tensorflow.keras.applications import EfficientNetB0
-from tensorflow.keras.models import Model
-from tensorflow.keras.layers import Dense, GlobalAveragePooling2D, Input
-
-# Create the model using TensorFlow's native EfficientNet implementation
-input_tensor = Input(shape=(32, 32, 1))
-# Use a preprocessing layer to handle grayscale to RGB conversion
-x = tf.keras.layers.Rescaling(1.0 / 255)(input_tensor)
-# Repeat the grayscale channel to create a 3-channel input compatible with EfficientNet
-x = tf.keras.layers.Concatenate(axis=-1)([x, x, x])
-
-
-base_model = EfficientNetB0(include_top=False, weights=None, input_tensor=x)
-x = GlobalAveragePooling2D()(base_model.output)
-x = Dense(10, activation="softmax")(x)
-model = Model(inputs=input_tensor, outputs=x)
+from tensorflow import keras
+from tensorflow.keras.models import load_model
+from tensorflow.keras import layers
+from efficientnet.tfkeras import EfficientNetB0
+
+model = keras.Sequential(
+    [
+        layers.Input(shape=(32, 32, 1)),
+        layers.experimental.preprocessing.Rescaling(1.0 / 255),
+        EfficientNetB0(include_top=False, weights=None, input_shape=(32, 32, 1)),
+        layers.GlobalAveragePooling2D(),
+        layers.Dropout(0.2),
+        layers.Dense(10),
+    ]
+)
 
 model.compile(
-    optimizer=tf.keras.optimizers.Adam(learning_rate=1e-3),
-    loss=tf.keras.losses.SparseCategoricalCrossentropy(),
+    optimizer=keras.optimizers.Adam(learning_rate=1e-3),
+    loss=keras.losses.SparseCategoricalCrossentropy(from_logits=True),
     metrics=["accuracy"],
 )
 
-# Load weights if you have them
-try:
-    model.load_weights("my_model.h5")
-except:
-    print("Could not load weights. Using untrained model.")
+model = load_model("my_model.h5")
 
 
 def classify_image(image):
     # Preprocess the image
     image_gray = tf.image.rgb_to_grayscale(image)
-    image_tensor = tf.image.resize(image_gray, (32, 32))
+    image_tensor = tf.convert_to_tensor(image_gray)
 
-    # Add a batch dimension
+    # Resize the image to 28x28.
+    image_tensor = tf.image.resize(image_tensor, (32, 32))
+
+    # Cast the data to float32.
+    image_tensor = tf.cast(image_tensor, tf.float32)
+
+    # Add a batch dimension.
     image_tensor = tf.expand_dims(image_tensor, 0)
 
-    # Get the prediction
+    # Normalize the data.
+    image_tensor = image_tensor / 255.0
+
+    # Get the prediction.
     prediction = model.predict(image_tensor)
 
-    # Convert the prediction to a dictionary of label-confidence pairs
-    top_indices = tf.argsort(prediction[0])[-3:][::-1]  # Top 3 indices
-    confidences = {str(idx): float(prediction[0][idx]) for idx in top_indices}
+    # Convert the prediction to a string label.
+    prediction_label = str(prediction.argmax())
 
-    return confidences
+    return prediction_label
 
 
-title = "MNIST Model 98% Accuracy"
-description = "Model trained on MNIST dataset using EfficientNet to classify handwritten digits with 98% accuracy"
-article = "For source code, visit [my GitHub](https://github.com/Bijan-K/Tensorflow-MNIST-98Acc.git)"
+title = "MNIST Model 98%acc"
+description = "Model trained on MNIST dataset using efficientnet to classify MNIST images with 98% accuracy"
+article = "for source code you can visit [my github](https://github.com/Bijan-K/Tensorflow-MNIST-98Acc.git) (gradio + training code)."
 
-example_list = []
-if os.path.exists("examples"):
-    example_list = [
-        ["examples/" + example]
-        for example in os.listdir("examples")
-        if os.path.isfile(os.path.join("examples", example))
-    ]
+example_list = [["examples/" + example] for example in os.listdir("examples")]
 
 interface = gr.Interface(
     fn=classify_image,
     inputs=gr.Image(type="pil"),
-    outputs=gr.Label(num_top_classes=3),
+    outputs=gr.Label(num_top_classes=3, label="Predictions"),
     examples=example_list,
     title=title,
     description=description,
     article=article,
 )
 
+
 interface.launch()

requirements.txt

@@ -1,4 +1,3 @@
-tensorflow==2.9.1
-keras==2.9.0
+tensorflow==2.6.0
 efficientnet==1.1.1
-gradio==3.32.0
+gradio