Initial commit

.gitignore

@@ -0,0 +1 @@
+/saved_model

app.py

@@ -0,0 +1,52 @@
+import tensorflow as tf
+import gradio as gr
+from scipy.ndimage import zoom
+import cv2
+import numpy as np 
+
+# Load the pre-trained model
+from huggingface_hub import from_pretrained_keras
+
+model = from_pretrained_keras("manufy/mnist_model_keras")
+#model = tf.keras.models.load_model('mnist_model.keras')
+
+def recognize_digit(input):
+    image = input['layers'][0]
+    print("Type of image variable:", type(image))
+    dimensions = image.shape
+    print("Dimensions of the image:", dimensions)
+    # Resize image to 28x28 using interpolation
+    resized_image = cv2.resize(image, (28, 28))
+     # Convert image to grayscale
+    gray_image = cv2.cvtColor(resized_image, cv2.COLOR_RGBA2GRAY)
+    # Reshape the image to add a single channel dimension
+    final_image = np.expand_dims(gray_image, axis=-1)
+    
+    print("Final Dimensions of the image:", final_image.shape)
+   
+    #if image.shape[-1] == 3:  # If the image has 3 color channels (RGB)
+    #    image = image.mean(axis=-1)  # Convert to grayscale
+    # Resize the image to 28x28
+    #image = zoom(image, (28 / image.shape[0], 28 / image.shape[1]))
+    final_image = final_image.reshape((1, 28, 28, 1)).astype('float32') / 255
+    prediction = model.predict(final_image)
+    return {str(i): float(prediction[0][i]) for i in range(10)}, input['layers'][0]
+#final_image[0, :, :, 0]
+
+    #return ''
+
+# Configure the Sketchpad
+sketchpad = gr.Sketchpad()
+
+# Create the Gradio interface
+interface = gr.Interface(
+    fn=recognize_digit, 
+    inputs="sketchpad",
+    outputs=[gr.Label(num_top_classes=3), "image"],
+    live=True,
+    title="Digit Recognizer",
+    description="Draw a digit and let the model recognize it."
+)
+
+# Launch the interface
+interface.launch()

load-model-from-huggingface.py

@@ -0,0 +1,3 @@
+from huggingface_hub import from_pretrained_keras
+
+model = from_pretrained_keras("manufy/mnist_model_keras")

push-model-to-huggingface-with-cli.sh

@@ -0,0 +1,7 @@
+#/bin/bash
+# PUSH model to HuggingFace using CLI
+# HUGGINGFACE_TOKEN should be set on your environment variables
+# commit message is not customizable
+# https://huggingface.co/docs/huggingface_hub/guides/upload
+
+huggingface-cli upload manufy/mnist_model_keras saved_model . --token $HUGGINGFACE_TOKEN

push-model-to-huggingface-with-commit.py

@@ -0,0 +1,36 @@
+# https://huggingface.co/docs/huggingface_hub/guides/upload
+
+# PUSH model to HuggingFace manufy/mnist_model_keras
+# with custom commit message
+
+###############################
+# 1 - Initial generic imports #
+###############################
+
+from icecream import ic
+ic("--- Importing generic libraries ---")
+from dotenv import load_dotenv, find_dotenv
+import os
+from datetime import datetime
+
+###############################################################################
+# 2 - Setting up environment variable HUGGINGFACE_TOKEN with write permission #
+###############################################################################
+
+ic("--- Setting HF API token ---")
+load_dotenv(find_dotenv())
+token = os.getenv("HUGGINGFACE_TOKEN")
+formatted_now = datetime.now().strftime("%d/%m/%Y, %H:%M:%S")
+ic("--- Importing HF API ---")
+from huggingface_hub import HfApi, CommitOperationAdd, CommitOperationDelete
+ic("--- Executing commit with custom message ---")
+api = HfApi()
+operations = [
+    CommitOperationAdd(path_in_repo="saved_model", path_or_fileobj="mnist_model.keras")
+]
+api.create_commit(
+    repo_id="manufy/mnist_model_keras",
+    operations=operations,
+    commit_message=f"Updated model at  {formatted_now}.",
+    token=token
+)

push-model-to-huggingface-with-upload-folder.py

@@ -0,0 +1,76 @@
+# https://huggingface.co/docs/huggingface_hub/guides/upload
+
+# PUSH model to HuggingFace manufy/mnist_model_keras
+# using upload_folder, commit message is set by defaulr and not customizable
+
+###############################
+# 1 - Initial generic imports #
+###############################
+
+from icecream import ic
+ic("--- Importing generic libraries ---")
+from dotenv import load_dotenv, find_dotenv
+import os
+
+###############################################################################
+# 2 - Setting up environment variable HUGGINGFACE_TOKEN with write permission #
+###############################################################################
+
+ic("--- Importing HF API token ---")
+load_dotenv(find_dotenv())
+token = os.getenv("HUGGINGFACE_TOKEN")
+
+# The code in the comments is not needed
+# the model doen't need to be loaded to be uploaded
+
+################################################################
+# 3 - Importing keras load_model #
+################################################################
+
+# ic("--- Importing tensorflow load_model ---")
+# from tensorflow.keras.models import load_model
+# ic("--- Importing huggingface_hub ---")
+
+###########################
+# 3 - Loading keras model #
+###########################
+
+# ic("--- Loading local keras model ---"  )
+# model = load_model('mnist_model.keras')
+
+######################################
+# 3 - Uploading model to HuggingFace #
+######################################
+
+ic ("--- Importing HF API ---")
+from huggingface_hub import HfApi
+ic("--- Settig up HF API ---")
+HuggingFace_api = HfApi()
+ic("--- Uploading model ---")
+HuggingFace_api.upload_folder(
+    folder_path="saved_model/",
+    repo_id="manufy/mnist_model_keras",
+    repo_type="model",
+    token=token)
+
+########################################################################
+# The following are failed attempts to upload the model to HuggingFace #
+########################################################################
+
+# upload_folder("manufy/mnist_model_keras")
+
+# not supported in keras 3
+# NotImplementedError: Cannot use 'save_pretrained_keras': 
+# Keras 3.x is not supported.
+# Please save models manually and upload them
+# using `upload_folder` or `huggingface-cli upload`.
+
+#ic("--- Saving huggingface model ---"   )
+#huggingface_hub.save_pretrained_keras('saved_model/')
+
+
+#push_to_hub_keras(model,
+#    "manufy/mnist_model_keras")
+#    include_optimizer = True,
+#    tags = ["object-detection", "some_other_tag"]
+#)

requirements.txt

@@ -0,0 +1,3 @@
+pencv-python
+tensorflow
+icecream

sketchpad.py

@@ -0,0 +1,31 @@
+
+import gradio as gr
+import numpy as np
+import cv2
+
+
+
+def classify(input):
+    image = input['layers'][0]
+    print("Type of image variable:", type(image))
+    dimensions = image.shape
+    print("Dimensions of the image:", dimensions)
+    # Resize image to 28x28 using interpolation
+    resized_image = cv2.resize(image, (28, 28))
+     # Convert image to grayscale
+    gray_image = cv2.cvtColor(resized_image, cv2.COLOR_RGBA2GRAY)
+    # Reshape the image to add a single channel dimension
+    final_image = np.expand_dims(gray_image, axis=-1)
+    
+    print("Final image shape:", final_image.shape)
+    print("Final image:", final_image)
+    
+    #input = np.reshape(input, (1, 28, 28))
+    #print(input)
+    return ''
+
+label = gr.Label(num_top_classes=10)
+
+interface = gr.Interface(fn=classify, inputs="sketchpad", outputs=label,
+live=True)
+interface.launch()

training.py

@@ -0,0 +1,74 @@
+
+from icecream import ic
+ic("--- Importing tensorflow ---")
+import tensorflow as tf
+from tensorflow.keras import layers, models
+from tensorflow. keras. utils import plot_model
+from tensorflow.keras import Input
+
+# load mnist dataset
+ic("------ Loading mnist dataset ------")
+(train_images, train_labels), (test_images, test_labels) = tf.keras.datasets.mnist.load_data()
+
+# normalize 60000 instances, 28x28 pixels 1 channel
+
+ic("------ Normalizing data ------")
+train_images = train_images.reshape((60000, 28, 28, 1)).astype('float32') / 255
+test_images  = test_images.reshape((10000, 28, 28, 1)).astype('float32') / 255
+
+# labels are output numbers, 0 to 9 we need to convert them to one-hot encoding
+
+ic("------ One-hot encoding labels ------")
+train_labels = tf.keras.utils.to_categorical(train_labels)
+
+# 1 for correct digit, 0 for incorrect 
+
+
+ic("------ Creating model ------")
+# define model
+model = models.Sequential()
+
+
+# Add an Input layer
+model.add(Input(shape=(28, 28, 1)))
+
+# create convolutional layer 
+# 32 filters, 3x3 kernel, relu activation function, input shape 28x28x1
+# them create max pooling layer 2x2
+
+#model.add(layers.Conv2D(32, (3, 3), activation='relu', input_shape=(28, 28, 1)))
+model.add(layers.MaxPooling2D((2, 2)))
+
+model.add(layers.Conv2D(64, (3, 3), activation='relu',))
+model.add(layers.MaxPooling2D((2, 2)))
+
+model.add(layers.Conv2D(64, (3, 3), activation='relu',))
+model.add(layers.Flatten())
+
+model.add(layers.Dense(64, activation='relu'))
+
+# output layer, 
+
+model.add(layers.Dense(10, activation='softmax'))
+
+model.compile(optimizer='adam', loss='categorical_crossentropy', metrics=['accuracy'])
+
+# generate model.png with architecture plot
+
+ic("------ Plotting model ------")
+plot_model(model, to_file='static/model.png', show_shapes=True, show_layer_names=True)
+
+
+# train
+
+ic("------ Training model ------")
+#model.fit(train_images, train_labels, epochs=5, batch_size=64, validation_split=0.1)
+
+# save model
+
+ic("------ Saving .h5 model ------")
+model.save('saved_model/keras/mnist_model.h5')
+ic("------ Saving .keras model ------")
+model.save('saved_model/keras/mnist_model.keras')
+ic("------ Exporting .keras model ------")
+model.export('saved_model/exported')