csy3025-assignment-2/main.py

import numpy as np
import matplotlib.pyplot as plt
import tensorflow as tf 
import os 


#Import Data
PATH = os.path.join(os.getcwd(), "crop-data")

training_data = os.path.join(PATH, "train")
validation_data = os.path.join(PATH, "valid")
test_data = os.path.join(PATH, "test")

BATCH_SIZE = 64
IMG_SIZE = (160,160)

training_data = tf.keras.utils.image_dataset_from_directory(training_data,
                                                            shuffle=True,
                                                            batch_size=BATCH_SIZE,
                                                            image_size=IMG_SIZE)
validation_data = tf.keras.utils.image_dataset_from_directory(validation_data,
                                                            shuffle=True,
                                                            batch_size=BATCH_SIZE,
                                                            image_size=IMG_SIZE)

test_data = tf.keras.utils.image_dataset_from_directory(test_data,
                                                            shuffle=True,
                                                            batch_size=BATCH_SIZE,
                                                            image_size=IMG_SIZE)

class_names = training_data.class_names

#View Data
plt.figure(figsize=(10,10))
for images, labels in training_data.take(1):
    for i in range(9):
        ax = plt.subplot(3, 3, i+1)
        plt.imshow(images[i].numpy().astype("uint8"))
        plt.title(class_names[labels[i]])
        plt.axis("off")
plt.show()


#Init Prefetching
AUTOTUNE = tf.data.AUTOTUNE
training_data = training_data.prefetch(buffer_size=AUTOTUNE)
validation_data = validation_data.prefetch(buffer_size=AUTOTUNE)
test_data = test_data.prefetch(buffer_size=AUTOTUNE)


#Data Augmentation
data_augmentation = tf.keras.Sequential([
    tf.keras.layers.RandomFlip('horisontal'),
    tf.keras.layers.RandomRotation(0.2)
    ])

#Rescale Pixel Values
preprocess_input = tf.keras.applications.mobilenet_v3.preprocess_input

rescale = tf.keras.layers.Rescaling(1./127.5, offset=-1)

#Create Base Model From MobileNetV3
IMG_SHAPE = IMG_SIZE + (3,)
base_model = tf.keras.applications.MobileNetV3Large(
        input_shape=IMG_SHAPE,
        include_top=False,
        weights="imagenet"
        )

image_batch, label_batch = next(iter(training_data))
feature_batch = base_model(image_batch)

base_model.trainable = False


#View Base Model
base_model.summary()

#Add Classification Header


global_avg_layer = tf.keras.layers.GlobalAveragePooling2D()
feature_batch_avg = global_avg_layer(feature_batch)

prediction_layer = tf.keras.layers.Dense(38, activation="softmax")
predication_batch = prediction_layer(feature_batch_avg)

inputs = tf.keras.Input(shape=(160,160,3))
x = data_augmentation(inputs)
x = preprocess_input(x)
x = base_model(x, training=False)
x = global_avg_layer(x)
x = tf.keras.layers.Dropout(0.2)(x)
outputs = prediction_layer(x)

model = tf.keras.Model(inputs, outputs)

#View Model with Classification Head
model.summary()


#Compile the Model
base_learning_rate = 0.0001

training_data = training_data.map(lambda x,y: (x, tf.one_hot(y,38)))
validation_data = validation_data.map(lambda x,y: (x, tf.one_hot(y,38)))



optimizer = tf.keras.optimizers.Adam(learning_rate=base_learning_rate)
loss = tf.keras.losses.CategoricalCrossentropy()
metrics = [tf.keras.metrics.CategoricalAccuracy()]


phy_dev = tf.config.list_physical_devices("GPU")
for gpu in phy_dev:
    tf.config.experimental.set_memory_growth(gpu, True)

model.compile(optimizer=optimizer, loss=loss, metrics=metrics)

#Train the Model
initial_epochs = 50

loss0, accuracy0 = model.evaluate(validation_data)

print(f"initial loss: {loss0}")
print(f"initial accuracy: {accuracy0}")

lr_schedule = tf.keras.callbacks.ReduceLROnPlateau(
        monitor="val_loss",
        factor=0.1,
        patience=5,
        min_lr=1e-6
        )

earky_stopping = tf.keras.callbacks.EarlyStopping(
        monitor="val_loss",
        patience=10,
        restore_best_weights=True
        )

history = model.fit(training_data,
                    epochs=initial_epochs,
                    validation_data=validation_data,
                    callbacks=[lr_schedule])

model.save("crop-classifier-callbacks.keras")