import tensorflow as tf
import matplotlib.pyplot as plt
train_ds = tf.keras.utils.image_dataset_from_directory(
    "dataset/cats_and_dogs_filtered/train",
    image_size=(160,160), batch_size=32
)
val_ds = tf.keras.utils.image_dataset_from_directory(
    "dataset/cats_and_dogs_filtered/validation",
    image_size=(160,160), batch_size=32
)
print("Classes:", train_ds.class_names)
preprocess = tf.keras.applications.vgg16.preprocess_input  #resnet50.preprocess_input
train_ds = train_ds.map(lambda x,y: (preprocess(x),y)).prefetch(1)
val_ds = val_ds.map(lambda x,y: (preprocess(x),y)).prefetch(1)
base = tf.keras.applications.VGG16(weights='imagenet', include_top=False, input_shape=(160,160,3)) #ResNet50
base.trainable = False
model = tf.keras.Sequential([
    base,
    tf.keras.layers.GlobalAveragePooling2D(),
    tf.keras.layers.Dense(128, activation='relu'),
    tf.keras.layers.Dense(1, activation='sigmoid')
])
model.compile(optimizer='adam', loss='binary_crossentropy', metrics=['accuracy'])
h1 = model.fit(train_ds, validation_data=val_ds, epochs=3)
loss1, acc1 = model.evaluate(val_ds)
print("Frozen VGG16 Accuracy:", acc1*100, "%")
base.trainable = True
for layer in base.layers[:-4]:
    layer.trainable = False
model.compile(optimizer=tf.keras.optimizers.Adam(learning_rate=0.00001), loss='binary_crossentropy', metrics=['accuracy'])
h2 = model.fit(train_ds, validation_data=val_ds, epochs=3)
loss2, acc2 = model.evaluate(val_ds)
print("Fine-tuned VGG16 Accuracy:", acc2*100, "%")
plt.bar(['Frozen VGG16','Fine-tuned VGG16'], [acc1*100,acc2*100])
plt.ylabel("Accuracy (%)")
plt.title("VGG16 Performance Comparison")
plt.show()
plt.plot(h1.history['val_accuracy'], label='Frozen VGG16')
plt.plot([x+3 for x in range(3)], h2.history['val_accuracy'], label='Fine-tuned VGG16')
plt.axvline(x=2, color='r', linestyle='--', label='Fine-tuning')
plt.xlabel("Epochs")
plt.ylabel("Validation Accuracy")
plt.title("VGG16 Accuracy Comparison")
plt.legend()
plt.show()
