 View on TensorFlow.org
|
 View source on GitHub
|
 Download notebook
|
Overview
This tutorial demonstrates manual image manipulations and augmentation using tf.image.
Data augmentation is a common technique to improve results and avoid overfitting, see Overfitting and Underfitting for others.
Setup
pip install -q git+https://github.com/tensorflow/docs
import urllib
import tensorflow as tf
from tensorflow.keras.datasets import mnist
from tensorflow.keras import layers
AUTOTUNE = tf.data.experimental.AUTOTUNE
import tensorflow_docs as tfdocs
import tensorflow_docs.plots
import tensorflow_datasets as tfds
import PIL.Image
import matplotlib.pyplot as plt
import matplotlib as mpl
mpl.rcParams['figure.figsize'] = (12, 5)
import numpy as np
Let's check the data augmentation features on an image and then augment a whole dataset later to train a model.
Download this image, by Von.grzanka, for augmentation.
image_path = tf.keras.utils.get_file("cat.jpg", "https://storage.googleapis.com/download.tensorflow.org/example_images/320px-Felis_catus-cat_on_snow.jpg")
PIL.Image.open(image_path)
Downloading data from https://storage.googleapis.com/download.tensorflow.org/example_images/320px-Felis_catus-cat_on_snow.jpg 24576/17858 [=========================================] - 0s 0us/step
Read and decode the image to tensor format.
image_string=tf.io.read_file(image_path)
image=tf.image.decode_jpeg(image_string,channels=3)
A function to visualize and compare the original and augmented image side by side.
def visualize(original, augmented):
fig = plt.figure()
plt.subplot(1,2,1)
plt.title('Original image')
plt.imshow(original)
plt.subplot(1,2,2)
plt.title('Augmented image')
plt.imshow(augmented)
Augment a single image
Flipping the image
Flip the image either vertically or horizontally.
flipped = tf.image.flip_left_right(image)
visualize(image, flipped)
Grayscale the image
Grayscale an image.
grayscaled = tf.image.rgb_to_grayscale(image)
visualize(image, tf.squeeze(grayscaled))
plt.colorbar()
<matplotlib.colorbar.Colorbar at 0x7f08102b1160>
Saturate the image
Saturate an image by providing a saturation factor.
saturated = tf.image.adjust_saturation(image, 3)
visualize(image, saturated)
Change image brightness
Change the brightness of image by providing a brightness factor.
bright = tf.image.adjust_brightness(image, 0.4)
visualize(image, bright)
Rotate the image
Rotate an image by 90 degrees.
rotated = tf.image.rot90(image)
visualize(image, rotated)
Center crop the image
Crop the image from center upto the image part you desire.
cropped = tf.image.central_crop(image, central_fraction=0.5)
visualize(image,cropped)
See the tf.image reference for details about available augmentation options.
Augment a dataset and train a model with it
Train a model on an augmented dataset.
dataset, info = tfds.load('mnist', as_supervised=True, with_info=True)
train_dataset, test_dataset = dataset['train'], dataset['test']
num_train_examples= info.splits['train'].num_examples
Downloading and preparing dataset mnist/3.0.0 (download: 11.06 MiB, generated: Unknown size, total: 11.06 MiB) to /home/kbuilder/tensorflow_datasets/mnist/3.0.0... Warning:absl:Dataset mnist is hosted on GCS. It will automatically be downloaded to your local data directory. If you'd instead prefer to read directly from our public GCS bucket (recommended if you're running on GCP), you can instead set data_dir=gs://tfds-data/datasets. HBox(children=(FloatProgress(value=0.0, description='Dl Completed...', max=4.0, style=ProgressStyle(descriptio… Dataset mnist downloaded and prepared to /home/kbuilder/tensorflow_datasets/mnist/3.0.0. Subsequent calls will reuse this data.
Write a function to augment the images. Map it over the the dataset. This returns a dataset that augments the data on the fly.
def convert(image, label):
image = tf.image.convert_image_dtype(image, tf.float32) # Cast and normalize the image to [0,1]
return image, label
def augment(image,label):
image,label = convert(image, label)
image = tf.image.convert_image_dtype(image, tf.float32) # Cast and normalize the image to [0,1]
image = tf.image.resize_with_crop_or_pad(image, 34, 34) # Add 6 pixels of padding
image = tf.image.random_crop(image, size=[28, 28, 1]) # Random crop back to 28x28
image = tf.image.random_brightness(image, max_delta=0.5) # Random brightness
return image,label
BATCH_SIZE = 64
# Only use a subset of the data so it's easier to overfit, for this tutorial
NUM_EXAMPLES = 2048
Create the augmented dataset.
augmented_train_batches = (
train_dataset
# Only train on a subset, so you can quickly see the effect.
.take(NUM_EXAMPLES)
.cache()
.shuffle(num_train_examples//4)
# The augmentation is added here.
.map(augment, num_parallel_calls=AUTOTUNE)
.batch(BATCH_SIZE)
.prefetch(AUTOTUNE)
)
And a non-augmented one for comparison.
non_augmented_train_batches = (
train_dataset
# Only train on a subset, so you can quickly see the effect.
.take(NUM_EXAMPLES)
.cache()
.shuffle(num_train_examples//4)
# No augmentation.
.map(convert, num_parallel_calls=AUTOTUNE)
.batch(BATCH_SIZE)
.prefetch(AUTOTUNE)
)
Setup the validation dataset. This doesn't change whether or not you're using the augmentation.
validation_batches = (
test_dataset
.map(convert, num_parallel_calls=AUTOTUNE)
.batch(2*BATCH_SIZE)
)
Create and compile the model. The model is a two layered, fully-connected neural network without convolution.
def make_model():
model = tf.keras.Sequential([
layers.Flatten(input_shape=(28, 28, 1)),
layers.Dense(4096, activation='relu'),
layers.Dense(4096, activation='relu'),
layers.Dense(10)
])
model.compile(optimizer = 'adam',
loss=tf.losses.SparseCategoricalCrossentropy(from_logits=True),
metrics=['accuracy'])
return model
Train the model, without augmentation:
model_without_aug = make_model()
no_aug_history = model_without_aug.fit(non_augmented_train_batches, epochs=50, validation_data=validation_batches)
Epoch 1/50 32/32 [==============================] - 1s 38ms/step - loss: 0.7947 - accuracy: 0.7485 - val_loss: 0.3258 - val_accuracy: 0.9008 Epoch 2/50 32/32 [==============================] - 1s 23ms/step - loss: 0.1551 - accuracy: 0.9512 - val_loss: 0.3558 - val_accuracy: 0.9033 Epoch 3/50 32/32 [==============================] - 1s 23ms/step - loss: 0.0937 - accuracy: 0.9692 - val_loss: 0.3041 - val_accuracy: 0.9170 Epoch 4/50 32/32 [==============================] - 1s 22ms/step - loss: 0.0413 - accuracy: 0.9839 - val_loss: 0.3262 - val_accuracy: 0.9207 Epoch 5/50 32/32 [==============================] - 1s 22ms/step - loss: 0.0206 - accuracy: 0.9927 - val_loss: 0.3151 - val_accuracy: 0.9272 Epoch 6/50 32/32 [==============================] - 1s 23ms/step - loss: 0.0179 - accuracy: 0.9932 - val_loss: 0.3467 - val_accuracy: 0.9246 Epoch 7/50 32/32 [==============================] - 1s 22ms/step - loss: 0.0351 - accuracy: 0.9907 - val_loss: 0.3731 - val_accuracy: 0.9227 Epoch 8/50 32/32 [==============================] - 1s 23ms/step - loss: 0.0772 - accuracy: 0.9751 - val_loss: 0.3367 - val_accuracy: 0.9200 Epoch 9/50 32/32 [==============================] - 1s 22ms/step - loss: 0.0579 - accuracy: 0.9805 - val_loss: 0.4197 - val_accuracy: 0.9141 Epoch 10/50 32/32 [==============================] - 1s 22ms/step - loss: 0.0341 - accuracy: 0.9878 - val_loss: 0.4159 - val_accuracy: 0.9221 Epoch 11/50 32/32 [==============================] - 1s 23ms/step - loss: 0.0444 - accuracy: 0.9897 - val_loss: 0.4530 - val_accuracy: 0.9180 Epoch 12/50 32/32 [==============================] - 1s 23ms/step - loss: 0.0440 - accuracy: 0.9888 - val_loss: 0.3899 - val_accuracy: 0.9208 Epoch 13/50 32/32 [==============================] - 1s 22ms/step - loss: 0.0102 - accuracy: 0.9961 - val_loss: 0.4007 - val_accuracy: 0.9254 Epoch 14/50 32/32 [==============================] - 1s 22ms/step - loss: 0.0085 - accuracy: 0.9976 - val_loss: 0.3633 - val_accuracy: 0.9352 Epoch 15/50 32/32 [==============================] - 1s 22ms/step - loss: 0.0060 - accuracy: 0.9985 - val_loss: 0.4455 - val_accuracy: 0.9220 Epoch 16/50 32/32 [==============================] - 1s 23ms/step - loss: 0.0384 - accuracy: 0.9922 - val_loss: 0.3923 - val_accuracy: 0.9291 Epoch 17/50 32/32 [==============================] - 1s 23ms/step - loss: 0.0142 - accuracy: 0.9956 - val_loss: 0.3724 - val_accuracy: 0.9293 Epoch 18/50 32/32 [==============================] - 1s 23ms/step - loss: 0.0019 - accuracy: 1.0000 - val_loss: 0.3417 - val_accuracy: 0.9360 Epoch 19/50 32/32 [==============================] - 1s 22ms/step - loss: 3.0518e-04 - accuracy: 1.0000 - val_loss: 0.3471 - val_accuracy: 0.9406 Epoch 20/50 32/32 [==============================] - 1s 23ms/step - loss: 2.2373e-04 - accuracy: 1.0000 - val_loss: 0.3493 - val_accuracy: 0.9405 Epoch 21/50 32/32 [==============================] - 1s 22ms/step - loss: 1.0900e-04 - accuracy: 1.0000 - val_loss: 0.3590 - val_accuracy: 0.9389 Epoch 22/50 32/32 [==============================] - 1s 22ms/step - loss: 8.8505e-05 - accuracy: 1.0000 - val_loss: 0.3616 - val_accuracy: 0.9393 Epoch 23/50 32/32 [==============================] - 1s 25ms/step - loss: 6.8111e-05 - accuracy: 1.0000 - val_loss: 0.3655 - val_accuracy: 0.9391 Epoch 24/50 32/32 [==============================] - 1s 23ms/step - loss: 5.1576e-05 - accuracy: 1.0000 - val_loss: 0.3731 - val_accuracy: 0.9392 Epoch 25/50 32/32 [==============================] - 1s 22ms/step - loss: 3.7954e-05 - accuracy: 1.0000 - val_loss: 0.3832 - val_accuracy: 0.9394 Epoch 26/50 32/32 [==============================] - 1s 22ms/step - loss: 2.6865e-05 - accuracy: 1.0000 - val_loss: 0.3915 - val_accuracy: 0.9396 Epoch 27/50 32/32 [==============================] - 1s 23ms/step - loss: 2.0358e-05 - accuracy: 1.0000 - val_loss: 0.4018 - val_accuracy: 0.9394 Epoch 28/50 32/32 [==============================] - 1s 22ms/step - loss: 1.5839e-05 - accuracy: 1.0000 - val_loss: 0.4086 - val_accuracy: 0.9394 Epoch 29/50 32/32 [==============================] - 1s 22ms/step - loss: 1.2895e-05 - accuracy: 1.0000 - val_loss: 0.4148 - val_accuracy: 0.9393 Epoch 30/50 32/32 [==============================] - 1s 23ms/step - loss: 1.0775e-05 - accuracy: 1.0000 - val_loss: 0.4208 - val_accuracy: 0.9394 Epoch 31/50 32/32 [==============================] - 1s 22ms/step - loss: 9.2570e-06 - accuracy: 1.0000 - val_loss: 0.4253 - val_accuracy: 0.9394 Epoch 32/50 32/32 [==============================] - 1s 23ms/step - loss: 8.1208e-06 - accuracy: 1.0000 - val_loss: 0.4302 - val_accuracy: 0.9396 Epoch 33/50 32/32 [==============================] - 1s 25ms/step - loss: 7.1906e-06 - accuracy: 1.0000 - val_loss: 0.4337 - val_accuracy: 0.9398 Epoch 34/50 32/32 [==============================] - 1s 23ms/step - loss: 6.5134e-06 - accuracy: 1.0000 - val_loss: 0.4373 - val_accuracy: 0.9398 Epoch 35/50 32/32 [==============================] - 1s 22ms/step - loss: 5.8839e-06 - accuracy: 1.0000 - val_loss: 0.4412 - val_accuracy: 0.9398 Epoch 36/50 32/32 [==============================] - 1s 22ms/step - loss: 5.3458e-06 - accuracy: 1.0000 - val_loss: 0.4449 - val_accuracy: 0.9395 Epoch 37/50 32/32 [==============================] - 1s 22ms/step - loss: 4.9340e-06 - accuracy: 1.0000 - val_loss: 0.4474 - val_accuracy: 0.9395 Epoch 38/50 32/32 [==============================] - 1s 22ms/step - loss: 4.5733e-06 - accuracy: 1.0000 - val_loss: 0.4503 - val_accuracy: 0.9394 Epoch 39/50 32/32 [==============================] - 1s 22ms/step - loss: 4.2514e-06 - accuracy: 1.0000 - val_loss: 0.4525 - val_accuracy: 0.9394 Epoch 40/50 32/32 [==============================] - 1s 23ms/step - loss: 3.9590e-06 - accuracy: 1.0000 - val_loss: 0.4553 - val_accuracy: 0.9394 Epoch 41/50 32/32 [==============================] - 1s 23ms/step - loss: 3.7258e-06 - accuracy: 1.0000 - val_loss: 0.4568 - val_accuracy: 0.9393 Epoch 42/50 32/32 [==============================] - 1s 22ms/step - loss: 3.4758e-06 - accuracy: 1.0000 - val_loss: 0.4596 - val_accuracy: 0.9391 Epoch 43/50 32/32 [==============================] - 1s 23ms/step - loss: 3.3092e-06 - accuracy: 1.0000 - val_loss: 0.4624 - val_accuracy: 0.9391 Epoch 44/50 32/32 [==============================] - 1s 23ms/step - loss: 3.1065e-06 - accuracy: 1.0000 - val_loss: 0.4645 - val_accuracy: 0.9389 Epoch 45/50 32/32 [==============================] - 1s 23ms/step - loss: 2.9453e-06 - accuracy: 1.0000 - val_loss: 0.4658 - val_accuracy: 0.9391 Epoch 46/50 32/32 [==============================] - 1s 23ms/step - loss: 2.7940e-06 - accuracy: 1.0000 - val_loss: 0.4682 - val_accuracy: 0.9392 Epoch 47/50 32/32 [==============================] - 1s 23ms/step - loss: 2.6627e-06 - accuracy: 1.0000 - val_loss: 0.4696 - val_accuracy: 0.9392 Epoch 48/50 32/32 [==============================] - 1s 23ms/step - loss: 2.5305e-06 - accuracy: 1.0000 - val_loss: 0.4716 - val_accuracy: 0.9392 Epoch 49/50 32/32 [==============================] - 1s 23ms/step - loss: 2.4067e-06 - accuracy: 1.0000 - val_loss: 0.4733 - val_accuracy: 0.9393 Epoch 50/50 32/32 [==============================] - 1s 22ms/step - loss: 2.3043e-06 - accuracy: 1.0000 - val_loss: 0.4741 - val_accuracy: 0.9394
Train it again with augmentation:
model_with_aug = make_model()
aug_history = model_with_aug.fit(augmented_train_batches, epochs=50, validation_data=validation_batches)
Epoch 1/50 32/32 [==============================] - 1s 35ms/step - loss: 2.3106 - accuracy: 0.3115 - val_loss: 1.2095 - val_accuracy: 0.6655 Epoch 2/50 32/32 [==============================] - 1s 23ms/step - loss: 1.3821 - accuracy: 0.5371 - val_loss: 0.7040 - val_accuracy: 0.7943 Epoch 3/50 32/32 [==============================] - 1s 22ms/step - loss: 0.9761 - accuracy: 0.6709 - val_loss: 0.5867 - val_accuracy: 0.8518 Epoch 4/50 32/32 [==============================] - 1s 23ms/step - loss: 0.7830 - accuracy: 0.7437 - val_loss: 0.4247 - val_accuracy: 0.8816 Epoch 5/50 32/32 [==============================] - 1s 23ms/step - loss: 0.6546 - accuracy: 0.7715 - val_loss: 0.3521 - val_accuracy: 0.8943 Epoch 6/50 32/32 [==============================] - 1s 23ms/step - loss: 0.5795 - accuracy: 0.8057 - val_loss: 0.3036 - val_accuracy: 0.9034 Epoch 7/50 32/32 [==============================] - 1s 22ms/step - loss: 0.4934 - accuracy: 0.8350 - val_loss: 0.2950 - val_accuracy: 0.9088 Epoch 8/50 32/32 [==============================] - 1s 23ms/step - loss: 0.4900 - accuracy: 0.8447 - val_loss: 0.2640 - val_accuracy: 0.9179 Epoch 9/50 32/32 [==============================] - 1s 23ms/step - loss: 0.4573 - accuracy: 0.8408 - val_loss: 0.2528 - val_accuracy: 0.9191 Epoch 10/50 32/32 [==============================] - 1s 23ms/step - loss: 0.4360 - accuracy: 0.8623 - val_loss: 0.2463 - val_accuracy: 0.9224 Epoch 11/50 32/32 [==============================] - 1s 23ms/step - loss: 0.3768 - accuracy: 0.8765 - val_loss: 0.2337 - val_accuracy: 0.9238 Epoch 12/50 32/32 [==============================] - 1s 24ms/step - loss: 0.3681 - accuracy: 0.8823 - val_loss: 0.2051 - val_accuracy: 0.9351 Epoch 13/50 32/32 [==============================] - 1s 23ms/step - loss: 0.3783 - accuracy: 0.8804 - val_loss: 0.2184 - val_accuracy: 0.9315 Epoch 14/50 32/32 [==============================] - 1s 24ms/step - loss: 0.3429 - accuracy: 0.8887 - val_loss: 0.1834 - val_accuracy: 0.9430 Epoch 15/50 32/32 [==============================] - 1s 23ms/step - loss: 0.3271 - accuracy: 0.8945 - val_loss: 0.2174 - val_accuracy: 0.9310 Epoch 16/50 32/32 [==============================] - 1s 23ms/step - loss: 0.3005 - accuracy: 0.8975 - val_loss: 0.2344 - val_accuracy: 0.9302 Epoch 17/50 32/32 [==============================] - 1s 23ms/step - loss: 0.3343 - accuracy: 0.8926 - val_loss: 0.1898 - val_accuracy: 0.9398 Epoch 18/50 32/32 [==============================] - 1s 23ms/step - loss: 0.3340 - accuracy: 0.8867 - val_loss: 0.1854 - val_accuracy: 0.9413 Epoch 19/50 32/32 [==============================] - 1s 23ms/step - loss: 0.2615 - accuracy: 0.9116 - val_loss: 0.2155 - val_accuracy: 0.9310 Epoch 20/50 32/32 [==============================] - 1s 23ms/step - loss: 0.3074 - accuracy: 0.8950 - val_loss: 0.1897 - val_accuracy: 0.9412 Epoch 21/50 32/32 [==============================] - 1s 23ms/step - loss: 0.2912 - accuracy: 0.9087 - val_loss: 0.1575 - val_accuracy: 0.9523 Epoch 22/50 32/32 [==============================] - 1s 23ms/step - loss: 0.2821 - accuracy: 0.9067 - val_loss: 0.1591 - val_accuracy: 0.9515 Epoch 23/50 32/32 [==============================] - 1s 23ms/step - loss: 0.2324 - accuracy: 0.9233 - val_loss: 0.1755 - val_accuracy: 0.9436 Epoch 24/50 32/32 [==============================] - 1s 23ms/step - loss: 0.2290 - accuracy: 0.9224 - val_loss: 0.1562 - val_accuracy: 0.9515 Epoch 25/50 32/32 [==============================] - 1s 23ms/step - loss: 0.2217 - accuracy: 0.9307 - val_loss: 0.1724 - val_accuracy: 0.9487 Epoch 26/50 32/32 [==============================] - 1s 23ms/step - loss: 0.2533 - accuracy: 0.9160 - val_loss: 0.1754 - val_accuracy: 0.9452 Epoch 27/50 32/32 [==============================] - 1s 23ms/step - loss: 0.2780 - accuracy: 0.9111 - val_loss: 0.1715 - val_accuracy: 0.9484 Epoch 28/50 32/32 [==============================] - 1s 23ms/step - loss: 0.2000 - accuracy: 0.9341 - val_loss: 0.1744 - val_accuracy: 0.9459 Epoch 29/50 32/32 [==============================] - 1s 23ms/step - loss: 0.2195 - accuracy: 0.9341 - val_loss: 0.1528 - val_accuracy: 0.9534 Epoch 30/50 32/32 [==============================] - 1s 23ms/step - loss: 0.2188 - accuracy: 0.9277 - val_loss: 0.1650 - val_accuracy: 0.9480 Epoch 31/50 32/32 [==============================] - 1s 24ms/step - loss: 0.2226 - accuracy: 0.9287 - val_loss: 0.1505 - val_accuracy: 0.9543 Epoch 32/50 32/32 [==============================] - 1s 23ms/step - loss: 0.2362 - accuracy: 0.9248 - val_loss: 0.1791 - val_accuracy: 0.9472 Epoch 33/50 32/32 [==============================] - 1s 23ms/step - loss: 0.2175 - accuracy: 0.9277 - val_loss: 0.1589 - val_accuracy: 0.9508 Epoch 34/50 32/32 [==============================] - 1s 24ms/step - loss: 0.2393 - accuracy: 0.9248 - val_loss: 0.1736 - val_accuracy: 0.9455 Epoch 35/50 32/32 [==============================] - 1s 23ms/step - loss: 0.1797 - accuracy: 0.9370 - val_loss: 0.1602 - val_accuracy: 0.9515 Epoch 36/50 32/32 [==============================] - 1s 24ms/step - loss: 0.2059 - accuracy: 0.9326 - val_loss: 0.1544 - val_accuracy: 0.9525 Epoch 37/50 32/32 [==============================] - 1s 24ms/step - loss: 0.1548 - accuracy: 0.9492 - val_loss: 0.1524 - val_accuracy: 0.9537 Epoch 38/50 32/32 [==============================] - 1s 24ms/step - loss: 0.1691 - accuracy: 0.9473 - val_loss: 0.1896 - val_accuracy: 0.9385 Epoch 39/50 32/32 [==============================] - 1s 24ms/step - loss: 0.1962 - accuracy: 0.9360 - val_loss: 0.1591 - val_accuracy: 0.9516 Epoch 40/50 32/32 [==============================] - 1s 24ms/step - loss: 0.1725 - accuracy: 0.9429 - val_loss: 0.1522 - val_accuracy: 0.9533 Epoch 41/50 32/32 [==============================] - 1s 23ms/step - loss: 0.1502 - accuracy: 0.9468 - val_loss: 0.1502 - val_accuracy: 0.9566 Epoch 42/50 32/32 [==============================] - 1s 23ms/step - loss: 0.1620 - accuracy: 0.9409 - val_loss: 0.1744 - val_accuracy: 0.9504 Epoch 43/50 32/32 [==============================] - 1s 24ms/step - loss: 0.1784 - accuracy: 0.9443 - val_loss: 0.1562 - val_accuracy: 0.9549 Epoch 44/50 32/32 [==============================] - 1s 25ms/step - loss: 0.1676 - accuracy: 0.9399 - val_loss: 0.1541 - val_accuracy: 0.9542 Epoch 45/50 32/32 [==============================] - 1s 23ms/step - loss: 0.2119 - accuracy: 0.9351 - val_loss: 0.2007 - val_accuracy: 0.9441 Epoch 46/50 32/32 [==============================] - 1s 23ms/step - loss: 0.1873 - accuracy: 0.9395 - val_loss: 0.1683 - val_accuracy: 0.9515 Epoch 47/50 32/32 [==============================] - 1s 23ms/step - loss: 0.1826 - accuracy: 0.9409 - val_loss: 0.1621 - val_accuracy: 0.9519 Epoch 48/50 32/32 [==============================] - 1s 23ms/step - loss: 0.1718 - accuracy: 0.9458 - val_loss: 0.1673 - val_accuracy: 0.9515 Epoch 49/50 32/32 [==============================] - 1s 23ms/step - loss: 0.1484 - accuracy: 0.9521 - val_loss: 0.1644 - val_accuracy: 0.9520 Epoch 50/50 32/32 [==============================] - 1s 24ms/step - loss: 0.1538 - accuracy: 0.9473 - val_loss: 0.1562 - val_accuracy: 0.9534
Conclusion:
In this example the augmented model converges to an accuracy ~95% on validation set. This is slightly higher (+1%) than the model trained without data augmentation.
plotter = tfdocs.plots.HistoryPlotter()
plotter.plot({"Augmented": aug_history, "Non-Augmented": no_aug_history}, metric = "accuracy")
plt.title("Accuracy")
plt.ylim([0.75,1])
(0.75, 1.0)
In terms of loss, the non-augmented model is obviously in the overfitting regime. The augmented model, while a few epoch slower, is still training correctly and clearly not overfitting.
plotter = tfdocs.plots.HistoryPlotter()
plotter.plot({"Augmented": aug_history, "Non-Augmented": no_aug_history}, metric = "loss")
plt.title("Loss")
plt.ylim([0,1])
(0.0, 1.0)
