pdfcoffee

Chapter 5
IMG_SIZE = 160 # All images will be resized to 160x160
def format_example(image, label):
image = tf.cast(image, tf.float32)
image = (image/127.5) - 1
image = tf.image.resize(image, (IMG_SIZE, IMG_SIZE))
return image, label
train = raw_train.map(format_example)
validation = raw_validation.map(format_example)
test = raw_test.map(format_example)
Then, we shuffle and batch the training set, and batch the validation set and test set:
BATCH_SIZE = 32
SHUFFLE_BUFFER_SIZE = 2000
train_batches = train.shuffle(SHUFFLE_BUFFER_SIZE).batch(BATCH_SIZE)
validation_batches = validation.batch(BATCH_SIZE)
test_batches = test.batch(BATCH_SIZE)
Now, we can use MobileNet with input (160, 160, 3) where 3 is the number of color
channels. The top layers are omitted (include_top=False) since we are going to use
our own top layer, and all the internal layers are frozen because we use the weights
pretrained on ImageNet:
IMG_SHAPE = (IMG_SIZE, IMG_SIZE, 3)
base_model = tf.keras.applications.MobileNetV2(input_shape=IMG_SHAPE,
include_top=False,
weights='imagenet')
base_model.trainable = False
base_model.summary()
Let's inspect a batch and see if the shape is correct:
for image_batch, label_batch in train_batches.take(1):
pass
print (image_batch.shape)
(32, 160, 160, 3)
(32, 160, 160, 3) is the correct shape!
MobileNetV2 transforms each 160×160×3 image into a 5×5×1280 block of features.
For instance, let's see the transformation applied to the batch:
feature_batch = base_model(image_batch)
print(feature_batch.shape)
(32, 5, 5, 1280)
[ 155 ]