[Tensorflow] Pix2Pix

이번에 포스팅할 모델은 Pix2Pix입니다.

Pix2Pix는 그림의 도면이 주어지면, 이를 토대로 원본 이미지를 생성하는 모델입니다.

위의 Input 이미지로부터 Ground Truth 이미지를 생성할 수 있는 목적을 가지는 것입니다.

​

1. 사용할 패키지 불러오기.

 

import tensorflow as tf

import os
import time

from matplotlib import pyplot as plt
from IPython import display

 

import를 활용하여 사용할 패키지를 불어왔습니다.

​

2. 데이터 불러오기.

 

_URL = 'https://people.eecs.berkeley.edu/~tinghuiz/projects/pix2pix/datasets/facades.tar.gz'

path_to_zip = tf.keras.utils.get_file('facades.tar.gz',
                                      origin=_URL,
                                      extract=True)

PATH = os.path.join(os.path.dirname(path_to_zip), 'facades/')

 

Pix2Pix 데이터는 아래의 링크에서 받으 실 수 있으며, 위의 코드로 파이썬 환경에 불러옵니다.

 

https://people.eecs.berkeley.edu/~tinghuiz/projects/pix2pix/datasets/

Index of /~tinghuiz/projects/pix2pix/datasets

 

BUFFER_SIZE = 400
BATCH_SIZE = 1
IMG_WIDTH = 256
IMG_HEIGHT = 256

 

이미지 데이터셋을 만들기 위한 하이퍼파라미터를 정의하였습니다.

 

def load(image_file):
  image = tf.io.read_file(image_file)
  image = tf.image.decode_jpeg(image)

  w = tf.shape(image)[1]

  w = w // 2
  real_image = image[:, :w, :]
  input_image = image[:, w:, :]

  input_image = tf.cast(input_image, tf.float32)
  real_image = tf.cast(real_image, tf.float32)

  return input_image, real_image

inp, re = load(PATH+'train/100.jpg')
# casting to int for matplotlib to show the image
plt.figure()
plt.imshow(inp/255.0)
plt.figure()
plt.imshow(re/255.0)

 

실제로 파이썬에 불러와진 이미지의 예시를 살펴본 결과가 위와 같습니다.

이미지의 도면이 주어지고, 그에 대한 정답인 Ground Truth가 저장되어 있습니다.

 

def resize(input_image, real_image, height, width):
  input_image = tf.image.resize(input_image, [height, width],
                                method=tf.image.ResizeMethod.NEAREST_NEIGHBOR)
  real_image = tf.image.resize(real_image, [height, width],
                               method=tf.image.ResizeMethod.NEAREST_NEIGHBOR)

  return input_image, real_image

def random_crop(input_image, real_image):
  stacked_image = tf.stack([input_image, real_image], axis=0)
  cropped_image = tf.image.random_crop(
      stacked_image, size=[2, IMG_HEIGHT, IMG_WIDTH, 3])

  return cropped_image[0], cropped_image[1]

# normalizing the images to [-1, 1]

def normalize(input_image, real_image):
  input_image = (input_image / 127.5) - 1
  real_image = (real_image / 127.5) - 1

  return input_image, real_image

@tf.function()
def random_jitter(input_image, real_image):
  # resizing to 286 x 286 x 3
  input_image, real_image = resize(input_image, real_image, 286, 286)

  # randomly cropping to 256 x 256 x 3
  input_image, real_image = random_crop(input_image, real_image)

  if tf.random.uniform(()) > 0.5:
    # random mirroring
    input_image = tf.image.flip_left_right(input_image)
    real_image = tf.image.flip_left_right(real_image)

  return input_image, real_image

 

그리고 이미지에 대한 전처리로, Resize, Cropping, Normalize를 수행해줍니다.

마지막으로 이미지를 랜덤 확률에 따라 좌우반전을 하여 데이터의 양을 늘려줄 수 있습니다.

 

def load_image_train(image_file):
  input_image, real_image = load(image_file)
  input_image, real_image = random_jitter(input_image, real_image)
  input_image, real_image = normalize(input_image, real_image)

  return input_image, real_image

train_dataset = tf.data.Dataset.list_files(PATH+'train/*.jpg')
train_dataset = train_dataset.map(load_image_train,
                                  num_parallel_calls=tf.data.experimental.AUTOTUNE)
train_dataset = train_dataset.shuffle(BUFFER_SIZE)
train_dataset = train_dataset.batch(BATCH_SIZE)

def load_image_test(image_file):
  input_image, real_image = load(image_file)
  input_image, real_image = resize(input_image, real_image,
                                   IMG_HEIGHT, IMG_WIDTH)
  input_image, real_image = normalize(input_image, real_image)

  return input_image, real_image

test_dataset = tf.data.Dataset.list_files(PATH+'test/*.jpg')
test_dataset = test_dataset.map(load_image_test)
test_dataset = test_dataset.batch(BATCH_SIZE)

 

Train 이미지셋과 Test 이미지셋을 만들어줍니다.

​

3. Pix2Pix 모델 정의

​

(1) 생성자

​

Pix2Pix의 생성자는 Adjusted U-Net의 구조를 가지고 있습니다.

생성자의 Encoder 내부의 하나의 Block은 (Conv > Batch Norm > Leaky ReLU)의 과정을 가집니다.

반면에 Decoder 내부의 하나의 Block은 (Transposed Conv > Batch Norm > Dropout > ReLU)의 과정을 가집니다.

마지막으로 Encoder와 Decoder 사이에는 Skip Connection이 존재합니다.

 

OUTPUT_CHANNELS = 3

def downsample(filters, size, apply_batchnorm=True):
  initializer = tf.random_normal_initializer(0., 0.02)

  result = tf.keras.Sequential()
  result.add(
      tf.keras.layers.Conv2D(filters, size, strides=2, padding='same',
                             kernel_initializer=initializer, use_bias=False))

  if apply_batchnorm:
    result.add(tf.keras.layers.BatchNormalization())

  result.add(tf.keras.layers.LeakyReLU())

  return result

def upsample(filters, size, apply_dropout=False):
  initializer = tf.random_normal_initializer(0., 0.02)

  result = tf.keras.Sequential()
  result.add(
    tf.keras.layers.Conv2DTranspose(filters, size, strides=2,
                                    padding='same',
                                    kernel_initializer=initializer,
                                    use_bias=False))

  result.add(tf.keras.layers.BatchNormalization())

  if apply_dropout:
      result.add(tf.keras.layers.Dropout(0.5))

  result.add(tf.keras.layers.ReLU())

  return result

def Generator():
  inputs = tf.keras.layers.Input(shape=[256,256,3])

  down_stack = [
    downsample(64, 4, apply_batchnorm=False), # (bs, 128, 128, 64)
    downsample(128, 4), # (bs, 64, 64, 128)
    downsample(256, 4), # (bs, 32, 32, 256)
    downsample(512, 4), # (bs, 16, 16, 512)
    downsample(512, 4), # (bs, 8, 8, 512)
    downsample(512, 4), # (bs, 4, 4, 512)
    downsample(512, 4), # (bs, 2, 2, 512)
    downsample(512, 4), # (bs, 1, 1, 512)
  ]

  up_stack = [
    upsample(512, 4, apply_dropout=True), # (bs, 2, 2, 1024)
    upsample(512, 4, apply_dropout=True), # (bs, 4, 4, 1024)
    upsample(512, 4, apply_dropout=True), # (bs, 8, 8, 1024)
    upsample(512, 4), # (bs, 16, 16, 1024)
    upsample(256, 4), # (bs, 32, 32, 512)
    upsample(128, 4), # (bs, 64, 64, 256)
    upsample(64, 4), # (bs, 128, 128, 128)
  ]

  initializer = tf.random_normal_initializer(0., 0.02)
  last = tf.keras.layers.Conv2DTranspose(OUTPUT_CHANNELS, 4,
                                         strides=2,
                                         padding='same',
                                         kernel_initializer=initializer,
                                         activation='tanh') # (bs, 256, 256, 3)

  x = inputs

  # Downsampling through the model
  skips = []
  for down in down_stack:
    x = down(x)
    skips.append(x)

  skips = reversed(skips[:-1])

  # Upsampling and establishing the skip connections
  for up, skip in zip(up_stack, skips):
    x = up(x)
    x = tf.keras.layers.Concatenate()([x, skip])

  x = last(x)

  return tf.keras.Model(inputs=inputs, outputs=x)

generator = Generator()
tf.keras.utils.plot_model(generator, show_shapes=True, dpi=64)

다음으로 생성자의 Loss를 구현하겠습니다.

생성자의 Loss는 GAN Loss 와 L1 Loss로 구성됩니다.

 

LAMBDA = 100

def generator_loss(disc_generated_output, gen_output, target):
  gan_loss = loss_object(tf.ones_like(disc_generated_output), disc_generated_output)

  # mean absolute error
  l1_loss = tf.reduce_mean(tf.abs(target - gen_output))

  total_gen_loss = gan_loss + (LAMBDA * l1_loss)

  return total_gen_loss, gan_loss, l1_loss

 

(2) 판별자

​

Pix2Pix의 판별자는 PatchGAN의 구조를 가지고 있습니다.

 

def Discriminator():
  initializer = tf.random_normal_initializer(0., 0.02)

  inp = tf.keras.layers.Input(shape=[256, 256, 3], name='input_image')
  tar = tf.keras.layers.Input(shape=[256, 256, 3], name='target_image')

  x = tf.keras.layers.concatenate([inp, tar]) # (bs, 256, 256, channels*2)

  down1 = downsample(64, 4, False)(x) # (bs, 128, 128, 64)
  down2 = downsample(128, 4)(down1) # (bs, 64, 64, 128)
  down3 = downsample(256, 4)(down2) # (bs, 32, 32, 256)

  zero_pad1 = tf.keras.layers.ZeroPadding2D()(down3) # (bs, 34, 34, 256)
  conv = tf.keras.layers.Conv2D(512, 4, strides=1,
                                kernel_initializer=initializer,
                                use_bias=False)(zero_pad1) # (bs, 31, 31, 512)

  batchnorm1 = tf.keras.layers.BatchNormalization()(conv)

  leaky_relu = tf.keras.layers.LeakyReLU()(batchnorm1)

  zero_pad2 = tf.keras.layers.ZeroPadding2D()(leaky_relu) # (bs, 33, 33, 512)

  last = tf.keras.layers.Conv2D(1, 4, strides=1,
                                kernel_initializer=initializer)(zero_pad2) # (bs, 30, 30, 1)

  return tf.keras.Model(inputs=[inp, tar], outputs=last)

discriminator = Discriminator()
tf.keras.utils.plot_model(discriminator, show_shapes=True, dpi=64)

판별자의 손실 함수를 정의해보겠습니다.

 

loss_object = tf.keras.losses.BinaryCrossentropy(from_logits=True)

def discriminator_loss(disc_real_output, disc_generated_output):
  real_loss = loss_object(tf.ones_like(disc_real_output), disc_real_output)

  generated_loss = loss_object(tf.zeros_like(disc_generated_output), disc_generated_output)

  total_disc_loss = real_loss + generated_loss

  return total_disc_loss

 

4. 학습

​

모델 학습에 앞서 옵티마이저를 정의하겠습니다.

 

generator_optimizer = tf.keras.optimizers.Adam(2e-4, beta_1=0.5)
discriminator_optimizer = tf.keras.optimizers.Adam(2e-4, beta_1=0.5)

checkpoint_dir = './training_checkpoints'
checkpoint_prefix = os.path.join(checkpoint_dir, "ckpt")
checkpoint = tf.train.Checkpoint(generator_optimizer=generator_optimizer,
                                 discriminator_optimizer=discriminator_optimizer,
                                 generator=generator,
                                 discriminator=discriminator)

 

판별자와 생성자 모두 Adam을 사용하며, Learning Rate는 0.0002로 동일하게 설정하였습니다.

 

def generate_images(model, test_input, tar):
  prediction = model(test_input, training=True)
  plt.figure(figsize=(15,15))

  display_list = [test_input[0], tar[0], prediction[0]]
  title = ['Input Image', 'Ground Truth', 'Predicted Image']

  for i in range(3):
    plt.subplot(1, 3, i+1)
    plt.title(title[i])
    # getting the pixel values between [0, 1] to plot it.
    plt.imshow(display_list[i] * 0.5 + 0.5)
    plt.axis('off')
  plt.show()

 

이미지를 생성하여 시각화하는 함수를 정의하였습니다.

​

이제 Training Code를 구현하겠습니다.

 

EPOCHS = 150

import datetime
log_dir="logs/"

summary_writer = tf.summary.create_file_writer(
  log_dir + "fit/" + datetime.datetime.now().strftime("%Y%m%d-%H%M%S"))

@tf.function
def train_step(input_image, target, epoch):
  with tf.GradientTape() as gen_tape, tf.GradientTape() as disc_tape:
    gen_output = generator(input_image, training=True)

    disc_real_output = discriminator([input_image, target], training=True)
    disc_generated_output = discriminator([input_image, gen_output], training=True)

    gen_total_loss, gen_gan_loss, gen_l1_loss = generator_loss(disc_generated_output, gen_output, target)
    disc_loss = discriminator_loss(disc_real_output, disc_generated_output)

  generator_gradients = gen_tape.gradient(gen_total_loss,
                                          generator.trainable_variables)
  discriminator_gradients = disc_tape.gradient(disc_loss,
                                               discriminator.trainable_variables)

  generator_optimizer.apply_gradients(zip(generator_gradients,
                                          generator.trainable_variables))
  discriminator_optimizer.apply_gradients(zip(discriminator_gradients,
                                              discriminator.trainable_variables))

  with summary_writer.as_default():
    tf.summary.scalar('gen_total_loss', gen_total_loss, step=epoch)
    tf.summary.scalar('gen_gan_loss', gen_gan_loss, step=epoch)
    tf.summary.scalar('gen_l1_loss', gen_l1_loss, step=epoch)
    tf.summary.scalar('disc_loss', disc_loss, step=epoch)

def fit(train_ds, epochs, test_ds):
  for epoch in range(epochs):
    start = time.time()

    display.clear_output(wait=True)

    for example_input, example_target in test_ds.take(1):
      generate_images(generator, example_input, example_target)
    print("Epoch: ", epoch)

    # Train
    for n, (input_image, target) in train_ds.enumerate():
      print('.', end='')
      if (n+1) % 100 == 0:
        print()
      train_step(input_image, target, epoch)
    print()

    # saving (checkpoint) the model every 20 epochs
    if (epoch + 1) % 20 == 0:
      checkpoint.save(file_prefix = checkpoint_prefix)

    print ('Time taken for epoch {} is {} sec\n'.format(epoch + 1,
                                                        time.time()-start))
  checkpoint.save(file_prefix = checkpoint_prefix)

 

해당 코드의 Epoch는 150회이며, 20회마다 체크포인트를 저장하였습니다.

 

fit(train_dataset, EPOCHS, test_dataset)

 

위의 코드로 모델 훈련을 실행시킬 수 있습니다.