def display_examples(n_images):
"""
Display an example of MNIST & CelebA
:param n_images: number of images to display
"""
show_n_images = n_images
get_ipython().magic('matplotlib inline')
mnist_images = helper.get_batch(glob(os.path.join(DATA_DIR, 'mnist/*.jpg'))[:show_n_images], 28, 28, 'L')
plt.imshow(helper.images_square_grid(mnist_images, 'L'), cmap='gray')
plt.show()
mnist_images = helper.get_batch(glob(os.path.join(DATA_DIR, 'img_align_celeba/*.jpg'))[:show_n_images], 28, 28, 'RGB')
plt.imshow(helper.images_square_grid(mnist_images, 'RGB'))
plt.show()
def train(self, vectorized_songs):
for iter in tqdm(range(self.num_training_iterations)):
x_batch, y_batch = helper.get_batch(vectorized_songs,
self.seq_length,
self.batch_size)
loss = self.train_step(x_batch, y_batch)
self.history.append(loss.numpy().mean())
if iter % 100 == 0:
self.model.save_weights(self.checkpoint_prefix)
self.model.save_weights(self.checkpoint_prefix)
def load_image():
data_dir = './data'
# helper.download_extract(data_dir)
show_n_images = 10000
image_size = 56
plt.figure(figsize=(10, 10))
images = helper.get_batch(
glob(os.path.join(data_dir, 'img_align_celeba/*.jpg'))[:show_n_images],
image_size, image_size, 'RGB')
return images
def _run_train(sess, model, x_train, y_train, x_eval, y_eval, train_config):
for epoch in range(train_config.EPOCHS):
# train
for batch_x, batch_y in helper.get_batch(x_train,
y_train,
step=train_config.BATCH_SIZE):
model.train(sess,
batch_x,
batch_y,
dropout_keep_prob=train_config.DROPOUT_KEEP_PROB)
# eval
loss = model.evaluate_loss(sess, x_eval, y_eval, dropout_keep_prob=1.0)
_print_eval_results(loss, epoch)
def _run_train_with_summary(sess, model, x_train, y_train, x_eval, y_eval,
train_config):
saver = tf.train.Saver()
helper.mkdir(train_config.SUMMARY_DIR)
summary_writer = tf.summary.FileWriter(logdir=train_config.SUMMARY_DIR)
summary_writer.add_graph(sess.graph)
for epoch in range(train_config.EPOCHS):
# train
for batch_x, batch_y in helper.get_batch(x_train,
y_train,
step=train_config.BATCH_SIZE):
summary, _ = model.train(
sess,
batch_x,
batch_y,
dropout_keep_prob=train_config.DROPOUT_KEEP_PROB)
summary_writer.add_summary(summary)
# eval
summary, loss = model.evaluate_loss(sess,
batch_x,
batch_y,
dropout_keep_prob=1.0)
summary_writer.add_summary(summary)
_print_eval_results(loss, epoch)
def run(): train_map, dev_map, test_map = get_datamap(composers=composers, split=True) train_size = get_size(train_map) dev_size = get_size(dev_map) print(train_size, dev_size) # Create Train and Dev generators train_gen = randomized_generator(train_map, batch_size=batch_size, timesteps=timesteps, n_classes=n_classes) dev_gen = randomized_generator(dev_map, batch_size=batch_size, timesteps=timesteps, n_classes=n_classes) train_iterations = int(train_size/(10*batch_size)) dev_iterations = int(dev_size/(10*batch_size)) print(train_iterations, dev_iterations) # Train Model composer_clf = Composer_Classifier(n_classes=n_classes) composer_clf.create_model() composer_clf.train(train_generator=train_gen, dev_generator=dev_gen, steps_per_epoch=train_iterations, validation_steps=dev_iterations, epochs=n_epochs) # Test x_test, y_test = get_batch(test_map, batch_size=test_batch_size, timesteps=100) composer_clf.test(x_test, y_test)
# ### MNIST # [MNIST](http://yann.lecun.com/exdb/mnist/) 是一个手写数字的图像数据集。你可以更改 `show_n_images` 探索此数据集。 # In[90]: show_n_images = 25 """ DON'T MODIFY ANYTHING IN THIS CELL """ get_ipython().magic(u'matplotlib inline') import os from glob import glob from matplotlib import pyplot mnist_images = helper.get_batch(glob(os.path.join(data_dir, 'mnist/*.jpg'))[:show_n_images], 28, 28, 'L') pyplot.imshow(helper.images_square_grid(mnist_images, 'L'), cmap='gray') # ### CelebA # [CelebFaces Attributes Dataset (CelebA)](http://mmlab.ie.cuhk.edu.hk/projects/CelebA.html) 是一个包含 20 多万张名人图片及相关图片说明的数据集。你将用此数据集生成人脸，不会用不到相关说明。你可以更改 `show_n_images` 探索此数据集。 # In[91]: show_n_images = 25 """ DON'T MODIFY ANYTHING IN THIS CELL """ mnist_images = helper.get_batch(glob(os.path.join(data_dir, 'img_align_celeba/*.jpg'))[:show_n_images], 28, 28, 'RGB') pyplot.imshow(helper.images_square_grid(mnist_images, 'RGB'))
import helper
import os
from glob import glob
import numpy as np
from matplotlib import pyplot
from distutils.version import LooseVersion
import warnings
import tensorflow as tf
import problem_unittests as tests
show_n_images = 40
celeb_images = helper.get_batch(
glob(os.path.join('img_align_celeba/*.jpg'))[:show_n_images], 50, 50,
'RGB')
pyplot.imshow(helper.images_square_grid(celeb_images, 'RGB'))
# pyplot.show()
# Check TensorFlow Version
assert LooseVersion(tf.__version__) >= LooseVersion(
'1.0'
), 'Please use TensorFlow version 1.0 or newer. You are using {}'.format(
tf.__version__)
print('TensorFlow Version: {}'.format(tf.__version__))
if not tf.test.gpu_device_name():
warnings.warn(
'No GPU found. Please use a GPU to train your neural network.')
else:
print('Default GPU Device: {}'.format(tf.test.gpu_device_name()))
# In[47]:
def output_fig(images_array, file_name):
plt.figure(figsize=(6, 6), dpi=100)
plt.imshow(helper.images_square_grid(images_array))
plt.axis("off")
plt.savefig(file_name+'.png', bbox_inches='tight', pad_inches=0)
# In[50]:
for i in range(500):
width = height = 64
generated_images = helper.get_batch(glob('fake_images/*.png')[0+(i*9):9+(i*9)], 64, 64, 'RGB')
output_fig(generated_images, file_name="./fake_images_9x9/img%03d.png"%i)
del generated_images
# In[48]:
output_fig(generated_images)
# In[ ]:
"./fake_images_9x9/img%03d.png"%
train_loss_hist = []
val_loss_hist = []
lr_hist = []
with tf.device("/gpu:0"):
with tf.Session() as sess:
train_writer = tf.summary.FileWriter(
'../../outputs/summary/train_2' + cost_type, sess.graph)
val_writer = tf.summary.FileWriter('../../outputs/summary/test_2' +
cost_type)
sess.run(tf.global_variables_initializer())
init_fn(sess)
if data_bias == "balancing":
iterations = iterations_number
for step in range(iterations):
batch = hp.get_batch(cluster_mini_batches, mini_size,
clusters)
images, labels = hp.load_images_labels2(batch, train, path)
sess.run(
optimiser, {
md.input: images,
md.target: labels,
md.kp: kp,
md.lr: learning_rate
})
if step % print_step == 0:
train_loss, train_summary = sess.run(
[md.loss, merged], {
md.input: images,
md.target: labels,
md.kp: 1.0
dacr_sum = np.zeros(2)
dacf_sum = np.zeros(2)
gac_sum = np.zeros(2)
for epch in range(maxepch):
print('Iteration No.', epch)
for bcnt in range(0, dir.shape[0], batch_size):
#random.shuffle(dir)
n_critic = 1
for ct in range(n_critic):
# Label smoothing
#label_r = ones + (np.random.sample([batch_size,1]) * 0.01)
#label_f = zeros + (np.random.sample([batch_size, 1]) * 0.01)
select = np.random.randint(0, dir.shape[0], batch_size)
real_image = (hp.get_batch(dir[select], img_size, img_size, 'RGB') - 127.5) / 127.5
nd_noise = np.random.normal(0, 1, (batch_size, noise_size))
# nd_noise = np.random.uniform(-1, 1, (batch_size, noise_size))
fake_img = G.predict(nd_noise)
dacr_sum += D.train_on_batch(real_image, mones)
dacf_sum += D.train_on_batch(fake_img, ones)
# Clip discriminator weights
for l in D.layers:
weights = l.get_weights()
weights = [np.clip(w, -0.05, 0.05) for w in weights]
l.set_weights(weights)
# Train generator
nd_noise = np.random.normal(0,1,(batch_size, noise_size))
# nd_noise = np.random.uniform(-1, 1, (batch_size, noise_size))
image = Image.open(image_path)
return np.array(image.convert(mode))
def get_batch(image_files, width, height, mode):
data_batch = np.array(
[get_image(sample_file, width, height, mode) for sample_file in image_files]).astype(np.float32)
# Make sure the images are in 4 dimensions
if len(data_batch.shape) < 4:
data_batch = data_batch.reshape(data_batch.shape + (1,))
return data_batch
show_n_images = 25
mnist_images = helper.get_batch(glob(os.path.join(data_resized_dir, '*.jpg'))[:show_n_images], 64, 64, 'RGB')
plt.imshow(helper.images_square_grid(mnist_images, 'RGB'))
# Taken from Udacity face generator project
from distutils.version import LooseVersion
import warnings
# Check TensorFlow Version
assert LooseVersion(tf.__version__) >= LooseVersion('1.0'), 'Please use TensorFlow version 1.0 or newer. You are using {}'.format(tf.__version__)
print('TensorFlow Version: {}'.format(tf.__version__))
# -*- coding: utf-8 -*-
"""
Created on Thu Nov 7 22:39:07 2019
@author: Keke
"""
import os
import helper
import matplotlib.pyplot as plt
from glob import glob
data_dir = '../../cs-ioc5008-hw2/pytorch_GAN_zoo/output_datasets'
file_dir = '../../cs-ioc5008-hw2/images/'
def output_fig(images_array, file_name="./results"):
plt.figure(figsize=(6, 6), dpi=100)
plt.imshow(helper.images_square_grid(images_array))
plt.axis("off")
plt.savefig(file_name + '.png', bbox_inches='tight', pad_inches=0)
for i in range(5):
generated_images = helper.get_batch(
glob(os.path.join(data_dir, '*.png'))[i * 9:(i + 1) * 9], 32, 32,
'RGB')
print(generated_images.shape)
output_fig(generated_images,
file_name=file_dir + "{}_image".format(str.zfill(str(i), 3)))
tests.test_discriminator(discriminator, tf)
tests.test_generator(generator, tf)
tests.test_model_loss(model_loss)
tests.test_model_opt(model_opt, tf)
# 학습 Parameters
show_n_images = 25
batch_size = 64
z_dim = 100
learning_rate = 0.00025
beta1 = 0.45
epochs = 100 # 시간/장비의 제약으로 인해 11번째 epoch 까지만 진행
# celeba.zip 데이터셋 다운로드
helper.download_extract('celeba', './celeba')
mnist_images = helper.get_batch(glob(os.path.join('./celeba', 'img_align_celeba/*.jpg'))[:show_n_images], 28, 28, 'RGB')
pyplot.imshow(helper.images_square_grid(mnist_images, 'RGB')) # 샘플 데이터 1개 보여주기
# 다운로드 받은 celeba 이미지들을 학습시킬 수 있는 데이터셋으로 변환
celeba_dataset = helper.Dataset('celeba', glob(os.path.join('./celeba', 'img_align_celeba/*.jpg')))
# GPU 사용 여부 확인
if not tf.test.gpu_device_name():
warnings.warn('GPU가 없습니다. - GPU 사용 권장')
else:
print('사용중인 GPU: {}'.format(tf.test.gpu_device_name()))
# 실제 학습시키고 output을 생성하는 부분 (시간이 오래 걸립니다)
with tf.Graph().as_default():
train(epochs, batch_size, z_dim, learning_rate, beta1, celeba_dataset.get_batches, celeba_dataset.shape, celeba_dataset.image_mode)
def show_images(ids):
show_n_images = 16
celeb_images = helper.get_batch(ids[:show_n_images], 28, 28, 'RGB')
plt.imshow(helper.images_square_grid(celeb_images, 'RGB'))
"""
import helper
helper.download_extract('mnist', data_dir)
helper.download_extract('celeba', data_dir)
show_n_images = 25
"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
import os
from glob import glob
from matplotlib import pyplot
mnist_images = helper.get_batch(glob(os.path.join(data_dir, 'mnist/*.jpg'))[:show_n_images], 28, 28, 'L')
pyplot.imshow(helper.images_square_grid(mnist_images, 'L'), cmap='gray')
import problem_unittests as tests
"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
from distutils.version import LooseVersion
import warnings
import tensorflow as tf
# Check TensorFlow Version
assert LooseVersion(tf.__version__) >= LooseVersion(
'1.0'), 'Please use TensorFlow version 1.0 or newer. You are using {}'.format(tf.__version__)
print('TensorFlow Version: {}'.format(tf.__version__))
# ## 探索数据（Explore the Data）
# ### MNIST
# [MNIST](http://yann.lecun.com/exdb/mnist/) 是一个手写数字的图像数据集。你可以更改 `show_n_images` 探索此数据集。
# In[2]:
show_n_images = 25
"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
get_ipython().run_line_magic('matplotlib', 'inline')
import os
from glob import glob
from matplotlib import pyplot
mnist_images = helper.get_batch(
glob(os.path.join(data_dir, 'mnist/*.jpg'))[:show_n_images], 28, 28, 'L')
pyplot.imshow(helper.images_square_grid(mnist_images, 'L'), cmap='gray')
# ### CelebA
# [CelebFaces Attributes Dataset (CelebA)](http://mmlab.ie.cuhk.edu.hk/projects/CelebA.html) 是一个包含 20 多万张名人图片及相关图片说明的数据集。你将用此数据集生成人脸，不会用不到相关说明。你可以更改 `show_n_images` 探索此数据集。
# In[3]:
show_n_images = 25
"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
mnist_images = helper.get_batch(
glob(os.path.join(data_dir, 'img_align_celeba/*.jpg'))[:show_n_images], 28,
28, 'RGB')
pyplot.imshow(helper.images_square_grid(mnist_images, 'RGB'))
# In[1]:
show_n_images = 25
import helper
import numpy as np
"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
get_ipython().run_line_magic('matplotlib', 'inline')
import os
from glob import glob
from matplotlib import pyplot
mnist_images = helper.get_batch(
glob(os.path.join('F:/uav+ai/data_for_neural_network/',
'mnist/*.jpg'))[:show_n_images], 28, 28, 'L')
print(np.shape(mnist_images))
pyplot.imshow(helper.images_square_grid(mnist_images, 'L'), cmap='gray')
# ### CelebA
# [CelebFaces Attributes Dataset (CelebA)](http://mmlab.ie.cuhk.edu.hk/projects/CelebA.html) 是一个包含 20 多万张名人图片及相关图片说明的数据集。你将用此数据集生成人脸，不会用不到相关说明。你可以更改 `show_n_images` 探索此数据集。
# In[2]:
show_n_images = 25
"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
mnist_images = helper.get_batch(
glob(
