def run(self, debug=False):
batch_size = self.initial_batch_size
for _ in range(self.number_of_interation):
self.run_iteration(batch_size=batch_size, epochs=self.epochs, debug=debug)
batch_size *= 2
if self.all_image_buffers:
make_animation(self.all_image_buffers, '/opt/host/Downloads/quadGAN')
def show_sample(path, interval=200):
original, img = get_image(path)
patch_list = []
for line in get_lines(img):
patch_list.append((line, 'blue', ''))
for char in get_chars(img, line):
patch_list.append((char, 'green', ''))
make_animation(original, patch_list, interval)
def show_sample(path, interval=200):
original, img = get_image(path)
patch_list = []
for line in get_lines(img):
patch_list.append((line, 'blue', ''))
for char in get_chars(img, line):
patch_list.append((char, 'green', ''))
make_animation(original, patch_list, interval)
def train(self, debug):
image_buffers = []
image_buffers_test = []
for ep in tqdm.trange(self.epochs, desc='Epoch Loop'):
if ep < self.ep_cnt:
continue
# update epoch counter
self.ep_cnt.assign_add(1)
# train for an epoch
for A, B in tqdm.tqdm(self.A_B_dataset,
desc='Inner Epoch Loop',
total=self.len_dataset):
G_loss_dict, D_loss_dict = self.train_step(A, B)
# # summary
tl.summary(G_loss_dict,
step=self.G_optimizer.iterations,
name='G_losses')
tl.summary(D_loss_dict,
step=self.G_optimizer.iterations,
name='D_losses')
tl.summary(
{
'learning rate':
self.G_lr_scheduler.current_learning_rate
},
step=self.G_optimizer.iterations,
name='learning rate')
# sample
snapshot_period = min(10, (self.epochs // 70) + 1)
if self.G_optimizer.iterations.numpy() % snapshot_period == 0:
image_buffers.append(
self.snapshot(A, B, 'train_iter-%09d.jpg',
debug=debug))
A_test, B_test = next(self.test_iter)
image_buffers_test.append(
self.snapshot(A_test,
B_test,
'test_iter-%09d.jpg',
debug=debug))
# save checkpoint
self.checkpoint.save(ep)
if image_buffers:
image_buffers.extend(image_buffers_test)
make_animation(image_buffers, 'animations/cycleGAN')
def process_image():
numbers = list(map(int, open('numbers.txt').read().split(' ')))
o, _ = get_image('numbers.png')
slices = [(s, 'green', n) for ((_, s), n) in zip(get_digits_fake(o), numbers)]
ani = make_animation(o, slices, 300)
# plt.show()
return ani
def process_image():
numbers = list(map(int, open('numbers.txt').read().split(' ')))
o, _ = get_image('numbers.png')
slices = [(s, 'green', n)
for ((_, s), n) in zip(get_digits_fake(o), numbers)]
ani = make_animation(o, slices, 300)
# plt.show()
return ani
from digits_classifiers import get_predictor
from image_processing import get_digits, get_image
from utils import make_animation
import matplotlib.pylab as plt
predictor = get_predictor('Logistic regression')
original, img = get_image('numbers.png')
predictions = []
for digit, slice_ in get_digits(img):
p = predictor.predict(digit)[0]
predictions.append((slice_, 'green', p))
ani = make_animation(original, predictions, interval=300)
plt.show()
