def get_img(self, img_path):
img = cv2.imread(img_path)
crop = img_ops.center_crop(
img, self.crop_size,
self.crop_size) if self.crop_size is not None else img
resized_img = cv2.resize(
crop, (self.out_size,
self.out_size)) if self.out_size is not None else crop
return img_ops.bound_image_values(resized_img)
def generate_train_mb(self):
train_size = self.lab_data.shape[0]
for i in xrange(0, train_size, self.batch_size):
lab_data_batch = self.lab_data[i:i + self.batch_size, :, :, :]
labels_batch = self.labels[i:i + self.batch_size]
lab_data_batch, labels_batch = self.fill_batch(
lab_data_batch, labels_batch, self.lab_data, self.labels)
yield bound_image_values(lab_data_batch).astype(
np.float32), self.one_hot_labels(labels_batch)
def generate_train_mb(self):
train_size = self.unlab_data.shape[0]
for i in xrange(0, train_size, self.batch_size):
data_batch = self.unlab_data[i:i + self.batch_size, :, :, :]
if data_batch.shape[0] < self.batch_size:
data_batch = np.concatenate(
(data_batch,
self.unlab_data[0:self.batch_size -
data_batch.shape[0], :, :, :]))
yield bound_image_values(data_batch).astype(np.float32)
def generate_test_mb(self):
for i in xrange(0, self.test_data.shape[0], self.batch_size):
data_batch = self.test_data[i:i + self.batch_size, :, :, :]
labels_batch = self.test_labels[i:i + self.batch_size]
data_batch, labels_batch = self.fill_batch(data_batch,
labels_batch,
self.test_data,
self.test_labels)
data_batch = bound_image_values(data_batch).astype(np.float32)
yield data_batch, self.one_hot_labels(labels_batch)
def generate_train_mb(self):
train_size = self.unlab_data.shape[0]
inds = self.rng.permutation(train_size)
unlab_data = self.unlab_data[inds, :, :, :]
lab_data, labels = self.extend_labeled_data(self.lab_data, self.labels,
unlab_data)
for i in xrange(0, train_size, self.batch_size):
lab_data_batch = lab_data[i:i + self.batch_size, :, :, :]
unlab_data_batch = unlab_data[i:i + self.batch_size, :, :, :]
labels_batch = labels[i:i + self.batch_size]
lab_data_batch, labels_batch = self.fill_batch(
lab_data_batch, labels_batch, lab_data, labels)
if unlab_data_batch.shape[0] < self.batch_size:
unlab_data_batch = np.concatenate(
(unlab_data_batch, unlab_data[0:self.batch_size -
unlab_data_batch.shape[0]]),
axis=0)
yield bound_image_values(lab_data_batch).astype(np.float32), \
bound_image_values(unlab_data_batch).astype(np.float32), self.one_hot_labels(labels_batch)
def transform(img_arr, scale_img_opt):
if scale_img_opt == '0_to_1':
return img_ops.bound_image_values_01(img_arr).astype(np.float32)
else:
return img_ops.bound_image_values(img_arr).astype(np.float32)