def _flow_on_training(self):
# get minibatch indices
with self.lock:
index_array, current_index, current_batch_size = next(
self.index_generator)
image_batch = np.array(self.x[index_array])
image_batch = normalize(image_batch, self.normalize_mode)
# label_batch = np.array(self.y[index_array])
return image_batch
def preprocessing(x, target_size=None):
image = x.convert('L')
if target_size is not None and target_size != image.size:
image = image.resize(target_size, Image.BILINEAR)
image_array = np.asarray(image)
image_array = normalize(image_array)
return image_array
def load_image(path, channel=None):
"""
:param paths: (channel, ) string list
:param target_size:
:return:
"""
image_array = np.asarray(Image.open(path))
image_array = normalize(image_array)
split_array = np.array(np.split(image_array, channel, axis=1))
image_array = split_array.transpose((1, 2, 0))
return image_array
def _flow_on_test(self):
# create indices
indexes = np.arange(self.n)
# calculate steps per a test
steps = self.n // self.batch_size
if self.n % self.batch_size != 0:
steps += 1
# yield loop
for i in range(steps):
index_array = indexes[i * self.batch_size:(i + 1) *
self.batch_size]
image_batch = self.x[index_array]
image_batch = normalize(image_batch, self.normalize_mode)
# label_batch = self.y[index_array]
yield image_batch