def __getitem__(self, i, type=None):
filename = self.files[i]
image = load_image(filename)
label = data_processor.binarize(self.annotations[i], self.num_classes)
label = np.reshape(label, (self.num_classes))
return (image, label)
def __getitem__(self, i, type=None):
files, labels = self.files_and_annotations
image = self._get_image(files[i])
label = data_processor.binarize(labels[i], self.num_classes)
label = np.reshape(label, (self.num_classes))
return (image, label)
def __getitem__(self, i, type=None):
target_file = self.files[i]
image = load_image(target_file)
label = self.get_label(target_file)
label = data_processor.binarize(label, self.num_classes)
label = np.reshape(label, (self.num_classes))
return (image, label)
def __getitem__(self, i, type=None):
filename = self.files[i]
image = PIL.Image.open(filename)
# sometime image data be gray.
image = image.convert("RGB")
image = np.array(image)
label = data_processor.binarize(self.annotations[i], self.num_classes)
label = np.reshape(label, (self.num_classes))
return (image, label)
def feed(self):
"""Returns numpy array of batch size data."""
images, labels = zip(*[self._element() for _ in range(self.batch_size)])
labels = data_processor.binarize(labels, self.num_classes)
images = np.array(images)
if self.data_format == 'NCHW':
images = np.transpose(images, [0, 3, 1, 2])
return images, labels
def feed(self):
"""Batch size numpy array of images and ground truth boxes.
Returns:
images: images numpy array. shape is [batch_size, height, width]
labels: labels numpy array. shape is [batch_size, num_classes]
"""
images, labels_list = zip(
*[self._element() for _ in range(self.batch_size)])
images = np.array(images)
labels_list = data_processor.binarize(labels_list, self.num_classes)
if self.data_format == "NCHW":
images = np.transpose(images, [0, 3, 1, 2])
return images, labels_list
def feed(self):
"""Returns numpy array of batch size data.
Returns:
images: images numpy array. shape is [batch_size, height, width]
labels: one hot labels. shape is [batch_size, num_classes]
"""
images, labels = zip(
*[self._element() for _ in range(self.batch_size)])
labels = data_processor.binarize(labels, self.num_classes)
images = np.array(images)
if self.data_format == "NCHW":
images = np.transpose(images, [0, 3, 1, 2])
return images, labels
def _init_features(self):
iterator = self.dataset.make_one_shot_iterator()
next_element = iterator.get_next()
self.features = []
with tf.Session() as sess:
for _ in range(self.num_per_epoch):
element = sess.run(next_element)
image = element["image"]
label = element["label"]
# Converting grayscale images into RGB images.
# This workaround is needed because the method PIL.Image.fromarray()
# in pre_prpcessor requires images array to have a shape of (h, w, 3).
if image.shape[2] == 1:
image = np.stack([image] * 3, 3)
image = image.reshape(image.shape[:2] + (3,))
label = data_processor.binarize(label, self.num_classes)
label = np.reshape(label, (self.num_classes))
self.features.append((image, label))
def __getitem__(self, i, type=None):
image = self._get_image(i)
label = data_processor.binarize(self.labels[i], self.num_classes)
label = np.reshape(label, (self.num_classes))
return (image, label)
