def load_examples():
if a.input_dir is None or not os.path.exists(a.input_dir):
raise Exception("input_dir does not exist")
# synchronize seed for image operations so that we do the same operations to both
# input and output images
# seed = random.randint(0, 2**31 - 1)
decode = tf.image.decode_png
### Read inputs
input_paths = sorted(glob.glob(os.path.join(a.input_dir, "*.png")))
# print(input_paths)
with tf.name_scope("load_input_images"):
path_queue = tf.train.string_input_producer(input_paths,
shuffle=a.mode == "train")
reader = tf.WholeFileReader()
paths, contents = reader.read(path_queue)
raw_input = decode(contents, channels=1)
raw_input = tf.image.convert_image_dtype(raw_input, dtype=tf.float32)
# assertion = tf.assert_equal(tf.shape(raw_input)[2], 3, message="image does not have 3 channels")
# with tf.control_dependencies([assertion]):
raw_input = tf.identity(raw_input)
raw_input.set_shape([350, 256, 1])
input_images = preprocess(raw_input)
def get_name(path):
name, _ = os.path.splitext(os.path.basename(path))
return name
input_dir = a.input_dir
if len(input_paths) == 0:
raise Exception("input_dir contains no image files")
### Read targets
target_dir = a.target_dir
target_paths = sorted(glob.glob(os.path.join(target_dir, "*.png")))
if len(target_paths) == 0:
raise Exception("input_dir contains no image files")
with tf.name_scope("load_target_images"):
targets_path_queue = tf.train.string_input_producer(
target_paths, shuffle=a.mode == "train")
targets_reader = tf.WholeFileReader()
targets_paths, targets_contents = targets_reader.read(
targets_path_queue)
raw_input = decode(targets_contents, channels=1)
raw_input = tf.image.convert_image_dtype(raw_input, dtype=tf.float32)
# assertion = tf.assert_equal(tf.shape(raw_input)[2], 3, message="image does not have 3 channels")
# with tf.control_dependencies([assertion]):
raw_input = tf.identity(raw_input)
raw_input.set_shape([29, 21, 1])
target_images = preprocess(raw_input)
input_paths_batch, inputs_batch, targets_batch = tf.train.batch(
[paths, input_images, target_images], batch_size=a.batch_size)
steps_per_epoch = int(math.ceil(len(input_paths) / a.batch_size))
return Examples(input_paths=input_paths_batch,
inputs=inputs_batch,
targets=targets_batch,
count=len(input_paths),
steps_per_epoch=steps_per_epoch)
def main(unused_argv):
os.environ["CUDA_VISIBLE_DEVICES"] = cmd_args.cuda_device
tf.logging.set_verbosity(tf.logging.INFO)
# Read list of images.
tf.logging.info('Reading list of images...')
image_paths = _ReadImageList(cmd_args.curr_idx)
num_images = len(image_paths)
tf.logging.info('done! Found %d images', num_images)
# Parse DelfConfig proto.
config = delf_config_pb2.DelfConfig()
with tf.gfile.FastGFile(cmd_args.config_path, 'r') as f:
text_format.Merge(f.read(), config)
# Create output directory if necessary.
if not os.path.exists(cmd_args.output_dir):
os.makedirs(cmd_args.output_dir, exist_ok=True)
# Tell TensorFlow that the model will be built into the default Graph.
with tf.Graph().as_default():
# Reading list of images.
filename_queue = tf.train.string_input_producer(image_paths,
shuffle=False)
reader = tf.WholeFileReader()
_, value = reader.read(filename_queue)
image_tf = tf.image.decode_image(value, channels=3)
count = 0
with tf.Session(config=sess_config) as sess:
init_op = tf.global_variables_initializer()
sess.run(init_op)
extractor_fn = MakeExtractor(sess, config)
# Start input enqueue threads.
coord = tf.train.Coordinator()
threads = tf.train.start_queue_runners(sess=sess, coord=coord)
start = time.clock()
for i in range(num_images):
# Write to log-info once in a while.
if i == 0:
tf.logging.info(
'Starting to extract DELF features from images...')
elif i % _STATUS_CHECK_ITERATIONS == 0:
elapsed = (time.clock() - start)
tf.logging.info(
'Processing image %d out of %d, last %d '
'images took %f seconds', i, num_images,
_STATUS_CHECK_ITERATIONS, elapsed)
start = time.clock()
try:
im = sess.run(image_tf)
except tf.errors.NotFoundError as e:
count += 1
tf.logging.warning(
f"{'*' * 15} Decode error {e} - {count}")
continue
if len(im.shape) == 4 or im.shape[-1] != 3:
tf.logging.warning(
f"{'*' * 15} Skipping image {i} which probably is a GIF or broken image"
)
continue
# If descriptor already exists, skip its computation.
out_desc_filename = os.path.splitext(
os.path.basename(image_paths[i]))[0] + _DELF_EXT
out_desc_parent_dir = os.path.join(
cmd_args.output_dir, '/'.join(out_desc_filename[:3]))
if not os.path.exists(out_desc_parent_dir):
os.makedirs(out_desc_parent_dir)
out_desc_fullpath = os.path.join(out_desc_parent_dir,
out_desc_filename)
if tf.gfile.Exists(out_desc_fullpath):
tf.logging.info('Skipping %s', image_paths[i])
continue
# Extract and save features.
(locations_out, descriptors_out, feature_scales_out,
attention_out) = extractor_fn(im)
feature_io.WriteToFile(out_desc_fullpath, locations_out,
feature_scales_out, descriptors_out,
attention_out)
# Finalize enqueue threads.
coord.request_stop()
coord.join(threads)
def make_data_tensor(self, train=True):
if train:
folders = self.metatrain_character_folders
# number of tasks, not number of meta-iterations. (divide by metabatch size to measure)
num_total_batches = 200000
else:
folders = self.metaval_character_folders
num_total_batches = 600
# make list of files
print('Generating filenames')
if self.datasource == 'cifar':
if not train:
all_filenames = []
from datetime import datetime
start = datetime.now()
for i in range(num_total_batches):
if (i + 1) % 5000 == 0:
print('Generated {} tasks...'.format((i + 1)))
sampled_character_folders = random.sample(
folders, self.num_classes)
random.shuffle(sampled_character_folders)
if self.mode == 'pretrain':
sampled_character_folders = folders
labels_and_images = get_images(
sampled_character_folders,
range(self.num_classes),
nb_samples=self.num_samples_per_class,
shuffle=False)
# make sure the above isn't randomized order
labels = [li[0] for li in labels_and_images]
filenames = [li[1] for li in labels_and_images]
all_filenames.extend(filenames)
# temp
# save and reload pickle for fast iteration
else:
with open("cifar_filenames_25way1shot.pkl", 'rb') as file:
# pickle.dump(all_filenames, file)
all_filenames = pickle.load(file)
# quit()
labels = list(
np.concatenate(np.array(self.num_samples_per_class *
[list(range(self.num_classes))]).T,
axis=0))
elif self.datasource == 'miniimagenet':
if self.mode == 'pretrain':
import glob
all_filenames = glob.glob('data/miniImagenet/train/n*/*')
all_labels = list(
np.concatenate(np.array(600 * [list(range(64))]).T,
axis=0))
elif not train:
all_filenames = []
from datetime import datetime
start = datetime.now()
for i in range(num_total_batches):
if (i + 1) % 5000 == 0:
print('Generated {} tasks...'.format((i + 1)))
sampled_character_folders = random.sample(
folders, self.num_classes)
random.shuffle(sampled_character_folders)
if self.mode == 'pretrain':
sampled_character_folders = folders
labels_and_images = get_images(
sampled_character_folders,
range(self.num_classes),
nb_samples=self.num_samples_per_class,
shuffle=False)
# make sure the above isn't randomized order
labels = [li[0] for li in labels_and_images]
filenames = [li[1] for li in labels_and_images]
all_filenames.extend(filenames)
# temp
# save and reload pickle for fast iteration
else:
with open("miniimagenet_filenames_5way1shot.pkl",
'rb') as file:
# pickle.dump(all_filenames, file)
all_filenames = pickle.load(file)
# quit()
labels = list(
np.concatenate(np.array(self.num_samples_per_class *
[list(range(self.num_classes))]).T,
axis=0))
else:
all_filenames = []
from datetime import datetime
start = datetime.now()
for i in range(num_total_batches):
if (i + 1) % 5000 == 0:
print('Generated {} tasks...'.format((i + 1)))
sampled_character_folders = random.sample(
folders, self.num_classes)
random.shuffle(sampled_character_folders)
# temp
# sampled_character_folders = folders
labels_and_images = get_images(
sampled_character_folders,
range(self.num_classes),
nb_samples=self.num_samples_per_class,
shuffle=False)
# make sure the above isn't randomized order
labels = [li[0] for li in labels_and_images]
filenames = [li[1] for li in labels_and_images]
all_filenames.extend(filenames)
# make queue for tensorflow to read from
if self.mode == 'pretrain':
filename_queue = tf.train.string_input_producer(
tf.convert_to_tensor(all_filenames), shuffle=True, seed=42)
label_queue = tf.train.input_producer(
tf.convert_to_tensor(all_labels), shuffle=True, seed=42)
print('Generating image processing ops')
image_reader = tf.WholeFileReader()
_, image_file = image_reader.read(filename_queue)
# if self.datasource == 'miniimagenet':
image = tf.image.decode_jpeg(image_file, channels=3)
image.set_shape((self.img_size[0], self.img_size[1], 3))
image = tf.reshape(image, [self.dim_input])
image = tf.cast(image, tf.float32) / 255.0
num_preprocess_threads = 1
min_queue_examples = 256
images, labels = tf.train.batch(
[image, label_queue.dequeue()],
batch_size=self.batch_size,
num_threads=num_preprocess_threads,
capacity=min_queue_examples + 3 * self.batch_size,
)
labels = tf.one_hot(labels, self.num_classes)
return images, labels
filename_queue = tf.train.string_input_producer(
tf.convert_to_tensor(all_filenames), shuffle=False)
print('Generating image processing ops')
image_reader = tf.WholeFileReader()
_, image_file = image_reader.read(filename_queue)
if self.datasource == 'miniimagenet':
image = tf.image.decode_jpeg(image_file, channels=3)
image.set_shape((self.img_size[0], self.img_size[1], 3))
image = tf.reshape(image, [self.dim_input])
image = tf.cast(image, tf.float32) / 255.0
elif self.datasource == 'cifar':
image = tf.image.decode_png(image_file, channels=3)
image.set_shape((self.img_size[0], self.img_size[1], 3))
image = tf.reshape(image, [self.dim_input])
image = tf.cast(image, tf.float32) / 255.0
else:
image = tf.image.decode_png(image_file)
image.set_shape((self.img_size[0], self.img_size[1], 1))
image = tf.reshape(image, [self.dim_input])
image = tf.cast(image, tf.float32) / 255.0
image = 1.0 - image # invert
num_preprocess_threads = 1 # TODO - enable this to be set to >1
min_queue_examples = 256
examples_per_batch = self.num_classes * self.num_samples_per_class
batch_image_size = self.batch_size * examples_per_batch
print('Batching images')
images = tf.train.batch(
[image],
batch_size=batch_image_size,
num_threads=num_preprocess_threads,
capacity=min_queue_examples + 3 * batch_image_size,
)
all_image_batches, all_label_batches = [], []
print('Manipulating image data to be right shape')
for i in range(self.batch_size):
image_batch = images[i * examples_per_batch:(i + 1) *
examples_per_batch]
if self.datasource == 'omniglot':
# omniglot augments the dataset by rotating digits to create new classes
# get rotation per class (e.g. 0,1,2,0,0 if there are 5 classes)
rotations = tf.multinomial(tf.log([[1., 1., 1., 1.]]),
self.num_classes)
label_batch = tf.convert_to_tensor(labels)
new_list, new_label_list = [], []
for k in range(self.num_samples_per_class):
class_idxs = tf.range(0, self.num_classes)
class_idxs = tf.random_shuffle(class_idxs)
true_idxs = class_idxs * self.num_samples_per_class + k
new_list.append(tf.gather(image_batch, true_idxs))
if self.datasource == 'omniglot': # and FLAGS.train:
new_list[-1] = tf.stack([
tf.reshape(
tf.image.rot90(tf.reshape(
new_list[-1][ind],
[self.img_size[0], self.img_size[1], 1]),
k=tf.cast(
rotations[0, class_idxs[ind]],
tf.int32)), (self.dim_input, ))
for ind in range(self.num_classes)
])
new_label_list.append(tf.gather(label_batch, true_idxs))
new_list = tf.concat(
new_list, 0
) # has shape [self.num_classes*self.num_samples_per_class, self.dim_input]
new_label_list = tf.concat(new_label_list, 0)
all_image_batches.append(new_list)
all_label_batches.append(new_label_list)
all_image_batches = tf.stack(all_image_batches)
all_label_batches = tf.stack(all_label_batches)
all_label_batches = tf.one_hot(all_label_batches, self.num_classes)
return all_image_batches, all_label_batches
def load_examples():
if a.input_dir is None or not os.path.exists(a.input_dir):
raise Exception("input_dir does not exist")
input_paths = glob.glob(os.path.join(a.input_dir, "*.jpg"))
decode = tf.image.decode_jpeg
if len(input_paths) == 0:
input_paths = glob.glob(os.path.join(a.input_dir, "*.png"))
decode = tf.image.decode_png
if len(input_paths) == 0:
raise Exception("input_dir contains no image files")
def get_name(path):
name, _ = os.path.splitext(os.path.basename(path))
return name
# if the image names are numbers, sort by the value rather than asciibetically
# having sorted inputs means that the outputs are sorted in test mode
if all(get_name(path).isdigit() for path in input_paths):
input_paths = sorted(input_paths, key=lambda path: int(get_name(path)))
else:
input_paths = sorted(input_paths)
with tf.name_scope("load_images"):
path_queue = tf.train.string_input_producer(input_paths, shuffle=False)
reader = tf.WholeFileReader()
paths, contents = reader.read(path_queue)
raw_input = decode(contents)
raw_input = tf.image.convert_image_dtype(raw_input, dtype=tf.float32)
assertion = tf.assert_equal(tf.shape(raw_input)[2], 3, message="image does not have 3 channels")
with tf.control_dependencies([assertion]):
raw_input = tf.identity(raw_input)
raw_input.set_shape([None, None, 3])
# break apart image pair and move to range [-1, 1]
width = tf.shape(raw_input)[1] # [height, width, channels]
images = preprocess(raw_input[:,:,:])
inputs=images
# synchronize seed for image operations so that we do the same operations to both
# input and output images
seed = random.randint(0, 2**31 - 1)
def transform(image):
r = image
r = tf.image.resize_images(r, [a.scale_size, a.scale_size], method=tf.image.ResizeMethod.AREA)
return r
with tf.name_scope("input_images"):
input_images = transform(inputs)
paths_batch, inputs_batch, targets_batch = tf.train.batch([paths, input_images, input_images], batch_size=a.batch_size)
steps_per_epoch = int(math.ceil(len(input_paths) / a.batch_size))
return Examples(
paths=paths_batch,
inputs=inputs_batch,
count=len(input_paths),
steps_per_epoch=steps_per_epoch,
)
def save_model(self, saver, sess, step):
"""
save model with path error checking
"""
if not os.path.exists(os.path.join(self.model_path, 'Checkpoints')):
os.makedirs(os.path.join(self.model_path, 'Checkpoints'))
my_path = os.path.join(self.model_path, 'Checkpoints\my_model.ckpt')
saver.save(sess, os.path.join(os.getcwd(), my_path), global_step=step)
if __name__ == "__main__":
filenames= [('C:/Users/Daniel/Desktop/myc_e2f_data/figureD/distribution_myc_ef_2_%d.png' %i) for i in range(1,501)]
filename_queue = tf.train.string_input_producer(filenames, num_epochs=None)
reader = tf.WholeFileReader()
_, content = reader.read(filename_queue)
image = tf.image.decode_png(content, channels=3)
image = tf.cast(image, tf.float32)
resized_image= tf.image.resize_images(image, [64, 64])
result = tf.train.shuffle_batch([resized_image], batch_size=300, capacity=500, min_after_dequeue=100)
#print(result)
#pdb.set_trace()
train_size=500
valid_size=300
with tf.Session() as sess:
sess.run(tf.local_variables_initializer())
def single_JPEGimage_reader(filename_queue):
image_reader = tf.WholeFileReader()
_, image_file = image_reader.read(filename_queue)
image = (tf.to_float(tf.image.decode_jpeg(image_file, channels=3)))
image = tf.image.resize_images(image,[HEIGHT,WIDTH],method=tf.image.ResizeMethod.NEAREST_NEIGHBOR)
return image
# read csv
csv_file = tf.train.string_input_producer(["./label.csv"], shuffle=True)
csv_reader = tf.TextLineReader()
_, line = csv_reader.read(csv_file)
image_file, label_decoded = tf.decode_csv(line, record_defaults=[[""], [""]])
image_decoded = tf.image.decode_jpeg(tf.read_file(image_file), channels=1)
image_cast = tf.cast(image_decoded, tf.float32)
image = tf.reshape(image_cast,
[IMAGE_WIDTH, IMAGE_HEIGHT, 1]) # 64 by 64 , grayscale
test_batch = int(12500 / BATCH_SIZE)
test_image_list = [
'./resize_test/' + file_name for file_name in os.listdir('./resize_test/')
]
test_image_reader = tf.WholeFileReader()
test_image_name = tf.train.string_input_producer(test_image_list)
_, value = test_image_reader.read(test_image_name)
test_image_decode = tf.cast(tf.image.decode_jpeg(value, channels=1),
tf.float32)
test_image = tf.reshape(test_image_decode, [IMAGE_WIDTH, IMAGE_HEIGHT, 1])
# make batch file
image_batch, label_batch, test_batch_x = tf.train.shuffle_batch(
[image, label_decoded, test_image],
batch_size=BATCH_SIZE,
num_threads=4,
capacity=50000,
min_after_dequeue=10000)
# start session
ran_lst = []
PARTION_SIZE = 1000
if __name__ == "__main__":
md5_label_f = open("../data/md5_ctype", "r")
for items in md5_label_f.readlines():
try:
md5, ctype = items.strip().split("\t")
if len(md5) > 16:
md5_lst += [md5]
md5_dic[md5] = ctype
elif md5 not in md5_dic:
continue
else:
logger.warning("wrong length of md5ctype:%s" % items.strip())
except:
logger.warning("wrong formats of md5ctype:%s" % items.strip())
random.shuffle(md5_lst)
index = 0
while index < len(md5_lst):
ran_lst += [md5_lst[index: min(index+PARTION_SIZE, len(md5_lst))]]
index += PARTION_SIZE
print len(ran_lst)
for ran_part in ran_lst:
for md5 in ran_part:
ctype = md5_dic[md5]
key, value = tf.WholeFileReader().read(tf.train.string_input_producer([".".join(("../data/" + md5, "jpg"))]))
tf_image = tf.image.decode_png(value) # use png or jpg decoder based on your files.
def create_input_pipeline(files,
batch_size,
n_epochs,
shape,
crop_shape=None,
crop_factor=1.0,
n_threads=2):
"""Creates a pipefile from a list of image files.
Includes batch generator/central crop/resizing options.
The resulting generator will dequeue the images batch_size at a time until
it throws tf.errors.OutOfRangeError when there are no more images left in
the queue.
Parameters
----------
files : list
List of paths to image files.
batch_size : int
Number of image files to load at a time.
n_epochs : int
Number of epochs to run before raising tf.errors.OutOfRangeError
shape : list
[height, width, channels]
crop_shape : list
[height, width] to crop image to.
crop_factor : float
Percentage of image to take starting from center.
n_threads : int, optional
Number of threads to use for batch shuffling
"""
# We first create a "producer" queue. It creates a production line which
# will queue up the file names and allow another queue to deque the file
# names all using a tf queue runner.
# Put simply, this is the entry point of the computational graph.
# It will generate the list of file names.
# We also specify it's capacity beforehand.
producer = tf.train.string_input_producer(files, capacity=len(files))
# We need something which can open the files and read its contents.
reader = tf.WholeFileReader()
# We pass the filenames to this object which can read the file's contents.
# This will create another queue running which dequeues the previous queue.
keys, vals = reader.read(producer)
# And then have to decode its contents as we know it is a jpeg image
imgs = tf.image.decode_jpeg(
vals, channels=3 if len(shape) > 2 and shape[2] == 3 else 0)
# We have to explicitly define the shape of the tensor.
# This is because the decode_jpeg operation is still a node in the graph
# and doesn't yet know the shape of the image. Future operations however
# need explicit knowledge of the image's shape in order to be created.
imgs.set_shape(shape)
# Next we'll centrally crop the image to the size of 100x100.
# This operation required explicit knowledge of the image's shape.
if shape[0] > shape[1]:
rsz_shape = [
int(shape[0] / shape[1] * crop_shape[0] / crop_factor),
int(crop_shape[1] / crop_factor)
]
else:
rsz_shape = [
int(crop_shape[0] / crop_factor),
int(shape[1] / shape[0] * crop_shape[1] / crop_factor)
]
rszs = tf.image.resize_images(imgs, rsz_shape)
crops = (tf.image.resize_image_with_crop_or_pad(rszs, crop_shape[0],
crop_shape[1])
if crop_shape is not None else imgs)
# Now we'll create a batch generator that will also shuffle our examples.
# We tell it how many it should have in its buffer when it randomly
# permutes the order.
min_after_dequeue = len(files) // 100
# The capacity should be larger than min_after_dequeue, and determines how
# many examples are prefetched. TF docs recommend setting this value to:
# min_after_dequeue + (num_threads + a small safety margin) * batch_size
capacity = min_after_dequeue + (n_threads + 1) * batch_size
# Randomize the order and output batches of batch_size.
batch = tf.train.shuffle_batch([crops],
enqueue_many=False,
batch_size=batch_size,
capacity=capacity,
min_after_dequeue=min_after_dequeue,
num_threads=n_threads)
# alternatively, we could use shuffle_batch_join to use multiple reader
# instances, or set shuffle_batch's n_threads to higher than 1.
return batch
def read_data(filename_queue):
reader = tf.WholeFileReader()
_, image = reader.read(filename_queue)
image = tf.image.per_image_standardization(
tf.image.decode_jpeg(image, channels=1))
return tf.image.random_flip_left_right(image)
def load_images(data_root):
filename_queue = tf.train.string_input_producer(data_root)
image_reader = tf.WholeFileReader()
key, image_file = image_reader.read(filename_queue)
image = tf.image.decode_jpeg(image_file)
return image, key
def main(_):
os.environ["CUDA_VISIBLE_DEVICES"] = FLAGS.gpu_id
if not tf.gfile.Exists(FLAGS.output_dir):
tf.gfile.MakeDirs(FLAGS.output_dir)
with tf.Graph().as_default() as g:
with open(FLAGS.input_fname, 'r') as f:
filenames = [line.split(',')[0][:-4] for line in f.readlines()]
filenames = [
os.path.join(FLAGS.image_dir, name) for name in filenames \
if not os.path.exists(os.path.join(FLAGS.output_dir, name + '.npy'))
]
filename_queue = tf.train.string_input_producer(filenames)
reader = tf.WholeFileReader()
key, value = reader.read(filename_queue)
image = tf.image.decode_jpeg(value, channels=3)
image_size = resnet_v1.resnet_v1.default_image_size
processed_image = vgg_preprocessing.preprocess_image(
image, image_size, image_size, is_training=False
)
processed_images, keys = tf.train.batch(
[processed_image, key],
FLAGS.batch_size,
num_threads=8, capacity=8*FLAGS.batch_size*5,
allow_smaller_final_batch=True
)
# Create the model
with slim.arg_scope(resnet_v1.resnet_arg_scope()):
net, end_points = resnet_v1.resnet_v1_101(
processed_images, num_classes=1000, is_training=False
)
init_fn = slim.assign_from_checkpoint_fn(
FLAGS.checkpoint_dir, slim.get_model_variables()
)
pool5 = g.get_operation_by_name('resnet_v1_101/pool5').outputs[0]
pool5 = tf.transpose(pool5, perm=[0, 3, 1, 2]) # (batch_size, 2048, 1, 1)
with tf.Session() as sess:
init_fn(sess)
coord = tf.train.Coordinator()
threads = tf.train.start_queue_runners(coord=coord)
try:
for step in tqdm(
xrange(len(filenames) / FLAGS.batch_size + 1), ncols=70
):
if coord.should_stop():
break
file_names, pool5_value = sess.run([keys, pool5])
for i in xrange(len(file_names)):
np.save(
os.path.join(
FLAGS.output_dir,
os.path.basename(file_names[i]) + '.npy'
),
pool5_value[i].astype(np.float32)
)
except tf.errors.OutOfRangeError:
print "Done feature extraction -- epoch limit reached"
finally:
coord.request_stop()
coord.join(threads)
import tensorflow as tf from PIL import Image IMAGE_PATH = '../data/Face00001.png' LABEL_PATH = '../data/Label.csv' IMAGE_LIST = [IMAGE_PATH] LABEL_LIST = [LABEL_PATH] IMAGE_WIDTH = 49 IMAGE_HEIGHT = 61 label_queue = tf.train.string_input_producer(LABEL_LIST) image_queue = tf.train.string_input_producer(IMAGE_LIST) reader_csv = tf.TextLineReader() reader_image = tf.WholeFileReader() label_key, label_value = reader_csv.read(label_queue) key, value = reader_image.read(image_queue) image_decoded = tf.image.decode_png(value) label_decoded = tf.decode_csv(label_value, record_defaults=[[0]]) x = tf.cast(image_decoded, tf.float32) y_ = tf.cast(label_decoded, tf.float32) y_ = tf.reshape(y_, [-1, 1]) w1 = tf.Variable(tf.truncated_normal([5, 5, 1, 32])) b1 = tf.Variable(tf.zeros([32])) x_image = tf.reshape(x, [-1, IMAGE_WIDTH, IMAGE_HEIGHT, 1]) conv1 = tf.nn.conv2d(x_image, w1, strides=[1, 1, 1, 1], padding='SAME') h1 = tf.nn.relu(conv1 + b1)
def read_in_data(file_path):
##################################### READ IN DATA #####################################
# Make a queue of file names including all the JPEG images files in the relative
# image directory.
file_names = tf.train.match_filenames_once(file_path + "/*.jpg")
filename_queue = tf.train.string_input_producer(file_names)
# Used to read the entire file
file_reader = tf.WholeFileReader()
# Read a whole file from the queue. Not entirely sure how this works but it returns
# a key and value, that we can use to iterate through the entire file.
key, value = file_reader.read(filename_queue)
#Array to hold our image data and also our labels
data = []
lab = []
with tf.Session() as sess:
tf.initialize_all_variables().run()
# Coordinate the loading of image files.
coord = tf.train.Coordinator()
threads = tf.train.start_queue_runners(coord=coord)
seen = set()
# print len(file_names.eval())
# print len(set(file_names.eval()))
for i in range(len(file_names.eval())):
# I think what happens when you call eval on either key or value, it will iterate both key and
# val to the next item in the queue. I tried calling key.eval and value.eval both and that f****d it up.
# Key here is the file's name. Value is actually the file.
file_name = key.eval()
labels = file_name.decode().split("-")
if (i % 100 == 0 and i != 0):
print i
if file_name in seen:
print("seen:") + file_name
else:
seen.add(file_name)
# [male, female]
# labels[1][1] is race and sex
# if labels[1][1] == 'M': #'AM' stands for Asian Male, but I only have AM and AF currently in the file
# lab.append([1, 0])
# else: # 0 for girls
# lab.append([0, 1])
# [Happy Open, Happy Closed, Neutral, Angry, Fear]
image_emotion = labels[4][:-4].split('.')[0]
if (image_emotion == 'HO'):
lab.append([1, 0, 0, 0, 0])
elif (image_emotion == 'HC'):
lab.append([0, 1, 0, 0, 0])
elif (image_emotion == 'N'):
lab.append([0, 0, 1, 0, 0])
elif (image_emotion == 'A'):
lab.append([0, 0, 0, 1, 0])
elif (image_emotion == 'F'):
lab.append([0, 0, 0, 0, 1])
else:
print('ERROR: UNEXPECTED LABEL RECEIVED')
sys.exit()
# Read the in the image and decode it. It might be possible here to use "value" instead of,
# but I'm not sure. You guys can mess with it.
my_img = tf.image.decode_jpeg(tf.read_file(file_name), channels=1)
image = tf.image.resize_images(my_img, 31, 32)
leh = tf.reshape(image, [1, 992])
data.append(leh.eval()[0])
coord.request_stop()
coord.join(threads)
print("Done reading images.")
return data, lab
def get_loader(root,
batch_size,
scale_size,
data_format,
split=None,
is_grayscale=False,
seed=None):
dataset_name = os.path.basename(root)
if dataset_name in ['CelebA'] and split:
root = os.path.join(root, 'splits', split)
print("Root is {}".format(root))
for ext in ["jpg", "png"]:
paths = glob("{}/*.{}".format(root, ext))
if ext == "jpg":
tf_decode = tf.image.decode_jpeg
elif ext == "png":
tf_decode = tf.image.decode_png
if len(paths) != 0:
break
with Image.open(paths[0]) as img:
w, h = img.size
shape = [h, w, 3]
filename_queue = tf.train.string_input_producer(list(paths),
shuffle=False,
seed=seed)
reader = tf.WholeFileReader()
filename, data = reader.read(filename_queue)
image = tf_decode(data, channels=3)
if is_grayscale:
image = tf.image.rgb_to_grayscale(image)
image.set_shape(shape)
min_after_dequeue = 5000
capacity = min_after_dequeue + 3 * batch_size
queue = tf.train.shuffle_batch([image],
batch_size=batch_size,
num_threads=4,
capacity=capacity,
min_after_dequeue=min_after_dequeue,
name='synthetic_inputs')
if dataset_name in ['CelebA']:
queue = tf.image.crop_to_bounding_box(queue, 50, 25, 128, 128)
queue = tf.image.resize_nearest_neighbor(queue,
[scale_size, scale_size])
else:
queue = tf.image.resize_nearest_neighbor(queue,
[scale_size, scale_size])
if data_format == 'NCHW':
queue = tf.transpose(queue, [0, 3, 1, 2])
elif data_format == 'NHWC':
pass
else:
raise Exception("[!] Unkown data_format: {}".format(data_format))
return tf.to_float(queue)
def load_train_batch(self):
"""Load a batch of training instances.
"""
opt = self.opt
# Load the list of training files into queues
file_list = self.format_file_list(opt.dataset_dir, 'train')
# file_list['image_file_list']
# '/userhome/34/h3567721/dataset/kitti/kitti_raw_eigen/2011_09_30_drive_0034_sync_03/0000000558.jpg'
# file_list['cam_file_list']
# '/userhome/34/h3567721/dataset/kitti/kitti_raw_eigen/2011_09_30_drive_0034_sync_03/0000000558_cam.txt'
image_paths_queue = tf.train.string_input_producer(
file_list['image_file_list'], shuffle=False)
cam_paths_queue = tf.train.string_input_producer(
file_list['cam_file_list'], shuffle=False)
# Load images
img_reader = tf.WholeFileReader()
_, image_contents = img_reader.read(image_paths_queue)
image_seq = tf.image.decode_jpeg(image_contents)
# tgt_image: (128, 416, 3)
# src_image_stack: (128, 416, 3)
tgt_image, src_image_stack = \
self.unpack_image_sequence(
image_seq, opt.img_height, opt.img_width, opt.num_source)
# Load camera intrinsics
cam_reader = tf.TextLineReader()
_, raw_cam_contents = cam_reader.read(cam_paths_queue)
rec_def = []
for i in range(9):
rec_def.append([1.])
# rec_def: [[1.0], [1.0], [1.0], [1.0], [1.0], [1.0], [1.0], [1.0], [1.0]]
raw_cam_vec = tf.decode_csv(raw_cam_contents, record_defaults=rec_def)
raw_cam_vec = tf.stack(raw_cam_vec)
intrinsics = tf.reshape(raw_cam_vec, [3, 3])
# ------------------------------------------------------
# Form training batches
seed = random.randint(0, 2**31 - 1)
min_after_dequeue = 2048
# 2048 * 32 * 4 = 262144
capacity = min_after_dequeue + opt.num_threads * opt.batch_size
# src_image_stack: (4,128,416,6)
# tgt_image: (4,128,416,3)
# intrinsics: (4,3,3)
src_image_stack, tgt_image, intrinsics = \
tf.train.shuffle_batch([src_image_stack, tgt_image, intrinsics], opt.batch_size,
capacity, min_after_dequeue, opt.num_threads, seed)
# Data augmentation
image_all = tf.concat([tgt_image, src_image_stack], axis=3)
image_all, intrinsics = self.data_augmentation(image_all, intrinsics,
opt.img_height,
opt.img_width)
tgt_image = image_all[:, :, :, :3] # (4,128,416,3)
src_image_stack = image_all[:, :, :, 3:] # (4,128,416,6)
# intrinsics_mscale (4,4,3,3)
intrinsics = self.get_multi_scale_intrinsics(intrinsics,
opt.num_scales)
return tgt_image, src_image_stack, intrinsics
def load_examples():
# 判断input路径是否存在
if a.input_dir is None or not os.path.exists(a.input_dir):
raise Exception("input_dir does not exist")
'''
glob.glob(pathname)
pathname是一个字符串，允许出现“*”、“?”、"["、"]"等通配符，不能含有“~”
这个函数根据pathname筛选匹配的路径
这个函数是通过配合使用os.listdir()与fnmatch.fnmatch()实现的，而非调用一个子shell
'''
# 匹配出图片路径
input_paths = glob.glob(os.path.join(a.input_dir, "*.jpg"))
decode = tf.image.decode_jpeg
if len(input_paths) == 0:
input_paths = glob.glob(os.path.join(a.input_dir, "*.png"))
decode = tf.image.decode_png
if len(input_paths) == 0:
raise Exception("input_dir contains no image files")
'''
os.path.basename(path)
获取路径最后的文件名，当路径以“\”结尾时，返回空字符串
os.path.splitext(path)
分离文件路径的文件名和扩展名
'''
def get_name(path):
name, _ = os.path.splitext(os.path.basename(path))
return name
'''
string.isdigit()
判断字符串是否全为数字组成
all()
当元素全为True时返回True，否则返回False
ang()
当元素全为False时返回True，否则返回False
'''
# 排序
# if the image names are numbers, sort by the value rather than asciibetically
# having sorted inputs means that the outputs are sorted in test mode
if all(get_name(path).isdigit() for path in input_paths):
input_paths = sorted(input_paths, key=lambda path: int(get_name(path)))
else:
input_paths = sorted(input_paths)
with tf.name_scope("load_images"):
# 使用文件名队列读取数据
path_queue = tf.train.string_input_producer(input_paths,
shuffle=a.mode == "train")
reader = tf.WholeFileReader()
paths, contents = reader.read(path_queue)
# decode图片，转换到[0,1]区间
raw_input = decode(contents)
raw_input = tf.image.convert_image_dtype(raw_input, dtype=tf.float32)
# 断言图片必须有三通道
assertion = tf.assert_equal(tf.shape(raw_input)[2],
3,
message="image does not have 3 channels")
with tf.control_dependencies([assertion]):
raw_input = tf.identity(raw_input)
raw_input.set_shape([None, None, 3])
# 是否转换到LAP颜色空间
# 把数据转换到[-1,1]区间
if a.lab_colorization:
# load color and brightness from image, no B image exists here
lab = rgb_to_lab(raw_input)
L_chan, a_chan, b_chan = preprocess_lab(lab)
a_images = tf.expand_dims(L_chan, axis=2)
b_images = tf.stack([a_chan, b_chan], axis=2)
else:
# break apart image pair and move to range [-1, 1]
width = tf.shape(raw_input)[1] # [height, width, channels]
a_images = preprocess(raw_input[:, :width // 2, :])
b_images = preprocess(raw_input[:, width // 2:, :])
# 样例图中，左侧为有雾图像，右侧为无雾图像
if a.which_direction == "AtoB":
inputs, targets = [a_images, b_images]
elif a.which_direction == "BtoA":
inputs, targets = [b_images, a_images]
else:
raise Exception("invalid direction")
# synchronize seed for image operations so that we do the same operations to both
# input and output images
seed = random.randint(0, 2**31 - 1)
# 变换图像
# 水平翻转、裁切
def transform(image):
r = image
if a.flip:
r = tf.image.random_flip_left_right(r, seed=seed)
# area produces a nice downscaling, but does nearest neighbor for upscaling
# assume we're going to be doing downscaling here
r = tf.image.resize_images(r, [a.scale_size, a.scale_size],
method=tf.image.ResizeMethod.AREA)
offset = tf.cast(tf.floor(
tf.random_uniform([2], 0, a.scale_size - CROP_SIZE + 1,
seed=seed)),
dtype=tf.int32)
if a.scale_size > CROP_SIZE:
r = tf.image.crop_to_bounding_box(r, offset[0], offset[1],
CROP_SIZE, CROP_SIZE)
elif a.scale_size < CROP_SIZE:
raise Exception("scale size cannot be less than crop size")
return r
with tf.name_scope("input_images"):
input_images = transform(inputs)
with tf.name_scope("target_images"):
target_images = transform(targets)
# 分割batch、epoch
paths_batch, inputs_batch, targets_batch = tf.train.batch(
[paths, input_images, target_images], batch_size=a.batch_size)
steps_per_epoch = int(math.ceil(len(input_paths) / a.batch_size))
return Examples(
paths=paths_batch,
inputs=inputs_batch,
targets=targets_batch,
count=len(input_paths),
steps_per_epoch=steps_per_epoch,
)
def main(unused_argv):
tf.logging.set_verbosity(tf.logging.INFO)
# Read list of images.
tf.logging.info('Reading list of images...')
image_paths = _ReadImageList(cmd_args.list_images_path)
num_images = len(image_paths)
tf.logging.info('done! Found %d images', num_images)
# Parse DelfConfig proto.
config = delf_config_pb2.DelfConfig()
with tf.gfile.FastGFile(cmd_args.config_path, 'r') as f:
text_format.Merge(f.read(), config)
# Create output directory if necessary.
if not os.path.exists(cmd_args.output_dir):
os.makedirs(cmd_args.output_dir)
# Tell TensorFlow that the model will be built into the default Graph.
with tf.Graph().as_default():
# Reading list of images.
filename_queue = tf.train.string_input_producer(image_paths,
shuffle=False)
print("PRINTING THE fileNAME QUEUE", filename_queue)
reader = tf.WholeFileReader()
print("PRINTING THE reader", reader)
_, value = reader.read(filename_queue)
print("PRINTING the VALUE", value)
image_tf = tf.image.decode_jpeg(value, channels=3)
print("THE IMAGE", image_tf)
"""
ADDED FOR CHECKING PURPOSE ONLY
"""
imagetest = image_tf
imagetest = tf.expand_dims(imagetest, 0)
print("size of tensor === ", imagetest)
print("THE IMAGE FORMED", imagetest)
print("MAKING FIRST FUNCTION")
modelsssss = feature_extractor.BuildModel('resnet_v1_50/block3',
'softplus',
'use_l2_normalized_feature',
1)
print("CALLING THE MODEL")
buildedModel, _ = modelsssss(imagetest, False, False)
print("CAME BACK TO ORIGINAL")
with tf.Session() as sess:
# Initialize variables.
init_op = tf.global_variables_initializer()
print("SHOWING OPERATION", init_op)
tf.logging.info('Starting session...')
sess.run(init_op)
tf.logging.info('Starting to load the models to be used...')
# Loading model that will be used.
tf.saved_model.loader.load(sess,
[tf.saved_model.tag_constants.SERVING],
config.model_path)
graph = tf.get_default_graph()
input_image = graph.get_tensor_by_name('input_image:0')
input_score_threshold = graph.get_tensor_by_name(
'input_abs_thres:0')
input_image_scales = graph.get_tensor_by_name('input_scales:0')
input_max_feature_num = graph.get_tensor_by_name(
'input_max_feature_num:0')
boxes = graph.get_tensor_by_name('boxes:0')
raw_descriptors = graph.get_tensor_by_name('features:0')
feature_scales = graph.get_tensor_by_name('scales:0')
attention_with_extra_dim = graph.get_tensor_by_name('scores:0')
attention = tf.reshape(attention_with_extra_dim,
[tf.shape(attention_with_extra_dim)[0]])
locations, descriptors = feature_extractor.DelfFeaturePostProcessing(
boxes, raw_descriptors, config)
tf.logging.info('Finished loading the models ...')
print('Hi')
print('------------------------locations--------------- = ',
locations)
# Start input enqueue threads.
coord = tf.train.Coordinator()
threads = tf.train.start_queue_runners(sess=sess, coord=coord)
start = time.clock()
for i in range(num_images):
# Write to log-info once in a while.
if i == 0:
tf.logging.info(
'Starting to extract DELF features from images...')
elif i % _STATUS_CHECK_ITERATIONS == 0:
elapsed = (time.clock() - start)
tf.logging.info(
'Processing image %d out of %d, last %d '
'images took %f seconds', i, num_images,
_STATUS_CHECK_ITERATIONS, elapsed)
start = time.clock()
# # Get next image.
im = sess.run(image_tf)
# If descriptor already exists, skip its computation.
out_desc_filename = os.path.splitext(
os.path.basename(image_paths[i]))[0] + _DELF_EXT
out_desc_fullpath = os.path.join(cmd_args.output_dir,
out_desc_filename)
if tf.gfile.Exists(out_desc_fullpath):
tf.logging.info('Skipping %s', image_paths[i])
continue
# Extract and save features.
(locations_out, descriptors_out, feature_scales_out,
attention_out) = sess.run(
[locations, descriptors, feature_scales, attention],
feed_dict={
input_image:
im,
input_score_threshold:
config.delf_local_config.score_threshold,
input_image_scales:
list(config.image_scales),
input_max_feature_num:
config.delf_local_config.max_feature_num
})
feature_io.WriteToFile(out_desc_fullpath, locations_out,
feature_scales_out, descriptors_out,
attention_out)
# Finalize enqueue threads.
coord.request_stop()
coord.join(threads)
seed = 5555
image_names = tf.convert_to_tensor(
sorted(glob.glob(os.path.join(feat_dir, '*.jpg'))))
mask_names = tf.convert_to_tensor(
sorted(glob.glob(os.path.join(mask_dir, '*.png'))))
feature_queue = tf.train.string_input_producer(image_names,
shuffle=True,
seed=seed)
mask_queue = tf.train.string_input_producer(mask_names,
shuffle=True,
seed=seed)
image_reader = tf.WholeFileReader()
image_key, image_file = image_reader.read(feature_queue)
image = tf.image.decode_image(image_file)
mask_reader = tf.WholeFileReader()
mask_key, mask_file = image_reader.read(mask_queue)
mask = tf.image.decode_image(mask_file)
image = tf.divide(image, 255)
mask = tf.divide(mask, 255)
## Do a random crop
## Important? Do the crop before batching.
image_mask = tf.concat([image, mask], 2)
image_mask = tf.random_crop(image_mask, [crop_size, crop_size, 4])
image, mask = tf.split(image_mask, [3, 1], axis=2)
def dataloader(opt):
if opt.input_dir is None or not os.path.exists(opt.input_dir):
raise Exception("input_dir does not exist")
# return list of paht of matched files
input_path = glob.glob(os.path.join(opt.input_dir, "*.jpg"))
decode = tf.image.decode_jpeg
if len(input_path) == 0:
input_path = glob.glob(os.path.join(opt.input_dir, "*.png"))
decode = tf.image.decode_png
if len(input_path) == 0:
raise Exception("input_dir continput_dirains no image files")
# if the image names are numbers, sort by the value rather than asciibetically
# having sorted inputs means that the outputs are sorted in test mode
if all(get_filename(path).isdigit() for path in input_path):
input_path = sorted(input_path, key=lambda path: int(get_filename(path)))
else:
input_path = sorted(input_path)
with tf.name_scope("load_images"):
path_queue = tf.train.string_input_producer(input_path, shuffle=opt.mode == "train")
reader = tf.WholeFileReader()
paths, contents = reader.read(path_queue)
raw_input = decode(contents)
raw_input = tf.image.convert_image_dtype(raw_input, dtype=tf.float32)
assertion = tf.assert_equal(tf.shape(raw_input)[2], 3, message="image does not have 3 channels")
with tf.control_dependencies([assertion]):
raw_input = tf.identity(raw_input)
raw_input.set_shape([None, None, 3])
if opt.lab_colorization:
# load color and brightness from image, no B image exists here
lab = rgb_to_lab(raw_input)
L_chan, a_chan, b_chan = preprocess_lab(lab)
a_images = tf.expand_dims(L_chan, axis=2)
b_images = tf.stack([a_chan, b_chan], axis=2)
else:
# break apart image pair and move to range [-1, 1]
width = tf.shape(raw_input)[1] # [height, width, channels]
a_images = preprocess(raw_input[:, :width//2, :])
b_images = preprocess(raw_input[:, width//2:, :])
if opt.which_direction == "AtoB":
inputs, targets = [a_images, b_images]
elif opt.which_direction == "BtoA":
inputs, targets = [b_images, a_images]
else:
raise Exception("invalid direction")
# synchronize seed for image operations so that we do the same operations to both
# input and output images
seed = random.randint(0, 2**31 - 1)
with tf.name_scope("input_images"):
input_images = transform(inputs, opt)
with tf.name_scope("target_images"):
target_images = transform(targets, opt)
paths_batch, inputs_batch, targets_batch = tf.train.batch([paths, input_images, target_images], batch_size=opt.batch_size)
steps_per_epoch = int(math.ceil(len(input_path) / opt.batch_size))
return Examples(
paths=paths_batch,
inputs=inputs_batch,
targets=targets_batch,
count=len(input_path),
steps_per_epoch=steps_per_epoch,
)
def load_train_batch(self):
"""Load a batch of training instances.
"""
opt = self.opt
# Load the list of training files into queues
#TODO
if opt.train_lite:
file_list = self.format_file_list(opt.dataset_dir,
opt.filelist_dir, 'train_lite')
else:
file_list = self.format_file_list(opt.dataset_dir,
opt.filelist_dir, 'train')
image_paths_queue = tf.train.string_input_producer(
file_list['image_file_list'], shuffle=False)
cam_paths_queue = tf.train.string_input_producer(
file_list['cam_file_list'], shuffle=False)
# Load camera intrinsics
cam_reader = tf.TextLineReader()
_, raw_cam_contents = cam_reader.read(cam_paths_queue)
rec_def = []
for i in range(9):
rec_def.append([1.])
raw_cam_vec = tf.decode_csv(raw_cam_contents, record_defaults=rec_def)
raw_cam_vec = tf.stack(raw_cam_vec)
intrinsics = tf.reshape(raw_cam_vec, [3, 3])
# Load images
img_reader = tf.WholeFileReader()
_, image_contents = img_reader.read(image_paths_queue)
image_seq = tf.image.decode_jpeg(image_contents)
tgt_image, src_image_stack = \
self.unpack_image_sequence(
image_seq, opt.img_height, opt.img_width, opt.num_source)
#TODO Load Semantics
# See cityscape label defs in https://github.com/mcordts/cityscapesScripts/blob/master/cityscapesscripts/helpers/labels.py#L62
# Also notice that deeplabv3+ uses `train_id` https://github.com/tensorflow/models/blob/69b016449ffc797421bf003d8b7fd8545db866d7/research/deeplab/datasets/build_cityscapes_data.py#L46
# Color maps are in https://github.com/tensorflow/models/blob/69b016449ffc797421bf003d8b7fd8545db866d7/research/deeplab/utils/get_dataset_colormap.py#L207
if opt.sem_assist:
sem_paths_queue = tf.train.string_input_producer(
file_list['sem_image_file_list'], shuffle=False)
sem_reader = tf.WholeFileReader()
sem_keys, sem_contents = sem_reader.read(sem_paths_queue)
if opt.load_from_raw:
sem_seq = tf.reshape(
tf.decode_raw(sem_contents, tf.uint8),
[1, opt.img_height, (opt.num_source + 1) * opt.img_width])
else:
sem_seq = tf.py_func(read_npy_file, [sem_keys], [
tf.uint8,
])
#TODO Load Instances: we use COCO
# Two channels: class and id level. For id level we only use the edge
if opt.ins_assist:
ins_paths_queue = tf.train.string_input_producer(
file_list['ins_image_file_list'], shuffle=False)
ins_reader = tf.WholeFileReader()
ins_keys, ins_contents = ins_reader.read(ins_paths_queue)
if opt.load_from_raw:
ins_seq = tf.reshape(tf.decode_raw(ins_contents, tf.uint8), [
1, opt.img_height, (opt.num_source + 1) * opt.img_width, 2
])
else:
ins_seq = tf.py_func(read_npy_file, [ins_keys], [
tf.uint8,
])
#TODO 1. SHUFFLE BATCH
# Form training batches
seed = random.randint(0, 2**31 - 1)
min_after_dequeue = 2048
capacity = min_after_dequeue + opt.num_threads * opt.batch_size
if opt.sem_assist and opt.ins_assist:
src_image_stack, tgt_image, intrinsics, sem_seq, ins_seq = tf.train.shuffle_batch(
[
src_image_stack, tgt_image, intrinsics, sem_seq[0],
ins_seq[0]
], opt.batch_size, capacity, min_after_dequeue,
opt.num_threads, seed)
elif opt.sem_assist:
src_image_stack, tgt_image, intrinsics, sem_seq = tf.train.shuffle_batch(
[src_image_stack, tgt_image, intrinsics, sem_seq[0]],
opt.batch_size, capacity, min_after_dequeue, opt.num_threads,
seed)
elif opt.ins_assist:
src_image_stack, tgt_image, intrinsics, ins_seq = tf.train.shuffle_batch(
[src_image_stack, tgt_image, intrinsics, ins_seq[0]],
opt.batch_size, capacity, min_after_dequeue, opt.num_threads,
seed)
else:
src_image_stack, tgt_image, intrinsics = tf.train.shuffle_batch(
[src_image_stack, tgt_image, intrinsics], opt.batch_size,
capacity, min_after_dequeue, opt.num_threads, seed)
# semantic segmentation
tgt_sem = None
tgt_sem_map = None
tgt_sem_mask = None
tgt_sem_edge = None
src_sem_stack = None
src_sem_map_stack = None
src_sem_mask_stack = None
src_sem_edge_stack = None
# ins0 ~ instance level, but still class segmentation
tgt_ins0 = None
tgt_ins0_map = None
tgt_ins0_edge = None
src_ins0_stack = None
src_ins0_map_stack = None
src_ins0_edge_stack = None
# ins1 ~ instance level, but this is id segmentation
tgt_ins1_edge = None
src_ins1_edge_stack = None
#TODO 2. TRAMSFORMATION AND UNPACKING
if opt.sem_assist:
#TODO get one-hot encoded sem_oh_seq (4,128,1248,19)X{0,1}
sem_oh_seq = tf.cast(
tf.one_hot(sem_seq, on_value=1, depth=opt.sem_num_class),
tf.uint8)
#TODO decouple tgt_sem (4,128,1248,19)X{0,1} src_sem_stack (4,128,1248,2*19)X{0,1}
tgt_sem, src_sem_stack = self.unpack_sem_sequence_batch_atom(
sem_oh_seq, opt.sem_num_class)
#TODO get densemap sem_map_seq (4,128,1248,1)X{0,1,...,18}
sem_map_seq = tf.expand_dims(sem_seq, -1)
#TODO decouple tgt_sem_map (4,128,1248,1)X{0,1,...,18} src_sem_map_stack (4,128,1248,2*1)X{0,1,...,18}
tgt_sem_map, src_sem_map_stack = self.unpack_sem_sequence_batch_atom(
sem_map_seq, 1)
if opt.sem_mask_explore:
#TODO get sem mask sem_mask_seq (4,128,1248,c) here we assume c=1
sem_mask_seq = self.get_sem_mask_batch(sem_seq)
#TODO decouple tgt_sem_mask (4,128,1248,c) src_sem_mask_stack (4,128,1248,2*c)
tgt_sem_mask, src_sem_mask_stack = self.unpack_sem_sequence_batch_atom(
sem_mask_seq, 1)
if opt.sem_edge_explore:
#TODO get sem edge sem_edge_seq (4,128,1248,c) here we assume c=1
sem_edge_seq = self.get_sem_edge_batch(sem_seq)
#TODO decouple tgt_sem_edge (4,128,1248,c) src_sem_edge_stack (4,128,1248,2*c)
tgt_sem_edge, src_sem_edge_stack = self.unpack_sem_sequence_batch_atom(
sem_edge_seq, 1)
if opt.ins_assist:
ins0_seq = ins_seq[:, :, :, 0]
ins1_seq = ins_seq[:, :, :, 1]
#TODO get one-hot ins0_oh_seq (4,128,1248,81)X{0,1}
ins0_oh_seq = tf.cast(
tf.one_hot(ins0_seq, on_value=1, depth=opt.ins_num_class),
tf.uint8)
#ins1_oh_seq = tf.cast(tf.one_hot(ins1_seq, on_value=1, depth = 255), tf.uint8)
#TODO decouple tgt_ins0 (4,128,1248,81)X{0,1} src_ins0_stack (4,128,1248,2*81)X{0,1}
tgt_ins0, src_ins0_stack = self.unpack_sem_sequence_batch_atom(
ins0_oh_seq, opt.ins_num_class)
#tgt_ins1, src_ins1_stack = self.unpack_sem_sequence_batch_atom(ins1_oh_seq, opt.ins_num_class)
#TODO get densemap sem_ins0_seq (4,128,1248,1)X{0,1,...,80}
ins0_map_seq = ins_seq[:, :, :, :1]
ins1_map_seq = ins_seq[:, :, :, 1:]
#TODO decouple tgt_ins0_map (4,128,1248,1)X{0,1,...,80} src_ins0_map_stack (4,128,1248,2*1)X{0,1,...,80}
tgt_ins0_map, src_ins0_map_stack = self.unpack_sem_sequence_batch_atom(
ins0_map_seq, 1)
tgt_ins1_map, src_ins1_map_stack = self.unpack_sem_sequence_batch_atom(
ins1_map_seq, 1)
if opt.ins0_edge_explore:
#TODO get edge ins0_edge_seq (4,128,1248,c) here we assume c=1
ins0_edge_seq = self.get_sem_edge_batch(ins0_seq)
#TODO decouple tgt_ins0_edge (4,128,1248,c) src_ins0_edge_stack (4,128,1248,2*c)
tgt_ins0_edge, src_ins0_edge_stack = self.unpack_sem_sequence_batch_atom(
ins0_edge_seq, 1)
if opt.ins1_edge_explore:
#TODO get edge ins1_edge_seq (4,128,1248,c) here we assume c=1
ins1_edge_seq = self.get_sem_edge_batch(ins1_seq)
#TODO decouple tgt_ins1_edge (4,128,1248,c) src_ins1_edge_stack (4,128,1248,2*c)
tgt_ins1_edge, src_ins1_edge_stack = self.unpack_sem_sequence_batch_atom(
ins1_edge_seq, 1)
#TODO 3. DATA AUGMENTATION
image_all = tf.concat([tgt_image, src_image_stack], axis=3)
image_all, intrinsics, aug_params = self.data_augmentation(
image_all, intrinsics, opt.img_height,
opt.img_width) #TODO changed API
if opt.sem_assist:
##TODO Do the same data augmentation for semantic segmentations
tgt_sem, src_sem_stack = self.data_aug(tgt_sem, src_sem_stack,
aug_params, "bilinear")
tgt_sem_map, src_sem_map_stack = self.data_aug(
tgt_sem_map, src_sem_map_stack, aug_params, "neighbor")
if self.opt.sem_mask_explore:
tgt_sem_mask, src_sem_mask_stack = \
self.data_aug(tgt_sem_mask, src_sem_mask_stack, aug_params, "bilinear")
if self.opt.sem_edge_explore:
tgt_sem_edge, src_sem_edge_stack = \
self.data_aug(tgt_sem_edge, src_sem_edge_stack, aug_params, "bilinear")
#TODO maybe transfer needs this settings self.data_aug(tgt_sem_edge, src_sem_edge_stack, aug_params, "neighbor")
if opt.ins_assist:
##TODO Do the same data augmentation for instance segmentations
tgt_ins0, src_ins0_stack = self.data_aug(tgt_ins0, src_ins0_stack,
aug_params, "bilinear")
#tgt_ins1, src_ins1_stack = self.data_aug(tgt_ins1, src_ins1_stack, aug_params, "bilinear")
tgt_ins0_map, src_ins0_map_stack = self.data_aug(
tgt_ins0_map, src_ins0_map_stack, aug_params, "neighbor")
#tgt_ins1_map, src_ins1_map_stack = self.data_aug(tgt_ins1_map, src_ins1_map_stack, aug_params, "neighbor")
if self.opt.ins0_edge_explore:
tgt_ins0_edge, src_ins0_edge_stack = \
self.data_aug(tgt_ins0_edge, src_ins0_edge_stack, aug_params, "bilinear")
#TODO maybe transfer needs this settings self.data_aug(tgt_ins0_edge, src_ins0_edge_stack, aug_params, "neighbor")
if self.opt.ins1_edge_explore:
tgt_ins1_edge, src_ins1_edge_stack = \
self.data_aug(tgt_ins1_edge, src_ins1_edge_stack, aug_params, "bilinear")
#TODO maybe transfer needs this settings self.data_aug(tgt_ins1_edge, src_ins1_edge_stack, aug_params, "neighbor")
# 4. RETURN
# image_channels=3*opt.seq_length
tgt_image = image_all[:, :, :, :3]
src_image_stack = image_all[:, :, :, 3:] #3:image_channels]
intrinsics = self.get_multi_scale_intrinsics(intrinsics,
opt.num_scales)
# if opt.sem_assist and opt.ins_assist:
return tgt_image, src_image_stack, intrinsics, \
[tgt_sem, tgt_sem_map, tgt_sem_mask, tgt_sem_edge], \
[src_sem_stack, src_sem_map_stack, src_sem_mask_stack, src_sem_edge_stack], \
[tgt_ins0, tgt_ins0_map, tgt_ins0_edge, tgt_ins1_edge], \
[src_ins0_stack, src_ins0_map_stack, src_ins0_edge_stack, src_ins1_edge_stack]
def make_data_tensor(self, training=True):
"""
:param training:
:return:
"""
if training:
folders = self.metatrain_folders
num_total_batches = self.total_batch_num
else:
folders = self.metaval_folders
num_total_batches = 600
if training and os.path.exists('filelist.pkl'):
labels = np.arange(self.nway).repeat(self.nimg).tolist()
with open('filelist.pkl', 'rb') as f:
all_filenames = pickle.load(f)
print('load episodes from file, len:', len(all_filenames))
else: # test or not existed.
# 16 in one class, 16*5 in one task
# [task1_0_img0, task1_0_img15, task1_1_img0,]
all_filenames = []
for _ in tqdm.tqdm(range(num_total_batches),
'generating episodes'): # 200000
# from image folder sample 5 class randomly
sampled_folders = random.sample(folders, self.nway)
random.shuffle(sampled_folders)
# sample 16 images from selected folders, and each with label 0-4, (0/1..., path), orderly, no shuffle!
# len: 5 * 16
labels_and_images = get_images(sampled_folders,
range(self.nway),
nb_samples=self.nimg,
shuffle=False)
# make sure the above isn't randomized order
labels = [li[0] for li in labels_and_images]
filenames = [li[1] for li in labels_and_images]
all_filenames.extend(filenames)
if training: # only save for training.
with open('filelist.pkl', 'wb') as f:
pickle.dump(all_filenames, f)
print('save all file list to filelist.pkl')
# make queue for tensorflow to read from
print('creating pipeline ops')
filename_queue = tf.train.string_input_producer(
tf.convert_to_tensor(all_filenames), shuffle=False)
image_reader = tf.WholeFileReader()
_, image_file = image_reader.read(filename_queue)
image = tf.image.decode_jpeg(image_file, channels=3)
# tensorflow format: N*H*W*C
image.set_shape((self.imgsz[0], self.imgsz[1], 3))
# reshape(image, [84*84*3])
image = tf.reshape(image, [self.dim_input])
# convert to range(0,1)
image = tf.cast(image, tf.float32) / 255.0
examples_per_batch = self.nway * self.nimg # 5*16
# batch here means batch of meta-learning, including 4 tasks = 4*80
batch_image_size = self.meta_batchsz * examples_per_batch # 4* 80
print('batching images')
images = tf.train.batch(
[image],
batch_size=batch_image_size, # 4*80
num_threads=self.meta_batchsz,
capacity=256 + 3 * batch_image_size, # 256 + 3* 4*80
)
all_image_batches, all_label_batches = [], []
print('manipulating images to be right order')
# images contains current batch, namely 4 task, 4* 80
for i in range(self.meta_batchsz): # 4
# current task, 80 images
image_batch = images[i * examples_per_batch:(i + 1) *
examples_per_batch]
# as all labels of all task are the same, which is 0,0,..1,1,..2,2,..3,3,..4,4...
label_batch = tf.convert_to_tensor(labels)
new_list, new_label_list = [], []
# for each image from 0 to 15 in all 5 class
for k in range(self.nimg): # 16
class_idxs = tf.range(0, self.nway) # 0-4
class_idxs = tf.random_shuffle(class_idxs)
# it will cope with 5 images parallelly
# [0, 16, 32, 48, 64] or [1, 17, 33, 49, 65]
true_idxs = class_idxs * self.nimg + k
new_list.append(tf.gather(image_batch, true_idxs))
new_label_list.append(tf.gather(label_batch, true_idxs))
# [80, 84*84*3]
new_list = tf.concat(
new_list, 0
) # has shape [self.num_classes*self.num_samples_per_class, self.dim_input]
# [80]
new_label_list = tf.concat(new_label_list, 0)
all_image_batches.append(new_list)
all_label_batches.append(new_label_list)
# [4, 80, 84*84*3]
all_image_batches = tf.stack(all_image_batches)
# [4, 80]
all_label_batches = tf.stack(all_label_batches)
# [4, 80, 5]
all_label_batches = tf.one_hot(all_label_batches, self.nway)
print('image_b:', all_image_batches)
print('label_onehot_b:', all_label_batches)
return all_image_batches, all_label_batches
import tensorflow as tf
image_filename = "./images/chapter-05-object-recognition-and-classification/working-with-images/test-input-image.jpg"
# 获得文件名列表
filename_queue = tf.train.string_input_producer(
tf.train.match_filenames_once(image_filename))
filename_queue1 = tf.train.string_input_producer(
tf.train.match_filenames_once(image_filename))
# 生成文件名队列
image_reader = tf.WholeFileReader()
image_reader1 = tf.WholeFileReader()
_, image_file = image_reader.read(filename_queue)
_, image_file1 = image_reader1.read(filename_queue1)
# 通过阅读器返回一个键值对,其中value表示图像
image = tf.image.decode_jpeg(image_file)
image1 = tf.image.decode_jpeg(image_file1)
# 通过tf.image.decode_jpeg解码函数对图片进行解码,得到图像.
sess = tf.Session()
init_op = tf.group(tf.global_variables_initializer(),
tf.local_variables_initializer())
init_op1 = tf.group(tf.global_variables_initializer(),
tf.local_variables_initializer())
sess.run(init_op)
coord = tf.train.Coordinator()
threads = tf.train.start_queue_runners(coord=coord, sess=sess)
coord1 = tf.train.Coordinator()
print('the image is:', sess.run(image))
filename_queue.close(cancel_pending_enqueues=True)
# filename_queue1.close(cancel_pending_enqueue=True)
coord.request_stop()
#!/usr/bin/env python
# -*- coding: utf-8 -*-
# @Author: Lenovo
# @Date: 2019-05-25 11:18:55
# @Last Modified by: zy19950209
# @Last Modified time: 2019-05-25 13:08:22
import tensorflow as tf
with tf.Session() as sess:
filename = [
'./mnist_data/A.jpg', './mnist_data/B.jpg', './mnist_data/C.jpg'
]
filename_queue = tf.train.string_input_producer(filename,
shuffle=False,
num_epochs=5)
reader = tf.WholeFileReader() # 读取所有数据
key, value = reader.read(filename)
tf.local_variables_initializer().run()
threads = tf.train.start_queue_runners(sess=sess)
i = 0
while True:
i = i + 1
image_data = sess.run(value)
with open('read/test_%d.jpg' % i, 'wb') as f:
f.write(image_data)
def main(unused_argv):
tf.logging.set_verbosity(tf.logging.INFO)
# Read list of images.
tf.logging.info('Reading list of images...')
image_paths = _ReadImageList(cmd_args.list_images_path, cmd_args.prefix)
num_images = len(image_paths)
tf.logging.info('done! Found %d images', num_images)
# Parse DelfConfig proto.
config = delf_config_pb2.DelfConfig()
with tf.gfile.FastGFile(cmd_args.config_path, 'r') as f:
text_format.Merge(f.read(), config)
# Tell TensorFlow that the model will be built into the default Graph.
with tf.Graph().as_default():
# Reading list of images.
filename_queue = tf.train.string_input_producer(image_paths, shuffle=False)
reader = tf.WholeFileReader()
_, value = reader.read(filename_queue)
image_tf = tf.image.decode_jpeg(value, channels=3)
image_tf = tf.image.resize(image_tf, (448, 448))
with tf.Session() as sess:
init_op = tf.global_variables_initializer()
sess.run(init_op)
extractor_fn = extractor.MakeExtractor(sess, config)
# Start input enqueue threads.
coord = tf.train.Coordinator()
threads = tf.train.start_queue_runners(sess=sess, coord=coord)
start = time.clock()
for i in range(num_images):
# Write to log-info once in a while.
if i == 0:
tf.logging.info('Starting to extract DELF features from images...')
elif i % _STATUS_CHECK_ITERATIONS == 0:
elapsed = (time.clock() - start)
tf.logging.info(
'Processing image %d out of %d, last %d '
'images took %f seconds', i, num_images, _STATUS_CHECK_ITERATIONS,
elapsed)
start = time.clock()
# # Get next image.
im = sess.run(image_tf)
# If descriptor already exists, skip its computation.
save_path = image_paths[i]
save_path = save_path.replace('undist_images', 'reg_feat')
save_path = save_path.replace('.jpg', '.bin')
if cmd_args.skip_extracted and tf.gfile.Exists(save_path):
tf.logging.info('Skipping %s', image_paths[i])
continue
# Extract and save features.
reg_feat_out = extractor_fn(im)
reg_feat_out = np.squeeze(reg_feat_out, axis=0)
dir_name = os.path.dirname(save_path)
if not os.path.exists(dir_name):
os.mkdir(dir_name)
reg_feat_out.astype(np.float32).tofile(save_path)
# Finalize enqueue threads.
coord.request_stop()
coord.join(threads)
def data_loader(FLAGS):
with tf.device('/cpu:0'):
# Define the returned data batches
Data = collections.namedtuple(
'Data',
'paths_LR, paths_HR, inputs, targets, image_count, steps_per_epoch'
)
#Check the input directory
if (FLAGS.input_dir_LR == 'None') or (FLAGS.input_dir_HR == 'None'):
raise ValueError('Input directory is not provided')
if (not os.path.exists(FLAGS.input_dir_LR)) or (not os.path.exists(
FLAGS.input_dir_HR)):
raise ValueError('Input directory not found')
image_list_LR = os.listdir(FLAGS.input_dir_LR)
image_list_LR = [_ for _ in image_list_LR if _.endswith('.png')]
if len(image_list_LR) == 0:
raise Exception('No png files in the input directory')
image_list_LR_temp = sorted(image_list_LR)
image_list_LR = [
os.path.join(FLAGS.input_dir_LR, _) for _ in image_list_LR_temp
]
image_list_HR = [
os.path.join(FLAGS.input_dir_HR, _) for _ in image_list_LR_temp
]
image_list_LR_tensor = tf.convert_to_tensor(image_list_LR,
dtype=tf.string)
image_list_HR_tensor = tf.convert_to_tensor(image_list_HR,
dtype=tf.string)
with tf.variable_scope('load_image'):
# define the image list queue
# image_list_LR_queue = tf.train.string_input_producer(image_list_LR, shuffle=False, capacity=FLAGS.name_queue_capacity)
# image_list_HR_queue = tf.train.string_input_producer(image_list_HR, shuffle=False, capacity=FLAGS.name_queue_capacity)
#print('[Queue] image list queue use shuffle: %s'%(FLAGS.mode == 'Train'))
output = tf.train.slice_input_producer(
[image_list_LR_tensor, image_list_HR_tensor],
shuffle=False,
capacity=FLAGS.name_queue_capacity)
# Reading and decode the images
reader = tf.WholeFileReader(name='image_reader')
image_LR = tf.read_file(output[0])
image_HR = tf.read_file(output[1])
input_image_LR = tf.image.decode_png(image_LR, channels=3)
input_image_HR = tf.image.decode_png(image_HR, channels=3)
input_image_LR = tf.image.convert_image_dtype(input_image_LR,
dtype=tf.float32)
input_image_HR = tf.image.convert_image_dtype(input_image_HR,
dtype=tf.float32)
assertion = tf.assert_equal(
tf.shape(input_image_LR)[2],
3,
message="image does not have 3 channels")
with tf.control_dependencies([assertion]):
input_image_LR = tf.identity(input_image_LR)
input_image_HR = tf.identity(input_image_HR)
# Normalize the low resolution image to [0, 1], high resolution to [-1, 1]
a_image = preprocessLR(input_image_LR)
b_image = preprocess(input_image_HR)
inputs, targets = [a_image, b_image]
# The data augmentation part
with tf.name_scope('data_preprocessing'):
with tf.name_scope('random_crop'):
# Check whether perform crop
if (FLAGS.random_crop is True) and FLAGS.mode == 'train':
print('[Config] Use random crop')
# Set the shape of the input image. the target will have 4X size
input_size = tf.shape(inputs)
target_size = tf.shape(targets)
offset_w = tf.cast(tf.floor(
tf.random_uniform([], 0,
tf.cast(input_size[1], tf.float32) -
FLAGS.crop_size)),
dtype=tf.int32)
offset_h = tf.cast(tf.floor(
tf.random_uniform([], 0,
tf.cast(input_size[0], tf.float32) -
FLAGS.crop_size)),
dtype=tf.int32)
if FLAGS.task == 'SRGAN' or FLAGS.task == 'SRResnet':
inputs = tf.image.crop_to_bounding_box(
inputs, offset_h, offset_w, FLAGS.crop_size,
FLAGS.crop_size)
targets = tf.image.crop_to_bounding_box(
targets, offset_h * 4, offset_w * 4,
FLAGS.crop_size * 4, FLAGS.crop_size * 4)
elif FLAGS.task == 'denoise':
inputs = tf.image.crop_to_bounding_box(
inputs, offset_h, offset_w, FLAGS.crop_size,
FLAGS.crop_size)
targets = tf.image.crop_to_bounding_box(
targets, offset_h, offset_w, FLAGS.crop_size,
FLAGS.crop_size)
# Do not perform crop
else:
inputs = tf.identity(inputs)
targets = tf.identity(targets)
with tf.variable_scope('random_flip'):
# Check for random flip:
if (FLAGS.flip is True) and (FLAGS.mode == 'train'):
print('[Config] Use random flip')
# Produce the decision of random flip
decision = tf.random_uniform([], 0, 1, dtype=tf.float32)
input_images = random_flip(inputs, decision)
target_images = random_flip(targets, decision)
else:
input_images = tf.identity(inputs)
target_images = tf.identity(targets)
if FLAGS.task == 'SRGAN' or FLAGS.task == 'SRResnet':
input_images.set_shape([FLAGS.crop_size, FLAGS.crop_size, 3])
target_images.set_shape(
[FLAGS.crop_size * 4, FLAGS.crop_size * 4, 3])
elif FLAGS.task == 'denoise':
input_images.set_shape([FLAGS.crop_size, FLAGS.crop_size, 3])
target_images.set_shape([FLAGS.crop_size, FLAGS.crop_size, 3])
if FLAGS.mode == 'train':
paths_LR_batch, paths_HR_batch, inputs_batch, targets_batch = tf.train.shuffle_batch(
[output[0], output[1], input_images, target_images],
batch_size=FLAGS.batch_size,
capacity=FLAGS.image_queue_capacity + 4 * FLAGS.batch_size,
min_after_dequeue=FLAGS.image_queue_capacity,
num_threads=FLAGS.queue_thread)
else:
paths_LR_batch, paths_HR_batch, inputs_batch, targets_batch = tf.train.batch(
[output[0], output[1], input_images, target_images],
batch_size=FLAGS.batch_size,
num_threads=FLAGS.queue_thread,
allow_smaller_final_batch=True)
steps_per_epoch = int(math.ceil(len(image_list_LR) / FLAGS.batch_size))
if FLAGS.task == 'SRGAN' or FLAGS.task == 'SRResnet':
inputs_batch.set_shape(
[FLAGS.batch_size, FLAGS.crop_size, FLAGS.crop_size, 3])
targets_batch.set_shape([
FLAGS.batch_size, FLAGS.crop_size * 4, FLAGS.crop_size * 4, 3
])
elif FLAGS.task == 'denoise':
inputs_batch.set_shape(
[FLAGS.batch_size, FLAGS.crop_size, FLAGS.crop_size, 3])
targets_batch.set_shape(
[FLAGS.batch_size, FLAGS.crop_size, FLAGS.crop_size, 3])
return Data(paths_LR=paths_LR_batch,
paths_HR=paths_HR_batch,
inputs=inputs_batch,
targets=targets_batch,
image_count=len(image_list_LR),
steps_per_epoch=steps_per_epoch)
def load_examples():
if a.input_dir is None or not os.path.exists(a.input_dir):
raise Exception("input_dir does not exist")
input_paths = glob.glob(os.path.join(a.input_dir, "*.jpg"))
decode = tf.image.decode_jpeg
if len(input_paths) == 0:
input_paths = glob.glob(os.path.join(a.input_dir, "*.png"))
decode = tf.image.decode_png
if len(input_paths) == 0:
raise Exception("input_dir contains no image files")
def get_name(path):
name, _ = os.path.splitext(os.path.basename(path))
return name
# if the image names are numbers, sort by the value rather than asciibetically
# having sorted inputs means that the outputs are sorted in test mode
if all(get_name(path).isdigit() for path in input_paths):
input_paths = sorted(input_paths, key=lambda path: int(get_name(path)))
else:
input_paths = sorted(input_paths)
with tf.name_scope("load_images"):
path_queue = tf.train.string_input_producer(input_paths, shuffle=a.mode == "train")
reader = tf.WholeFileReader()
paths, contents = reader.read(path_queue)
raw_input = decode(contents)
raw_input = tf.image.convert_image_dtype(raw_input, dtype=tf.float32)
assertion = tf.assert_equal(tf.shape(raw_input)[2], 3, message="image does not have 3 channels")
with tf.control_dependencies([assertion]):
raw_input = tf.identity(raw_input)
raw_input.set_shape([None, None, 3])
if a.lab_colorization:
# load color and brightness from image, no B image exists here
lab = rgb_to_lab(raw_input)
L_chan, a_chan, b_chan = preprocess_lab(lab)
a_images = tf.expand_dims(L_chan, axis=2)
b_images = tf.stack([a_chan, b_chan], axis=2)
else:
# break apart image pair and move to range [-1, 1]
width = tf.shape(raw_input)[1] # [height, width, channels]
a_images = preprocess(raw_input[:,:width//2,:])
b_images = preprocess(raw_input[:,width//2:,:])
if a.which_direction == "AtoB":
inputs, targets = [a_images, b_images]
elif a.which_direction == "BtoA":
inputs, targets = [b_images, a_images]
else:
raise Exception("invalid direction")
# synchronize seed for image operations so that we do the same operations to both
# input and output images
seed = random.randint(0, 2**31 - 1)
def transform(image):
r = image
if a.flip:
r = tf.image.random_flip_left_right(r, seed=seed)
# area produces a nice downscaling, but does nearest neighbor for upscaling
# assume we're going to be doing downscaling here
r = tf.image.resize_images(r, [a.scale_size, a.scale_size], method=tf.image.ResizeMethod.AREA)
offset = tf.cast(tf.floor(tf.random_uniform([2], 0, a.scale_size - CROP_SIZE + 1, seed=seed)), dtype=tf.int32)
if a.scale_size > CROP_SIZE:
r = tf.image.crop_to_bounding_box(r, offset[0], offset[1], CROP_SIZE, CROP_SIZE)
elif a.scale_size < CROP_SIZE:
raise Exception("scale size cannot be less than crop size")
return r
with tf.name_scope("input_images"):
input_images = transform(inputs)
with tf.name_scope("target_images"):
target_images = transform(targets)
classes = []
for xy in range(0, len(input_paths)):
classes.append(input_paths[xy][len(input_paths[xy]) - 6:len(input_paths[xy]) - 5])
to_categorical = tf.keras.utils.to_categorical
classes_hot = to_categorical(classes, num_classes=2)
classes_queue = tf.train.input_producer(classes_hot, shuffle=a.mode == "train")
paths_batch, classes_batch, inputs_batch, targets_batch = tf.train.batch([paths, classes_queue.dequeue(), input_images, target_images], batch_size=a.batch_size)
steps_per_epoch = int(math.ceil(len(input_paths) / a.batch_size))
return Examples(
paths=paths_batch,
classes=classes_batch,
inputs=inputs_batch,
targets=targets_batch,
count=len(input_paths),
steps_per_epoch=steps_per_epoch,
)
def image_input_fn(image_files):
filename_queue = tf.train.string_input_producer(image_files, shuffle=False)
reader = tf.WholeFileReader()
_, value = reader.read(filename_queue)
image_tf = tf.image.decode_jpeg(value, channels=3)
return tf.image.convert_image_dtype(image_tf, tf.float32)
def make_batch_tensor(self, network_config, restore_iter=0, train=True):
TEST_INTERVAL = 500
batch_image_size = (self.update_batch_size +
self.test_batch_size) * self.meta_batch_size
if train:
all_filenames = self.all_training_filenames
if restore_iter > 0:
all_filenames = all_filenames[batch_image_size *
(restore_iter + 1):]
else:
all_filenames = self.all_val_filenames
if restore_iter > 0:
all_filenames = all_filenames[batch_image_size * (
int(restore_iter / TEST_INTERVAL) + 1):]
im_height = network_config['image_height']
im_width = network_config['image_width']
num_channels = network_config['image_channels']
# make queue for tensorflow to read from
filename_queue = tf.train.string_input_producer(
tf.convert_to_tensor(all_filenames), shuffle=False)
print 'Generating image processing ops'
image_reader = tf.WholeFileReader()
_, image_file = image_reader.read(filename_queue)
image = tf.image.decode_gif(image_file)
# should be T x C x W x H
image.set_shape((self.T, im_height, im_width, num_channels))
image = tf.cast(image, tf.float32)
image /= 255.0
if FLAGS.hsv:
eps_min, eps_max = 0.5, 1.5
assert eps_max >= eps_min >= 0
# convert to HSV only fine if input images in [0, 1]
img_hsv = tf.image.rgb_to_hsv(image)
img_h = img_hsv[..., 0]
img_s = img_hsv[..., 1]
img_v = img_hsv[..., 2]
eps = tf.random_uniform([self.T, 1, 1], eps_min, eps_max)
img_v = tf.clip_by_value(eps * img_v, 0., 1.)
img_hsv = tf.stack([img_h, img_s, img_v], 3)
image_rgb = tf.image.hsv_to_rgb(img_hsv)
image = image_rgb
print('image.shape', image.shape)
image = tf.transpose(image,
perm=[0, 3, 2, 1
]) # transpose to mujoco setting for images
print('image.reshape', image.shape)
image = tf.reshape(image, [self.T, -1])
num_preprocess_threads = 1 # TODO - enable this to be set to >1
min_queue_examples = 64 #128 #256
print 'Batching images'
images = tf.train.batch(
[image],
batch_size=batch_image_size,
num_threads=num_preprocess_threads,
capacity=min_queue_examples + 3 * batch_image_size,
)
all_images = []
for i in xrange(self.meta_batch_size):
image = images[i * (self.update_batch_size +
self.test_batch_size):(i + 1) *
(self.update_batch_size + self.test_batch_size)]
# print('i',i,'image.shape',image.shape)
image = tf.reshape(
image,
[(self.update_batch_size + self.test_batch_size) * self.T, -1])
# print('append', 'image.shape', image.shape)
all_images.append(image)
print('make_batch_tensor', tf.stack(all_images).shape)
return tf.stack(all_images)
def ExampleGen(data_path, num_epochs=None):
"""Generates tf.Examples from path of data files.
ExampleGen
Binary data format: <length><blob>. <length> represents the byte size
of <blob>. <blob> is serialized tf.Example proto. The tf.Example contains
the tokenized article text and summary.
Args:
data_path: path to tf.Example data files.
num_epochs: Number of times to go through the data. None means infinite.
Yields:
Deserialized tf.Example.
If there are multiple files specified, they accessed in a random order.
"""
# epoch = 0
# while True:
# fileselection_path = data_path + "/*.json"
# filenames = tf.train.match_filenames_once(fileselection_path)
# count_num_files = tf.size(filenames)
# filename_queue = tf.train.string_input_producer(filenames)
# reader = tf.WholeFileReader()
# filename, _ = reader.read(filename_queue)
# counter = 0
# with tf.Session() as sess:
# sess.run([tf.global_variables_initializer(), tf.local_variables_initializer()])
# num_files = sess.run(count_num_files)
# if counter >= num_files:
# break
# coord = tf.train.Coordinator()
# threads = tf.train.start_queue_runners(coord=coord)
#
# for i in range(num_files):
# csv_file = sess.run(filename)
# tf_example = example_pb2.Example()
# filename_queue_sub = tf.train.string_input_producer([csv_file])
# reader_sub = tf.WholeFileReader()
# _, file_contents_sub = reader_sub.read(filename_queue_sub)
#
# threads_sub = tf.train.start_queue_runners(coord=coord)
# json_content = sess.run(file_contents_sub)
# json_text = json.loads(json_content)
# article = str(json_text['article'])
# summary = str(json_text["summary"])
# tf_example.features.feature['article'].bytes_list.value.extend([article])
# tf_example.features.feature['abstract'].bytes_list.value.extend([summary])
# counter += 1
# yield tf_example
#
# coord.request_stop()
# coord.join(threads)
# coord.join(threads_sub)
# epoch += 1
epoch = 0
filenames = tf.train.match_filenames_once(data_path +"/*.json")
count_num_files = tf.size(filenames)
filename_queue = tf.train.string_input_producer(filenames)
reader = tf.WholeFileReader()
filename, _ = reader.read(filename_queue)
with tf.Session() as sess:
sess.run([tf.global_variables_initializer(), tf.local_variables_initializer()])
num_files = sess.run(count_num_files)
counter =0
while True:
if counter >= num_files:
break
coord = tf.train.Coordinator()
threads = tf.train.start_queue_runners(coord=coord)
for i in range(num_files):
csv_file = sess.run(filename)
tf_example = example_pb2.Example()
filename_queue_sub = tf.train.string_input_producer([csv_file])
reader_sub = tf.WholeFileReader()
_, file_contents_sub = reader_sub.read(filename_queue_sub)
threads_sub = tf.train.start_queue_runners(coord=coord)
json_content = sess.run(file_contents_sub)
json_text = json.loads(json.dumps( ast.literal_eval(json_content)))
article = str(json_text.get('article').encode('ascii','ignore'))
summary = str(json_text.get("summary").encode('ascii','ignore'))
tf_example.features.feature['article'].bytes_list.value.extend([article])
tf_example.features.feature['abstract'].bytes_list.value.extend([summary])
counter = counter +1
yield(tf_example)
coord.request_stop()
coord.join(threads)
coord.join(threads_sub)
epoch += 1