def input_pipeline(file_pattern, mode, capacity=64):
keys_to_features = {
"source": tf.VarLenFeature(tf.int64),
"target": tf.VarLenFeature(tf.int64),
"source_length": tf.FixedLenFeature([1], tf.int64),
"target_length": tf.FixedLenFeature([1], tf.int64)
}
items_to_handlers = {
"source": tfexample_decoder.Tensor("source"),
"target": tfexample_decoder.Tensor("target"),
"source_length": tfexample_decoder.Tensor("source_length"),
"target_length": tfexample_decoder.Tensor("target_length")
}
# Now the non-trivial case construction.
with tf.name_scope("examples_queue"):
training = (mode == "train")
# Read serialized examples using slim parallel_reader.
num_epochs = None if training else 1
data_files = parallel_reader.get_data_files(file_pattern)
num_readers = min(4 if training else 1, len(data_files))
_, examples = parallel_reader.parallel_read([file_pattern],
tf.TFRecordReader,
num_epochs=num_epochs,
shuffle=training,
capacity=2 * capacity,
min_after_dequeue=capacity,
num_readers=num_readers)
decoder = tfexample_decoder.TFExampleDecoder(keys_to_features,
items_to_handlers)
decoded = decoder.decode(examples, items=list(items_to_handlers))
examples = {}
for (field, tensor) in zip(keys_to_features, decoded):
examples[field] = tensor
# We do not want int64s as they do are not supported on GPUs.
return {k: tf.to_int32(v) for (k, v) in six.iteritems(examples)}
def get_observation():
with tf.Session() as sess:
_, string_tensor = parallel_reader.parallel_read(
[DATA_PATH],
reader_class=tf.TFRecordReader,
num_epochs=1,
num_readers=1,
shuffle=True,
dtypes=[tf.string, tf.string],
capacity=500,
min_after_dequeue=200)
decoder = tf_example_decoder.TfExampleDecoder(
load_instance_masks=True,
instance_mask_type=input_reader_pb2.PNG_MASKS)
decoded_data = decoder.decode(string_tensor)
image = decoded_data['image']
box_list = decoded_data['groundtruth_boxes']
class_list = decoded_data['groundtruth_classes']
mask_list = decoded_data['groundtruth_instance_masks']
file_name = decoded_data['filename']
init_op = tf.group(tf.global_variables_initializer(),
tf.local_variables_initializer(),
tf.tables_initializer())
sess.run(init_op)
coord = tf.train.Coordinator()
threads = tf.train.start_queue_runners(coord=coord)
while True:
img, boxes, classes, masks, file_names = sess.run(
[image, box_list, class_list, mask_list, file_name])
print("current file {} contains {} masks".format(
file_names.decode(), len(boxes)))
sys.stdout.flush()
# if masks.shape[0] > 5:
# return masks
image_with_labels(img,
boxes,
masks,
classes,
label_dict=LABEL_DICT)
plt.tight_layout()
# if AUTO:
# plt.pause(0.5)
# elif plt.waitforbuttonpress():
# break
plt.waitforbuttonpress()
plt.close()
coord.request_stop()
coord.join(threads)
return 1
