我一直在创建自己的数据库来训练CNN。现在，我从tfrecord文件中读取数据时遇到问题。我已经成功保存了一个tfrecord文件，其中包含两个功能：图像和标签。当我尝试阅读它时，它只是读取第一批，然后我收到一条错误消息。
保存tfrecord文件的代码是（因为时间我只假设了5个文件）：
#SAVE TFRECORD FILE
import tensorflow as tf
import numpy as np
import Image
image_filename = [('/home/ag/Dropbox/DL/6_CNN_BD/data_resized/01GraspableGraspingRectangles_RGB/00%03d.png' % x) for x in range(1,6)]
records_filename = '/home/ag/Dropbox/DL/6_CNN_BD/data_resized/01GraspableGraspingRectangles_RGB/DS.tfrecord'
writer = tf.python_io.TFRecordWriter(records_filename)
original_images = []
for img_path in image_filename:
image = np.array(Image.open(img_path))
#img_label = 'GP'
img_label = b'\x01'
img_raw = image.tostring()
example = tf.train.Example(features=tf.train.Features(feature={
'image_raw': tf.train.Feature(bytes_list = tf.train.BytesList(value = [img_raw])),
'label': tf.train.Feature(bytes_list = tf.train.BytesList(value = [img_label])),
}))
writer.write(example.SerializeToString())
writer.close()
读取tfrecord文件的代码是：
#READ TFRECORD FILE
import tensorflow as tf
import skimage.io as io
IMAGE_HEIGHT = 24
IMAGE_WIDTH = 24
IMAGE_CHANNELS = 3
BATCH_SIZE = 2
tfrecords_filename = '/home/ag/Dropbox/DL/6_CNN_BD/data_resized/01GraspableGraspingRectangles_RGB/DS.tfrecord'
def read_and_decode(filename_queue):
reader = tf.TFRecordReader()
_, serialized_example = reader.read(filename_queue)
features = tf.parse_single_example(
serialized_example, features={
'image_raw': tf.FixedLenFeature([], tf.string),
'label': tf.FixedLenFeature([], tf.string),
})
image = tf.decode_raw(features['image_raw'], tf.uint8)
image_reshape = tf.reshape(image, [IMAGE_HEIGHT, IMAGE_WIDTH, IMAGE_CHANNELS])
label = tf.cast(features['label'], tf.string)
label_reshape = label
images, label = tf.train.shuffle_batch([image_reshape, label_reshape],
batch_size = 2,
capacity = 30,
num_threads = 2,
min_after_dequeue = 10)
return images, label
filename_queue = tf.train.string_input_producer([tfrecords_filename], num_epochs=10)
image, label = read_and_decode(filename_queue)
init_op = tf.group(tf.global_variables_initializer(), tf.local_variables_initializer())
with tf.Session() as sess:
sess.run(init_op)
coord = tf.train.Coordinator()
threads = tf.train.start_queue_runners(coord = coord)
for i in range(5):
img, label = sess.run([image, label])
print(img.shape)
print(label)
print('current batch')
io.imshow(img[0, :, :, :])
io.show()
io.imshow(img[1, :, :, :])
io.show()
coord.request_stop()
coord.join(threads)
值得一提的是，如果我为img, label = sess.run([image, label])更改img = sess.run(image)，我就没有错误。这让我觉得这个问题与标签功能的格式有关。
错误屏幕类似于：
>>>
(2, 24, 24, 3)
['\x01' '\x01']
current batch
Traceback (most recent call last):
File "/home/ag/Dropbox/DL/6_CNN_BD/DS2.py", line 52, in <module>
img, label = sess.run([image, label])
File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/client/session.py", line 767, in run
run_metadata_ptr)
File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/client/session.py", line 952, in _run
fetch_handler = _FetchHandler(self._graph, fetches, feed_dict_string)
File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/client/session.py", line 408, in __init__
self._fetch_mapper = _FetchMapper.for_fetch(fetches)
File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/client/session.py", line 230, in for_fetch
return _ListFetchMapper(fetch)
File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/client/session.py", line 337, in __init__
self._mappers = [_FetchMapper.for_fetch(fetch) for fetch in fetches]
File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/client/session.py", line 238, in for_fetch
return _ElementFetchMapper(fetches, contraction_fn)
File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/client/session.py", line 271, in __init__
% (fetch, type(fetch), str(e)))
TypeError: Fetch argument array(['\x01', '\x01'], dtype=object) has invalid type <type 'numpy.ndarray'>, must be a string or Tensor. (Can not convert a ndarray into a Tensor or Operation.)
我尝试过不同的方法但没有成功。对这个问题的任何建议？
答案 0 :(得分：0)
这是代码，有一些小改动有助于解决问题。
#READ TFRECORD FILE
import tensorflow as tf
import skimage.io as io
import Image
IMAGE_HEIGHT = 24
IMAGE_WIDTH = 24
IMAGE_CHANNELS = 3
BATCH_SIZE = 5
MIN_AFTER_DEQUEUE = 10000
CAPACITY = MIN_AFTER_DEQUEUE+3*BATCH_SIZE
NUM_THREADS = 2
tfrecords_filename = ['/home/ag/Dropbox/DL/6_CNN_BD/data_resized/TFrecords/DS01.tfrecord', '/home/ag/Dropbox/DL/6_CNN_BD/data_resized/TFrecords/DS02.tfrecord']
def read_and_decode(filename_queue):
reader = tf.TFRecordReader()
_, serialized_example = reader.read(filename_queue)
features = tf.parse_single_example(
serialized_example, features={
'image_raw': tf.FixedLenFeature([], tf.string),
'label': tf.FixedLenFeature([], tf.string),
})
image = tf.decode_raw(features['image_raw'], tf.uint8)
image_reshape = tf.reshape(image, [IMAGE_HEIGHT, IMAGE_WIDTH, IMAGE_CHANNELS])
label = tf.cast(features['label'], tf.string)
label_reshape = label
images, label = tf.train.shuffle_batch([image_reshape, label_reshape],
batch_size = BATCH_SIZE,
capacity = CAPACITY,
num_threads = NUM_THREADS,
min_after_dequeue = MIN_AFTER_DEQUEUE)
#images, label = tf.train.batch([image_reshape, label_reshape], batch_size = 2, capacity = 30, num_threads = 2, min_after_dequeue = 10)
return images, label
#return image_reshape, label_reshape
filename_queue = tf.train.string_input_producer(tfrecords_filename, num_epochs=10)
image, label = read_and_decode(filename_queue)
init_op = tf.group(tf.global_variables_initializer(), tf.local_variables_initializer())
with tf.Session() as sess:
sess.run(init_op)
coord = tf.train.Coordinator()
threads = tf.train.start_queue_runners(coord = coord)
for i in range(1000):
img, lbl = sess.run([image, label])
print(i, img.shape, lbl)
print('current batch')
#img_save = Image.fromarray(img, 'RGB')
#img_save.save("/home/ag/Dropbox/DL/6_CNN_BD/data_resized/02GraspableGraspingRectangles_RGB/" + str(i) + "-train.png")
#io.imshow(img[0, :, :, :])
#io.show()
#io.imshow(img[1, :, :, :])
#io.show()
coord.request_stop()
coord.join(threads)