Tensorflow维度:必须完全定义所有形状：[TensorShape([Dimension(None)，Dimension(None)，Dimension(3)])，TensorShape([])

Question

我想使用批处理在解码后从文件folder.But中读取图像，这在我使用tf.train.batch时可能会有一些问题。下面是代码。

def get_batch(image, label, batch_size, capacity):

image = tf.cast(image, tf.string)
label = tf.cast(label, tf.int32)

input_queue = tf.train.slice_input_producer([image, label])

label = input_queue[1]
image_contents = tf.read_file(input_queue[0])
image = tf.image.decode_jpeg(image_contents, channels=3)
image = tf.image.per_image_whitening(image) 

image_batch, label_batch = tf.train.batch([image, label],
                                            batch_size = batch_size,
                                            num_threads = 8, 
                                            capacity = capacity)

label_batch = tf.reshape(label_batch, [batch_size])
image_batch = tf.cast(image_batch, tf.float32)

return image_batch, label_batch

错误是说我没有定义一些张量形状。我不知道如何解码我没有在正确的way.Here中使用解码是错误的。

Traceback (most recent call last):
  File "input_data.py", line 118, in <module>
    image_batch, label_batch = get_batch(image_list, label_list, BATCH_SIZE, CAPACITY)
  File "input_data.py", line 90, in get_batch
    capacity = capacity)
  File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/training/input.py", line 538, in batch
    capacity=capacity, dtypes=types, shapes=shapes, shared_name=shared_name)
  File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/ops/data_flow_ops.py", line 453, in __init__
    shapes = _as_shape_list(shapes, dtypes)
  File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/ops/data_flow_ops.py", line 71, in _as_shape_list
    raise ValueError("All shapes must be fully defined: %s" % shapes)
ValueError: All shapes must be fully defined: [TensorShape([Dimension(None), Dimension(None), Dimension(3)]), TensorShape([])]

Answer 1

要批量处理的数据必须具有预定义的形状，在您的情况下，张量image没有预定义的形状，您需要使用image.set_shape或tf.image.resize_images指定形状