machine-learning - tensorflow 错误: FailedPeconditionError: attempting to use uninitialized variable

Question

我想将立体图像放入优化器中。这是我的代码:

tf.reset_default_graph()

# config
learning_rate = 0.5
training_epochs = 5

# init
init = tf.global_variables_initializer()

def conv2d(input_layer):
    conv1 = tf.layers.conv2d(
        inputs=input_layer,
        filters=32,
        kernel_size=[3, 3],
        padding='same',
        activation=tf.tanh,
        use_bias=False
    )

    conv2 = tf.layers.conv2d(
        inputs=conv1,
        filters=32,
        kernel_size=[3, 3],
        padding='same',
        activation=tf.tanh,
        use_bias=False
    )

    conv3 = tf.layers.conv2d(
        inputs=conv2,
        filters=32,
        kernel_size=[3, 3],
        padding='same',
        activation=tf.tanh,
        use_bias=False
    )

    conv4 = tf.layers.conv2d(
        inputs=conv3,
        filters=32,
        kernel_size=[3, 3],
        padding='same',
        activation=tf.tanh,
        use_bias=False
    )

    logits = tf.layers.conv2d(
        inputs=conv4,
        filters=32,
        kernel_size=[3, 3],
        padding='same',
        activation=tf.sigmoid,
        use_bias=False
    )

    return logits


if __name__ == '__main__':
    # read images
    # preprocessing: rgb converted to float, zero_mean, uni_variance
    images = reading_images()
    mask_tensor = images["mask"][1]

    # reshape images
    img0 = images["img0"][1]
    img1 = images["img1"][1]
    img0_rs = tf.reshape(img0, [1, int(1988 / 2), int(2880 / 2), 3])
    img1_rs = tf.reshape(img1, [1, int(1988 / 2), int(2880 / 2), 3])

    # define symbolic placeholders
    t_im0 = tf.placeholder(tf.float32, [1, None, None, 3])
    t_im1 = tf.placeholder(tf.float32, [1, None, None, 3])
    t_img = tf.concat([t_im0, t_im1], axis=3)

    input_layer = tf.reshape(t_img, [1, int(1988 / 2), int(2880 / 2), 6])
    logits = conv2d(input_layer)

    with tf.name_scope("cost_function") as scope:
        mask_tensor = tf.tile(mask_tensor, [1, 1, 3])
        cost_function = -tf.reduce_mean(mask_tensor * tf.log(logits) + (1. - mask_tensor) * tf.log(1. - logits))
    tf.summary.scalar("cost_function", cost_function)

    with tf.name_scope("train") as scope:
        optimizer = tf.train.GradientDescentOptimizer(learning_rate).minimize(cost_function)
    merged_summary_op = tf.summary.merge_all()

    with tf.Session() as sess:
        coord = tf.train.Coordinator()
        threads = tf.train.start_queue_runners(sess=sess, coord=coord)

        sess.run(init)
        # summary_writer = tf.summary.FileWriter('/tmp/tensorflow_logs', graph=sess.graph)
        for epoch in range(training_epochs):
            print("epoch ", epoch)

            avg_cost = 0.0
            mask = sess.run(mask_tensor)
            np_img0_rs = sess.run(img0_rs)
            np_img1_rs = sess.run(img1_rs)

            # res = t_img.eval(feed_dict={t_im0: img0_rs_, t_im1: img1_rs_})

            sess.run([optimizer], feed_dict={t_im0: np_img0_rs, t_im1: np_img1_rs})
    coord.request_stop()
    coord.join(threads)

但我总是收到这个错误。我不知道我必须改变什么。我可以尝试什么来调试它？我真的做了很多尝试来修复这个错误。

epoch  0
2017-07-17 10:26:03.719539: W tensorflow/core/kernels/queue_base.cc:294] _4_input_producer: Skipping cancelled enqueue attempt with queue not closed
2017-07-17 10:26:03.719610: W tensorflow/core/kernels/queue_base.cc:294] _5_input_producer_1: Skipping cancelled enqueue attempt with queue not closed
Traceback (most recent call last):
  File "/home/test/Dropbox/occlusion_thesis/occ_small/main.py", line 111, in <module>
    sess.run([optimizer], feed_dict={t_im0: np_img0_rs, t_im1: np_img1_rs})
  File "/home/test/Programs/anaconda2/lib/python2.7/site-packages/tensorflow/python/client/session.py", line 789, in run
    run_metadata_ptr)
  File "/home/test/Programs/anaconda2/lib/python2.7/site-packages/tensorflow/python/client/session.py", line 997, in _run
    feed_dict_string, options, run_metadata)
  File "/home/test/Programs/anaconda2/lib/python2.7/site-packages/tensorflow/python/client/session.py", line 1132, in _do_run
    target_list, options, run_metadata)
  File "/home/test/Programs/anaconda2/lib/python2.7/site-packages/tensorflow/python/client/session.py", line 1152, in _do_call
    raise type(e)(node_def, op, message)
tensorflow.python.framework.errors_impl.FailedPreconditionError: Attempting to use uninitialized value conv2d_4/kernel
     [[Node: conv2d_4/kernel/read = Identity[T=DT_FLOAT, _class=["loc:@conv2d_4/kernel"], _device="/job:localhost/replica:0/task:0/cpu:0"](conv2d_4/kernel)]]

Caused by op u'conv2d_4/kernel/read', defined at:
  File "/home/test/Dropbox/occlusion_thesis/occ_small/main.py", line 84, in <module>
    logits = conv2d(input_layer)
  File "/home/test/Dropbox/occlusion_thesis/occ_small/main.py", line 60, in conv2d
    use_bias=False
  File "/home/test/Programs/anaconda2/lib/python2.7/site-packages/tensorflow/python/layers/convolutional.py", line 551, in conv2d
    return layer.apply(inputs)
  File "/home/test/Programs/anaconda2/lib/python2.7/site-packages/tensorflow/python/layers/base.py", line 492, in apply
    return self.__call__(inputs, *args, **kwargs)
  File "/home/test/Programs/anaconda2/lib/python2.7/site-packages/tensorflow/python/layers/base.py", line 434, in __call__
    self.build(input_shapes[0])
  File "/home/test/Programs/anaconda2/lib/python2.7/site-packages/tensorflow/python/layers/convolutional.py", line 137, in build
    dtype=self.dtype)
  File "/home/test/Programs/anaconda2/lib/python2.7/site-packages/tensorflow/python/layers/base.py", line 374, in add_variable
    trainable=trainable and self.trainable)
  File "/home/test/Programs/anaconda2/lib/python2.7/site-packages/tensorflow/python/ops/variable_scope.py", line 1065, in get_variable
    use_resource=use_resource, custom_getter=custom_getter)
  File "/home/test/Programs/anaconda2/lib/python2.7/site-packages/tensorflow/python/ops/variable_scope.py", line 962, in get_variable
    use_resource=use_resource, custom_getter=custom_getter)
  File "/home/test/Programs/anaconda2/lib/python2.7/site-packages/tensorflow/python/ops/variable_scope.py", line 367, in get_variable
    validate_shape=validate_shape, use_resource=use_resource)
  File "/home/test/Programs/anaconda2/lib/python2.7/site-packages/tensorflow/python/ops/variable_scope.py", line 352, in _true_getter
    use_resource=use_resource)
  File "/home/test/Programs/anaconda2/lib/python2.7/site-packages/tensorflow/python/ops/variable_scope.py", line 725, in _get_single_variable
    validate_shape=validate_shape)
  File "/home/test/Programs/anaconda2/lib/python2.7/site-packages/tensorflow/python/ops/variables.py", line 200, in __init__
    expected_shape=expected_shape)
  File "/home/test/Programs/anaconda2/lib/python2.7/site-packages/tensorflow/python/ops/variables.py", line 319, in _init_from_args
    self._snapshot = array_ops.identity(self._variable, name="read")
  File "/home/test/Programs/anaconda2/lib/python2.7/site-packages/tensorflow/python/ops/gen_array_ops.py", line 1303, in identity
    result = _op_def_lib.apply_op("Identity", input=input, name=name)
  File "/home/test/Programs/anaconda2/lib/python2.7/site-packages/tensorflow/python/framework/op_def_library.py", line 767, in apply_op
    op_def=op_def)
  File "/home/test/Programs/anaconda2/lib/python2.7/site-packages/tensorflow/python/framework/ops.py", line 2506, in create_op
    original_op=self._default_original_op, op_def=op_def)
  File "/home/test/Programs/anaconda2/lib/python2.7/site-packages/tensorflow/python/framework/ops.py", line 1269, in __init__
    self._traceback = _extract_stack()

FailedPreconditionError (see above for traceback): Attempting to use uninitialized value conv2d_4/kernel
     [[Node: conv2d_4/kernel/read = Identity[T=DT_FLOAT, _class=["loc:@conv2d_4/kernel"], _device="/job:localhost/replica:0/task:0/cpu:0"](conv2d_4/kernel)]]

Answer 1

我不确定您的代码是否完整，但错误消息对我来说似乎很清楚:

FailedPreconditionError: Attempting to use uninitialized value conv2d_4/kernel

查看您的代码，我发现您有 sess.run(init) 但我在任何地方都找不到 init 的定义。尝试在 with tf.Session() as sess: 之前添加 init = tf.global_variables_initializer()，这应该可以修复“未初始化值”错误。

编辑:通过完整的代码，我发现问题来自:

# init
init = tf.global_variables_initializer() # <<<<<<<<< 1

def conv2d(input_layer):
    ## Bunch of code defining layers
    return logits

if __name__ == '__main__':
    ## bunch of other code

    logits = conv2d(input_layer) # <<<<<<<<< 2

我将您为到该点定义的所有变量定义初始化函数的点标记为1，并将该点标记为2您实际定义网络(以及其中的所有变量)的地方。 init 的定义必须在完成所有变量的定义之后，否则将会有未初始化的变量。

更新:

我在这里复制我对答案所做的评论，因为它可能是一个更好的放置位置。必须在定义图表之后调用tf.global_variables_initializer()。如果您在开始时定义它，然后将层添加到网络中，添加的层的权重将不会被初始化，因为它们在您创建初始化操作时尚未定义。始终将 init 定义为 with tf.Session() ... 之前的最后一个操作，以确保您不会错过初始化中的任何内容。