python - reshape 时的 ​​Tensorflow 尺寸错误

Question

我正在尝试调整非常简单的 MNIST 示例 ( https://github.com/tensorflow/tensorflow/blob/r1.8/tensorflow/examples/tutorials/layers/cnn_mnist.py ) 以适应我遇到的问题。问题似乎出在我的输入维度中的某个地方(我相信)

原始 MNIST 数据具有以下属性(在第 125 行插入):

print(train_data.shape) #    (55000, 784)
print(type(train_data)) #    <class 'numpy.ndarray'>
print(train_labels.shape) #  (55000, )
print(type(train_labels)) #  <class 'numpy.ndarray'>

我的数据具有以下形状

Train Data:  (10681, 9216)
Train Data:  <class 'numpy.ndarray'>
Train Labe:  (10681,)
Train Labe:  <class 'numpy.ndarray'>

特别是我们有 784=28*28 和 9216=96*96。因此，原始行 (31) 为

 input_layer = tf.reshape(features["x"], [-1, 28, 28, 1])

我把它替换为

 input_layer = tf.reshape(features["x"], [-1, 96, 96, 1])

但是当我运行这个时，我得到的错误是:

Traceback (most recent call last):
  File "/home/j-pc/.local/lib/python3.5/site-packages/tensorflow/python/framework/common_shapes.py", line686, in _call_cpp_shape_fn_impl
    input_tensors_as_shapes, status)
  File "/home/j-pc/.local/lib/python3.5/site-packages/tensorflow/python/framework/errors_impl.py", line 473, in __exit__
    c_api.TF_GetCode(self.status.status))
tensorflow.python.framework.errors_impl.InvalidArgumentError: Dimension size must be evenly divisible by 3136 but is 3686400 for 'Reshape_1' (op: 'Reshape') with input shapes: [100,24,24,64], [2] and with input tensors computed as partial shapes: input[1] = [?,3136].

During handling of the above exception, another exception occurred:
Traceback (most recent call last):
  File "tensorfuck/learn.py", line 252, in <module>
    tf.app.run()
  File "/home/j-pc/.local/lib/python3.5/site-packages/tensorflow/python/platform/app.py", line 124, in run
    _sys.exit(main(argv))
  File "tensorfuck/learn.py", line 238, in main
    hooks=[logging_hook])
  File "/home/j-pc/.local/lib/python3.5/site-packages/tensorflow/python/estimator/estimator.py", line 314, in train
    loss = self._train_model(input_fn, hooks, saving_listeners)
  File "/home/j-pc/.local/lib/python3.5/site-packages/tensorflow/python/estimator/estimator.py", line 743, in _train_model
    features, labels, model_fn_lib.ModeKeys.TRAIN, self.config)
  File "/home/j-pc/.local/lib/python3.5/site-packages/tensorflow/python/estimator/estimator.py", line 725, in _call_model_fn
    model_fn_results = self._model_fn(features=features, **kwargs)
  File "tensorfuck/learn.py", line 88, in cnn_model_fn
    pool2_flat = tf.reshape(pool2, [-1, 7 * 7 * 64])
  File "/home/j-pc/.local/lib/python3.5/site-packages/tensorflow/python/ops/gen_array_ops.py", line 3997,in reshape
    "Reshape", tensor=tensor, shape=shape, name=name)
  File "/home/j-pc/.local/lib/python3.5/site-packages/tensorflow/python/framework/op_def_library.py", line 787, in _apply_op_helper
    op_def=op_def)
  File "/home/j-pc/.local/lib/python3.5/site-packages/tensorflow/python/framework/ops.py", line 3162, in create_op
    compute_device=compute_device)
  File "/home/j-pc/.local/lib/python3.5/site-packages/tensorflow/python/framework/ops.py", line 3208, in _create_op_helper
    set_shapes_for_outputs(op)
  File "/home/j-pc/.local/lib/python3.5/site-packages/tensorflow/python/framework/ops.py", line 2427, in set_shapes_for_outputs
    return _set_shapes_for_outputs(op)
  File "/home/j-pc/.local/lib/python3.5/site-packages/tensorflow/python/framework/ops.py", line 2400, in _set_shapes_for_outputs
    shapes = shape_func(op)
  File "/home/j-pc/.local/lib/python3.5/site-packages/tensorflow/python/framework/ops.py", line 2330, in call_with_requiring
    return call_cpp_shape_fn(op, require_shape_fn=True)
  File "/home/j-pc/.local/lib/python3.5/site-packages/tensorflow/python/framework/common_shapes.py", line627, in call_cpp_shape_fn
    require_shape_fn)
  File "/home/j-pc/.local/lib/python3.5/site-packages/tensorflow/python/framework/common_shapes.py", line691, in _call_cpp_shape_fn_impl
    raise ValueError(err.message)
ValueError: Dimension size must be evenly divisible by 3136 but is 3686400 for 'Reshape_1' (op: 'Reshape') with input shapes: [100,24,24,64], [2] and with input tensors computed as partial shapes: input[1] = [?,3136].

有什么问题吗？附:我在Ubuntu16.04上使用Python3.5.2和Tensorflow1.5.0运行原始 MNIST 示例不会出现错误。

Answer 1

问题

这确实是一个维度问题。它与您的一般网络架构有关，该架构与您的新输入大小不匹配。您采用了 cnn_mnist.py 中的架构。这里第 72 行导致了这个问题:

pool2_flat = tf.reshape(pool2, [-1, 7 * 7 * 64]) // 7*7*64=3136 

该架构假设 [-1, 7, 7, 64]形状，但你正在路过[-1,24,24,64] .

如何解决

更改pool2_flat以匹配您的实际输出形状:

pool2_flat = tf.reshape(pool2, [-1, 24 * 24 * 64])