TensorFlow的代碼報錯，信息如下：

Traceback (most recent call last):
  File "main3.py", line 120, in <module>
    tf.app.run()
  File "C:\Python35\lib\site-packages\tensorflow\python\platform\app.py", line 48, in run     _sys.exit(main(_sys.argv[:1] + flags_passthrough))
  File "main3.py", line 99, in main
    data_file=FLAGS.imglist_file)
  File "D:\trunk\model3.py", line 90, in __init__     self.build_model()
  File "D:\trunk\model3.py", line 124, in build_model     self.output_height, self.output_width, 3, self.is_crop)
  File "D:\trunk\utils.py", line 30, in transform_with_tf     return image/127.5 - 1
TypeError: unsupported operand type(s) for /: 'Tensor' and 'float'

查看報錯的消息可能來自于tensorflow\python\kernel_tests\matmul_op_test.py文件，看來其實是沒有實現這個除法，但真的是這樣的話。多次運行出錯的代碼，發現代碼有時候能順利運行，有時候不能運行，會上述錯誤，或者報錯如下：

Traceback (most recent call last):
  File "main3.py", line 111, in <module>
    tf.app.run()
  File "C:\Python35\lib\site-packages\tensorflow\python\platform\app.py", line 48, in run
    _sys.exit(main(_sys.argv[:1] + flags_passthrough))
  File "main3.py", line 92, in main
    dcgan.train(FLAGS)
  File "D: \trunk\model3.py", line 227, in train
    is_grayscale=self.is_grayscale
  File "D: \trunk\utils.py", line 37, in get_batch_image
    img = transform_with_tf(img,input_height,input_width,resize_height,resize_wi
dth, depth, is_crop)
  File "D: \trunk\utils.py", line 58, in transform_with_tf
    return image/127.5 - 1.
  File "F:\Program Files\Python35\lib\site-packages\tensorflow\python\ops\math_o
ps.py", line 820, in binary_op_wrapper
    y = ops.convert_to_tensor(y, dtype=x.dtype.base_dtype, name="y")
  File "F:\Program Files\Python35\lib\site-packages\tensorflow\python\framework\
ops.py", line 639, in convert_to_tensor
    as_ref=False)
  File "F:\Program Files\Python35\lib\site-packages\tensorflow\python\framework\
ops.py", line 704, in internal_convert_to_tensor
    ret = conversion_func(value, dtype=dtype, name=name, as_ref=as_ref)
  File "F:\Program Files\Python35\lib\site-packages\tensorflow\python\framework\
constant_op.py", line 113, in _constant_tensor_conversion_function
    return constant(v, dtype=dtype, name=name)
  File "F:\Program Files\Python35\lib\site-packages\tensorflow\python\framework\
constant_op.py", line 102, in constant
    tensor_util.make_tensor_proto(value, dtype=dtype, shape=shape, verify_shape=
verify_shape))
  File "F:\Program Files\Python35\lib\site-packages\tensorflow\python\framework\
tensor_util.py", line 370, in make_tensor_proto
    _AssertCompatible(values, dtype)
  File "F:\Program Files\Python35\lib\site-packages\tensorflow\python\framework\
tensor_util.py", line 302, in _AssertCompatible
    (dtype.name, repr(mismatch), type(mismatch).__name__))
TypeError: Expected uint8, got 127.5 of type 'float' instead.

檢查renturn image / 127.5 – 1的上一句是
image = tf.image.resize_images(image,[resize_height, resize_width])
查看tf.image.resize_images的實現源碼，找到下列這段：

if all(x is not None
         for x in [new_width_const, width, new_height_const, height]) and (
             width == new_width_const and height == new_height_const):
    if not is_batch:
      images = array_ops.squeeze(images, squeeze_dims=[0])
    return images

  if method == ResizeMethod.BILINEAR:
    images = gen_image_ops.resize_bilinear(images,
                                           size,
                                           align_corners=align_corners)

意思就是如果resize設置的大小與原大小一致，就直接返回原來的數據了，而如果確實要改變圖片大小，則會根據設置的參數，調用對應的改變圖片大小的算法，在使用中發現，數據如果resize_images中改變了大小，那么不會報上述的錯誤，但是沒有改變的話，就會報上述的錯誤。檢查gen_image_ops.resize_bilinear對應源碼為ResizeBilinear的操作，在resize_bilinear_op.h文件中定義如下：

template <typename Device, typename T>
    struct ResizeBilinear {
      void operator()(const Device& d, typename TTypes<T, 4>::ConstTensor images,
                      const float height_scale, const float width_scale,
                      typename TTypes<float, 4>::Tensor resized_images);
    };

很顯然，這里resized_images變量被轉換成了float的類型，而圖片一般的數據類型是uint8類型，這里在處理圖片的時候，進行了隱性轉換，這就導致了代碼同樣的數據，不同參數調用會出現報錯與不報錯的區別。
解決的辦法其實很簡單：
在return image/127.5 -1之前加上顯式的類型轉換tf.cast即可，例如
image = tf.cast(image, tf.float32)
最后吐槽下：
1.同樣的函數，對于數據沒有做統一的格式處理，這里是不是算一個BUG，因此要盡量的減少隱性的數據轉換。
2.報錯信息的問題，報錯是不支持的Tensor和float類型的除法操作，其實際上是不支持uint8類型的Tensor與float類型的數據的除法操作（將上述報錯的內容改成image/2-1是可以正常運行的），這里說明報錯的信息不全，導致報錯的信息容易造成誤導（以偏概全的情況）。