python - "TypeError: Using a ` tf.Tensor ` as a Python ` bool ` is not allowed."在数据集上调用map函数时

Question

我正在尝试使用从每个图像中学习到的独特裁剪因子来加载和处理图像。我不断收到错误消息，指出我无法将张量用作 Python bool 值。

对于每个图像，我想对图像中心的一行像素进行阈值处理，并计算超过某个阈值的像素百分比。我想使用该百分比作为裁剪因子。

def preprocess_image(image):
    image = tf.image.decode_png(image, channels=3)
    print(tf.shape(image))
    halfpix = tf.shape(image)[0]//2
    row = tf.cast(tf.math.greater(image[:, :, 0][halfpix, :], 3), tf.float32)
    hor_scale_factor = tf.math.reduce_mean(row)
    image = tf.image.central_crop(image, hor_scale_factor)

    return image

def load_and_preprocess_image(path):
    image = tf.io.read_file(path)
    return preprocess_image(image)

train_image_ds = train_path_ds.map(load_and_preprocess_image, num_parallel_calls=AUTOTUNE)

我希望没有错误。我收到:“TypeError:不允许使用 tf.Tensor 作为 Python bool。使用 if t is not None: 而不是 if t: 测试是否定义了张量，并使用 tf.cond 等 TensorFlow 操作来执行以张量值为条件的子图。”

完整跟踪:

TypeError                                 Traceback (most recent call last) <ipython-input-89-b7d7da47ff6e> in <module>
----> 1 train_image_ds = train_path_ds.map(load_and_preprocess_image, num_parallel_calls=AUTOTUNE)
      2 test_image_ds = test_path_ds.map(load_and_preprocess_image, num_parallel_calls=AUTOTUNE)

c:\users\toby-pc\documents\code\blindness_kaggle\my_env\lib\site-packages\tensorflow\python\data\ops\dataset_ops.py in map(self, map_func, num_parallel_calls)    1144     else:    1145   return ParallelMapDataset(
-> 1146           self, map_func, num_parallel_calls, preserve_cardinality=True)    1147     1148   def flat_map(self, map_func):

c:\users\toby-pc\documents\code\blindness_kaggle\my_env\lib\site-packages\tensorflow\python\data\ops\dataset_ops.py in __init__(self, input_dataset, map_func, num_parallel_calls, use_inter_op_parallelism, preserve_cardinality, use_legacy_function)   3262         self._transformation_name(),    3263         dataset=input_dataset,
-> 3264         use_legacy_function=use_legacy_function)    3265     self._num_parallel_calls = ops.convert_to_tensor(    3266         num_parallel_calls, dtype=dtypes.int32, name="num_parallel_calls")

c:\users\toby-pc\documents\code\blindness_kaggle\my_env\lib\site-packages\tensorflow\python\data\ops\dataset_ops.py in __init__(self, func, transformation_name, dataset, input_classes, input_shapes, input_types, input_structure, add_to_graph, use_legacy_function, defun_kwargs)    2589       resource_tracker = tracking.ResourceTracker()    2590       with tracking.resource_tracker_scope(resource_tracker):
-> 2591         self._function = wrapper_fn._get_concrete_function_internal()    2592         if add_to_graph:    2593           self._function.add_to_graph(ops.get_default_graph())

c:\users\toby-pc\documents\code\blindness_kaggle\my_env\lib\site-packages\tensorflow\python\eager\function.py in _get_concrete_function_internal(self, *args, **kwargs)    1364     """Bypasses error checking when getting a graph function."""    1365   graph_function = self._get_concrete_function_internal_garbage_collected(
-> 1366         *args, **kwargs)    1367     # We're returning this concrete function to someone, and they may keep a    1368     # reference to the FuncGraph without keeping a reference to the

c:\users\toby-pc\documents\code\blindness_kaggle\my_env\lib\site-packages\tensorflow\python\eager\function.py in _get_concrete_function_internal_garbage_collected(self, *args,
**kwargs)    1358     if self.input_signature:    1359       args, kwargs = None, None
-> 1360     graph_function, _, _ = self._maybe_define_function(args, kwargs)    1361     return graph_function    1362 

c:\users\toby-pc\documents\code\blindness_kaggle\my_env\lib\site-packages\tensorflow\python\eager\function.py in _maybe_define_function(self, args, kwargs)    1646       graph_function = self._function_cache.primary.get(cache_key, None)    1647       if graph_function is None:
-> 1648         graph_function = self._create_graph_function(args, kwargs)    1649         self._function_cache.primary[cache_key] = graph_function    1650       return graph_function, args, kwargs

c:\users\toby-pc\documents\code\blindness_kaggle\my_env\lib\site-packages\tensorflow\python\eager\function.py in _create_graph_function(self, args, kwargs, override_flat_arg_shapes)    1539             arg_names=arg_names,    1540             override_flat_arg_shapes=override_flat_arg_shapes,
-> 1541             capture_by_value=self._capture_by_value),    1542         self._function_attributes)    1543 

c:\users\toby-pc\documents\code\blindness_kaggle\my_env\lib\site-packages\tensorflow\python\framework\func_graph.py in func_graph_from_py_func(name, python_func, args, kwargs, signature, func_graph, autograph, autograph_options, add_control_dependencies, arg_names, op_return_value, collections, capture_by_value, override_flat_arg_shapes)
    714                                           converted_func)
    715 
--> 716       func_outputs = python_func(*func_args, **func_kwargs)
    717 
    718       # invariant: `func_outputs` contains only Tensors, CompositeTensors,

c:\users\toby-pc\documents\code\blindness_kaggle\my_env\lib\site-packages\tensorflow\python\data\ops\dataset_ops.py in wrapper_fn(*args)    2583           attributes=defun_kwargs)    2584       def wrapper_fn(*args):  # pylint: disable=missing-docstring
-> 2585         ret = _wrapper_helper(*args)    2586         ret = self._output_structure._to_tensor_list(ret)    2587         return [ops.convert_to_tensor(t) for t in ret]

c:\users\toby-pc\documents\code\blindness_kaggle\my_env\lib\site-packages\tensorflow\python\data\ops\dataset_ops.py in _wrapper_helper(*args)    2528         nested_args = (nested_args,) 2529 
-> 2530       ret = func(*nested_args)    2531       # If `func` returns a list of tensors, `nest.flatten()` and    2532       # `ops.convert_to_tensor()` would conspire to attempt to stack

<ipython-input-86-61d7ec60892c> in load_and_preprocess_image(path)
      1 def load_and_preprocess_image(path):
      2     image = tf.io.read_file(path)
----> 3     return preprocess_image(image)

<ipython-input-85-4f3b9475e191> in preprocess_image(image)
     11     hor_scale_factor = tf.math.reduce_mean(row)
     12 #     print(hor_scale_factor)
---> 13     image = tf.image.central_crop(image, hor_scale_factor)
     14 #     print(type(hor_scale_factor))
     15     image = tf.image.resize(image, target_im_size, preserve_aspect_ratio=True) # Resize to final dimensions

c:\users\toby-pc\documents\code\blindness_kaggle\my_env\lib\site-packages\tensorflow\python\ops\image_ops_impl.py in central_crop(image, central_fraction)
    641   with ops.name_scope(None, 'central_crop', [image]):
    642     image = ops.convert_to_tensor(image, name='image')
--> 643     if central_fraction <= 0.0 or central_fraction > 1.0:
    644       raise ValueError('central_fraction must be within (0, 1]')
    645     if central_fraction == 1.0:

c:\users\toby-pc\documents\code\blindness_kaggle\my_env\lib\site-packages\tensorflow\python\framework\ops.py in __bool__(self)
    696       `TypeError`.
    697     """
--> 698     raise TypeError("Using a `tf.Tensor` as a Python `bool` is not allowed. "
    699                     "Use `if t is not None:` instead of `if t:` to test if a "
    700                     "tensor is defined, and use TensorFlow ops such as "

TypeError: Using a `tf.Tensor` as a Python `bool` is not allowed. Use `if t is not None:` instead of `if t:` to test if a tensor is defined, and use TensorFlow ops such as tf.cond to execute subgraphs conditioned on the value of a tensor.

Answer 1

tf.image.central_crop要求 central_fraction 参数是实际的浮点值，因此无法使用 TensorFlow 张量。不过，复制功能很容易，例如使用 tf.image.crop_to_bounding_box (或者甚至只是切片，这就是该函数真正的作用):

import tensorflow as tf

def central_crop_tf(image, central_fraction):
    s = tf.shape(image)
    h, w = s[-3], s[-2]
    h_box = tf.cast(tf.round(central_fraction * tf.cast(h, tf.float32)), tf.int32)
    w_box = tf.cast(tf.round(central_fraction * tf.cast(w, tf.float32)), tf.int32)
    h_off = (h - h_box) // 2
    w_off = (w - w_box) // 2
    return tf.image.crop_to_bounding_box(image, h_off, w_off, h_box, w_box)

# Test
with tf.Graph().as_default(), tf.Session() as sess:
    img = tf.reshape(tf.range(80), [1, 8, 10, 1])
    print(sess.run(img)[0, :, :, 0])
    # [[ 0  1  2  3  4  5  6  7  8  9]
    #  [10 11 12 13 14 15 16 17 18 19]
    #  [20 21 22 23 24 25 26 27 28 29]
    #  [30 31 32 33 34 35 36 37 38 39]
    #  [40 41 42 43 44 45 46 47 48 49]
    #  [50 51 52 53 54 55 56 57 58 59]
    #  [60 61 62 63 64 65 66 67 68 69]
    #  [70 71 72 73 74 75 76 77 78 79]]
    frac = tf.constant(0.4)
    res = central_crop_tf(img, frac)
    print(sess.run(res)[0, :, :, 0])
    # [[23 24 25 26]
    #  [33 34 35 36]
    #  [43 44 45 46]]