Tensorflow 2.5 model.evaluate报错Invalid argument: required broadcastable shapes at loc(unknown)

⚡插播一条老家自产的糖心苹果，多个品种，欢迎选购！有问题随时私信我⚡：🍎🍎来自雪域高原的馈赠——海拔2000米的大凉山高原生态糖心苹果，欢迎选购！！🍎🍎

大凉山高原生态糖心苹果

1.软件环境⚙️

Windows10 教育版64位
Python 3.6.3
Tensorflow-GPU 2.5.0
CUDA 11.1

2.问题描述🔍

我们在模型训练完时，都需要对模型的性能进行评估。而在Tensorflow.Keras中，往往通过.flow_from_directory函数读入本地的图片，然后使用model.evaluate对模型进行精度评估：

test_datagen = ImageDataGenerator(preprocessing_function=preprocessing_function)

test_generator = test_datagen.flow_from_directory(test_dir,
                                                  shuffle=False,
                                                  target_size=(299,299),
                                                  batch_size=32)

print("================开始模型评估======================")
model_evaluation = model.evaluate(test_generator, verbose=1)

比如我们这边有一个评估数据集val-fewer-sample，该数据集中包含dog、cat两类：

这些样本已经被打上了正确的标签（即文件夹名），我们训练出来的分类器对这些样本进行预测，如果标签对得上，那么该图片预测正确。
如果你训练的时候使用的是softmax那么不会有问题，但二分类问题我更喜欢用sigmoid，这个时候，如果是Tensorflow 2.5，就会出现报错：

Invalid argument: required broadcastable shapes at loc(unknown)

2022-09-01 10:31:58.676170: W tensorflow/core/framework/op_kernel.cc:1755] Invalid argument: required broadcastable shapes at loc(unknown)
2022-09-01 10:31:58.676482: W tensorflow/core/framework/op_kernel.cc:1755] Invalid argument: required broadcastable shapes at loc(unknown)
2022-09-01 10:31:58.676653: W tensorflow/core/framework/op_kernel.cc:1755] Invalid argument: required broadcastable shapes at loc(unknown)
Traceback (most recent call last):
  File "<input>", line 1, in <module>
  File "C:\Program Files\JetBrains\PyCharm 2020.1\plugins\python\helpers\pydev\_pydev_bundle\pydev_umd.py", line 197, in runfile
    pydev_imports.execfile(filename, global_vars, local_vars)  # execute the script
  File "C:\Program Files\JetBrains\PyCharm 2020.1\plugins\python\helpers\pydev\_pydev_imps\_pydev_execfile.py", line 18, in execfile
    exec(compile(contents+"\n", file, 'exec'), glob, loc)
  File "E:/Code/Python/classification model evaluation/model_evaluation_sigmoid.py", line 71, in <module>
    model_evaluation = model.evaluate(test_generator, verbose=1, workers=4, return_dict=True)
  File "C:\Users\Anaconda3\envs\tf2.5\lib\site-packages\tensorflow\python\keras\engine\training.py", line 1489, in evaluate
    tmp_logs = self.test_function(iterator)
  File "C:\Users\Anaconda3\envs\tf2.5\lib\site-packages\tensorflow\python\eager\def_function.py", line 889, in __call__
    result = self._call(*args, **kwds)
  File "C:\Users\Anaconda3\envs\tf2.5\lib\site-packages\tensorflow\python\eager\def_function.py", line 957, in _call
    filtered_flat_args, self._concrete_stateful_fn.captured_inputs)  # pylint: disable=protected-access
  File "C:\Users\Anaconda3\envs\tf2.5\lib\site-packages\tensorflow\python\eager\function.py", line 1961, in _call_flat
    ctx, args, cancellation_manager=cancellation_manager))
  File "C:\Users\Anaconda3\envs\tf2.5\lib\site-packages\tensorflow\python\eager\function.py", line 596, in call
    ctx=ctx)
  File "C:\Users\Anaconda3\envs\tf2.5\lib\site-packages\tensorflow\python\eager\execute.py", line 60, in quick_execute
    inputs, attrs, num_outputs)
tensorflow.python.framework.errors_impl.InvalidArgumentError: 2 root error(s) found.
  (0) Invalid argument:  required broadcastable shapes at loc(unknown)
	 [[node LogicalAnd (defined at E:/Code/Python/classification model evaluation/model_evaluation_sigmoid.py:71) ]]
	 [[assert_greater_equal_1/Assert/AssertGuard/else/_29/assert_greater_equal_1/Assert/AssertGuard/Assert/data_0/_65]]
  (1) Invalid argument:  required broadcastable shapes at loc(unknown)
	 [[node LogicalAnd (defined at E:/Code/Python/classification model evaluation/model_evaluation_sigmoid.py:71) ]]
0 successful operations.
0 derived errors ignored. [Op:__inference_test_function_5740]
Function call stack:
test_function -> test_function

可以看到，Tensorflow 报错原因是tensor的shape不一致，这就很尴尬了，为什么softmax训练出来的模型可以正常评估，但是sigmoid训练出来的模型就报错呢？

3.解决方法🐡

经过查询，发现是因为.flow_from_directory中的class_mode参数的默认值是categorical，而我们使用的是sigmoid进行训练：

  def flow_from_directory(self,
                          directory,
                          target_size=(256, 256),
                          color_mode='rgb',
                          classes=None,
                          class_mode='categorical',
                          batch_size=32,
                          shuffle=True,
                          seed=None,
                          save_to_dir=None,
                          save_prefix='',
                          save_format='png',
                          follow_links=False,
                          subset=None,
                          interpolation='nearest'):
    """Takes the path to a directory & generates batches of augmented data."""

因此需要将class_mode修改为binary才能用sigmoid适配，即：

test_generator = test_datagen.flow_from_directory(test_dir,
                                                  shuffle=False,
                                                  target_size=(299,299),
                                                  batch_size=32)
# 修改为：
test_generator = test_datagen.flow_from_directory(test_dir,
                                                  shuffle=False,
                                                  target_size=(299,299),
                                                  class_mode='binary',
                                                  batch_size=32)

4.结果预览🤔

修改完class_mode之后，发现模型评估可以正常运行了：

Found 12226 images belonging to 2 classes.
================开始模型评估======================
2022-09-01 13:47:56.720307: I tensorflow/compiler/mlir/mlir_graph_optimization_pass.cc:176] None of the MLIR Optimization Passes are enabled (registered 2)
2022-09-01 13:47:59.106484: I tensorflow/stream_executor/platform/default/dso_loader.cc:53] Successfully opened dynamic library cudnn64_8.dll
2022-09-01 13:48:00.179758: I tensorflow/stream_executor/cuda/cuda_dnn.cc:359] Loaded cuDNN version 8201
C:\Users\Anaconda3\envs\tf2.5\lib\site-packages\PIL\Image.py:976: UserWarning: Palette images with Transparency expressed in bytes should be converted to RGBA images
  "Palette images with Transparency expressed in bytes should be "
2022-09-01 13:48:02.140887: I tensorflow/stream_executor/platform/default/dso_loader.cc:53] Successfully opened dynamic library cublas64_11.dll
2022-09-01 13:48:02.827805: I tensorflow/stream_executor/platform/default/dso_loader.cc:53] Successfully opened dynamic library cublasLt64_11.dll
 42/192 [=====>........................] - ETA: 32s - loss: 0.5887 - accuracy: 0.9371 - precision: 0.8920 - recall: 0.9420

​
渣男！都看到这里了，还不赶紧点赞，评论，收藏走一波？