输入和滤波器的深度必须相同：使用MKL和channels first form时为32 vs 16

Question

为了加快我的DNN学习进度，我用MKL从源代码编译了tensorflow。我有一个ResNet模型，它是从tensorflow/models复制的。数据集是CIFAR-10。当我用channels last格式运行模型时，一切正常。但是为了使用MKL，也就是说只对channels first格式进行加速，我添加了一些代码将数据转换为NCHW格式，并运行它。然后我得到：

Caused by op 'stage/residual_v1/conv2d/Conv2D', defined at:
File "main.py", line 182, in <module>
  main(args)
File "main.py", line 83, in main
  tf.estimator.train_and_evaluate(estimator, train_spec, eval_spec)
File "/home/holmescn/.pyenv/versions/anaconda35.2.0/envs/tensorflow/lib/python3.6/site-packages/tensorflow/python/estimator/training.py", line 447, in train_and_evaluate
  return executor.run()
File "/home/holmescn/.pyenv/versions/anaconda3-5.2.0/envs/tensorflow/lib/python3.6/site-packages/tensorflow/python/estimator/training.py", line 531, in run
  return self.run_local()
File "/home/holmescn/.pyenv/versions/anaconda3-5.2.0/envs/tensorflow/lib/python3.6/site-packages/tensorflow/python/estimator/training.py", line 669, in run_local
hooks=train_hooks)
File "/home/holmescn/.pyenv/versions/anaconda3-5.2.0/envs/tensorflow/lib/python3.6/site-packages/tensorflow/python/estimator/estimator.py", line 366, in train
loss = self._train_model(input_fn, hooks, saving_listeners)
File "/home/holmescn/.pyenv/versions/anaconda3-5.2.0/envs/tensorflow/lib/python3.6/site-packages/tensorflow/python/estimator/estimator.py", line 1119, in _train_model
return self._train_model_default(input_fn, hooks, saving_listeners)
File "/home/holmescn/.pyenv/versions/anaconda3-5.2.0/envs/tensorflow/lib/python3.6/site-packages/tensorflow/python/estimator/estimator.py", line 1132, in _train_model_default
 features, labels, model_fn_lib.ModeKeys.TRAIN, self.config)
File "/home/holmescn/.pyenv/versions/anaconda3-5.2.0/envs/tensorflow/lib/python3.6/site-packages/tensorflow/python/estimator/estimator.py", line 1107, in _call_model_fn
model_fn_results = self._model_fn(features=features, **kwargs)
File "/home/holmescn/Work/deep-learning-practice/tensorflow/estimator/utils.py", line 18, in _model_fn
logits = build_model(input_layer, mode == tf.estimator.ModeKeys.TRAIN, params=params, args=args)
File "/home/holmescn/Work/deep-learning-practice/tensorflow/estimator/estimators/resnet.py", line 175, in build_model
return resnet.build_model(input_layer, args.num_layers)
File "/home/holmescn/Work/deep-learning-practice/tensorflow/estimator/estimators/resnet.py", line 56, in build_model
x = res_func(x, 3, filters[i], filters[i + 1], strides[i])
File "/home/holmescn/Work/deep-learning-practice/tensorflow/estimator/estimators/resnet.py", line 79, in _residual_v1
x = self._conv(x, kernel_size, out_filter, stride)
File "/home/holmescn/Work/deep-learning-practice/tensorflow/estimator/estimators/base.py", line 59, in _conv
name=name)
File "/home/holmescn/.pyenv/versions/anaconda3-5.2.0/envs/tensorflow/lib/python3.6/site-packages/tensorflow/python/layers/convolutional.py", line 427, in conv2d
return layer.apply(inputs)
File "/home/holmescn/.pyenv/versions/anaconda3-5.2.0/envs/tensorflow/lib/python3.6/site-packages/tensorflow/python/keras/engine/base_layer.py", line 759, in apply
return self.__call__(inputs, *args, **kwargs)
File "/home/holmescn/.pyenv/versions/anaconda3-5.2.0/envs/tensorflow/lib/python3.6/site-packages/tensorflow/python/layers/base.py", line 329, in __call__
outputs = super(Layer, self).__call__(inputs, *args, **kwargs)
File "/home/holmescn/.pyenv/versions/anaconda3-5.2.0/envs/tensorflow/lib/python3.6/site-packages/tensorflow/python/keras/engine/base_layer.py", line 688, in __call__
outputs = self.call(inputs, *args, **kwargs)
File "/home/holmescn/.pyenv/versions/anaconda3-5.2.0/envs/tensorflow/lib/python3.6/site-packages/tensorflow/python/keras/layers/convolutional.py", line 184, in call
outputs = self._convolution_op(inputs, self.kernel)
File "/home/holmescn/.pyenv/versions/anaconda3-5.2.0/envs/tensorflow/lib/python3.6/site-packages/tensorflow/python/ops/nn_ops.py", line 868, in __call__
return self.conv_op(inp, filter)
File "/home/holmescn/.pyenv/versions/anaconda3-5.2.0/envs/tensorflow/lib/python3.6/site-packages/tensorflow/python/ops/nn_ops.py", line 520, in __call__
return self.call(inp, filter)
File "/home/holmescn/.pyenv/versions/anaconda3-5.2.0/envs/tensorflow/lib/python3.6/site-packages/tensorflow/python/ops/nn_ops.py", line 204, in __call__
name=self.name)
File "/home/holmescn/.pyenv/versions/anaconda3-5.2.0/envs/tensorflow/lib/python3.6/site-packages/tensorflow/python/ops/gen_nn_ops.py", line 956, in conv2d
data_format=data_format, dilations=dilations, name=name)
File "/home/holmescn/.pyenv/versions/anaconda3-5.2.0/envs/tensorflow/lib/python3.6/site-packages/tensorflow/python/framework/op_def_library.py", line 787, in _apply_op_helper
op_def=op_def)
File "/home/holmescn/.pyenv/versions/anaconda3-5.2.0/envs/tensorflow/lib/python3.6/site-packages/tensorflow/python/framework/ops.py", line 3414, in create_op
 op_def=op_def)
File "/home/holmescn/.pyenv/versions/anaconda3-5.2.0/envs/tensorflow/lib/python3.6/site-packages/tensorflow/python/framework/ops.py", line 1740, in __init__
self._traceback = self._graph._extract_stack()  # pylint: disable=protected-access

InvalidArgumentError (see above for traceback): input and filter must have the same depth: 32 vs 16
 [[Node: stage/residual_v1/conv2d/Conv2D = _MklConv2D[T=DT_FLOAT, _kernel="MklOp", data_format="NCHW", dilations=[1, 1, 1, 1], padding="SAME", strides=[1, 1, 1, 1], use_cudnn_on_gpu=true, _device="/job:localhost/replica:0/task:0/device:CPU:0"](Relu, conv2d/kernel/read, Relu:1, DMT/_6)]]

最后一个trackback说input and filter must have the same depth，这意味着输入张量和滤波器的深度dim应该相同？但是如果我想生成更多的特征地图，我该怎么做呢？我该怎么办？

Answer 1

更改数据格式时无需转置数据。 您可以将数据格式作为channels first或channels last作为参数传递

例如， python cifar10_主.pydata dir=${PWD}/cifar-10-data data format=channels_first job dir=/tmp/cifar10