Keras负尺寸Conv2D

Question

我已经玩了一段时间的内核大小和频道安排没有运气。我不完全确定如何计算Conv2D层的校正参数，也不确定这些参数的变化会对论文中模型的相似性产生多大的影响

任何帮助都将不胜感激

我试图建立的模型基于文献中的设计

input_shape = (4, 30, 180)
model = Sequential()
model.add(Convolution2D(32, (8, 8), strides=(4,4), activation='relu', input_shape=(4,30,180), data_format='channels_first'))
model.add(Activation('relu'))
model.add(Convolution2D(64, (4, 4), strides=(2, 2)))
model.add(Activation('relu'))
model.add(Convolution2D(64, (3, 3), strides=(1, 1)))
model.add(Activation('relu'))
model.add(Flatten())
model.add(Dense(512))
model.add(Activation('relu'))
model.add(Dense(2))
model.add(Activation('linear'))

我收到的错误消息

Traceback (most recent call last):
  File "/Users/zacharyfrederick/opt/anaconda3/lib/python3.7/site-packages/tensorflow/python/framework/ops.py", line 1659, in _create_c_op
    c_op = c_api.TF_FinishOperation(op_desc)
tensorflow.python.framework.errors_impl.InvalidArgumentError: Negative dimension size caused by subtracting 3 from 2 for 'conv2d_3/convolution' (op: 'Conv2D') with input shapes: [?,15,2,64], [3,3,64,64].

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "stock_env.py", line 101, in <module>
    model.add(Convolution2D(64, (3, 3), strides=(1, 1)))
  File "/Users/zacharyfrederick/opt/anaconda3/lib/python3.7/site-packages/keras/engine/sequential.py", line 181, in add
    output_tensor = layer(self.outputs[0])
  File "/Users/zacharyfrederick/opt/anaconda3/lib/python3.7/site-packages/keras/engine/base_layer.py", line 457, in __call__
    output = self.call(inputs, **kwargs)
  File "/Users/zacharyfrederick/opt/anaconda3/lib/python3.7/site-packages/keras/layers/convolutional.py", line 171, in call
    dilation_rate=self.dilation_rate)
  File "/Users/zacharyfrederick/opt/anaconda3/lib/python3.7/site-packages/keras/backend/tensorflow_backend.py", line 3650, in conv2d
    data_format=tf_data_format)
  File "/Users/zacharyfrederick/opt/anaconda3/lib/python3.7/site-packages/tensorflow/python/ops/nn_ops.py", line 851, in convolution
    return op(input, filter)
  File "/Users/zacharyfrederick/opt/anaconda3/lib/python3.7/site-packages/tensorflow/python/ops/nn_ops.py", line 966, in __call__
    return self.conv_op(inp, filter)
  File "/Users/zacharyfrederick/opt/anaconda3/lib/python3.7/site-packages/tensorflow/python/ops/nn_ops.py", line 591, in __call__
    return self.call(inp, filter)
  File "/Users/zacharyfrederick/opt/anaconda3/lib/python3.7/site-packages/tensorflow/python/ops/nn_ops.py", line 208, in __call__
    name=self.name)
  File "/Users/zacharyfrederick/opt/anaconda3/lib/python3.7/site-packages/tensorflow/python/ops/gen_nn_ops.py", line 1026, in conv2d
    data_format=data_format, dilations=dilations, name=name)
  File "/Users/zacharyfrederick/opt/anaconda3/lib/python3.7/site-packages/tensorflow/python/framework/op_def_library.py", line 788, in _apply_op_helper
    op_def=op_def)
  File "/Users/zacharyfrederick/opt/anaconda3/lib/python3.7/site-packages/tensorflow/python/util/deprecation.py", line 507, in new_func
    return func(*args, **kwargs)
  File "/Users/zacharyfrederick/opt/anaconda3/lib/python3.7/site-packages/tensorflow/python/framework/ops.py", line 3300, in create_op
    op_def=op_def)
  File "/Users/zacharyfrederick/opt/anaconda3/lib/python3.7/site-packages/tensorflow/python/framework/ops.py", line 1823, in __init__
    control_input_ops)
  File "/Users/zacharyfrederick/opt/anaconda3/lib/python3.7/site-packages/tensorflow/python/framework/ops.py", line 1662, in _create_c_op
    raise ValueError(str(e))
ValueError: Negative dimension size caused by subtracting 3 from 2 for 'conv2d_3/convolution' (op: 'Conv2D') with input shapes: [?,15,2,64], [3,3,64,64].

Answer 1

出现此错误是因为内核和步幅对于输入来说太大，一个常见的开始是使用(3, 3)形状和步幅(1, 1)的内核

尝试阅读卷积是如何计算的，以直观地了解如何设置正确的内核/步幅大小：http://cs231n.github.io/convolutional-networks/

此外，您有一个channel first的输入，因此您将第一个conv设置为channel first，这很好，但是您可以为所有卷积执行此操作，因为默认情况下，keras卷积将使用channel last

例如，这是可行的：

input_shape = (4, 30, 180)
model = Sequential()
model.add(Conv2D(32, (8, 8), strides=(4, 4), activation='relu', input_shape=(4, 30, 180), data_format='channels_first'))
model.add(Activation('relu'))
model.add(Conv2D(64, (4, 4), strides=(1, 1), data_format='channels_first'))
model.add(Activation('relu'))
model.add(Conv2D(64, (3, 3), strides=(1, 1), data_format='channels_first'))
model.add(Activation('relu'))
model.add(Flatten())
model.add(Dense(512))
model.add(Activation('relu'))
model.add(Dense(2))
model.add(Activation('linear'))

Answer 2

在诊断中，另一个答案是正确的：在卷积之后，图像会缩小，在某个点上内核会比图像大。试一试

1）降低内核大小或

2）将, padding='same'添加到卷积层

使用Calculate the Output size in Convolution layer计算输出大小