tf.nn.conv2d(input, filter, strides, padding, use_cudnn_on_gpu=None, name=None)
- input: input tensor for converlution with shape [batch, in_height, in_width, in_channels]
- filter: kernel tensor with shape [filter_height, filter_width, in_channels, out_channels]
- stride: step size tensor with shape [step_batch, step_height, step_width, step_channel]
- padding: option for converlution: “SAME”->with boundary, “VALID”->no boundary
- use_cudnn_on_gpu: whether to accelerate calculation with cudnn
- return: feature map
normally we need a conv2d wrapper function, so that the input and kernel matrix can be formulated correctly and adding activation function after converlutional computation
tf.layers.conv2d()
more advanced api of tensorflow
conv2d(inputs, filters, kernel_size, strides=(1, 1), padding=’valid’, data_format=’channels_last’, dilation_rate=(1, 1), activation=None, use_bias=True, kernel_initializer=None, bias_initializer=<tensorflow.python.ops.init_ops.Zeros object at 0x000002596A1FD898>, kernel_regularizer=None, bias_regularizer=None, activity_regularizer=None, kernel_constraint=None, bias_constraint=None, trainable=True, name=None, reuse=None)
- imputs: [batch, in_height, in_width, in_channels]
- filters: out_channel
- kernel_size: (filter_height, filter_width) or one integer(height equal width)
padding
W: input size，F: filter size，S: stride，⌈⌉: ceiling
- VALID:
- SAME:
deconvolution
tf.nn.conv2d_transpose(input, filters, output_shape, strides, padding=’SAME’, data_format=’NHWC’, dilations=None, name=None)
- input: [batch, height, width, in_channels]
- filter: [height, width, output_channels, input_channels] (ouput_channels is before input_channels)
- output_shape: [batch, height, width, out_channels], the format should corresponde to the data_format sequence
pooling
tf.nn.max_pool(value, ksize, strides, padding, name=None)
- valuevalue: feature map tensor with shape [batch, height, width, cnannels]
- ksize: pooling tensor with shape [1, height, width, 1]
- strides: step size tensor with shape [1, stride, stride, 1]
- padding: “VALID” or “SAME”
- return: feature map [batch, height, width, channels]
