Method definition
conv2d(input, filter, strides, padding, use_cudnn_on_gpu=True, data_format="NHWC", dilations=[1, 1, 1, 1], name=None)
Parameter Description
- INPUT:
INPUT is a shape [batch, in_height, in_width, in_channels ] of Tensor, wherein the batch is the number of data in each batch; in_height, in_width height and width of the input matrix; in_channels is the number of input channels, grayscale the value 1, RGB graph 3.
- filter:
filter is shaped as a [filter_height, filter_width, in_channels, out_channels ] the tensor, filter_height, filter_width convolution kernel is high and wide; in_channels is the number of input channels, the input consistency of in_channels; out_channels output channel is the number is the number of nuclear convolution.
- Strides:
Strides is the size of the sliding window, the shape of [. 1, a stride of, a stride of,. 1] .
- padding:
String type is 'SAME' and the 'VALID', represents the convolution approach. Wherein 'SAME' boundary into account when the convolution, when filled with less than 0, 'VALID' boundaries not considered.
- use_cudnn_on_gpu:
whether to use cudnn acceleration, the default is true