tensorflow反卷积(conv2d_transpose)
2016-11-08 本文已影响2675人
Persistently
tf.nn.conv2d_transpose(value, filter, output_shape,
strides, padding='SAME', name=None)
参数的设置和conv2d(卷积还是有一定区别的),比如第二个参数:先写output_channels,再写in_channels
这里的filter与conv2d有一点区别,反卷积【height,width,output_channels,in_channels】;卷积【height,width,in_channels,output_channels】