tf.nn.conv2d(input, filter, strides, padding, use_cudnn_on_gpu=None, name=None), 第一个参数input:Tensor,[batch, in_height, in_width, in_c