如何在TensorFlow中进行全球平均合并？

Question

如何在TensorFlow中进行全球平均合并？如果我有一个形状张量obj = [ 1, 2, [3, 4], 5 ] count_even(obj) => count_even([ 1, 2, [3, 4], 5]) => is obj an `int`? No => for i in [ 1, 2, [3, 4], 5] => count_even(1) => is 1 an int? Yes => return 0 because 1 is odd count_even(2) is 2 an int? Yes => return 1 because 2 is even count_even([3, 4]) => is [3, 4] an int? No => for i in [3, 4] => count_even(3) => Is 3 an int? Yes => return 0 because 3 is odd count_even(4) => Is 4 and int? Yes => return 1 because 4 is even count_even(5) => is 5 an int? Yes => return 0 because 5 is odd ，只要使用batch_size, height, width, channels = 32, 11, 40, 100只要channel = classes就足够了吗？

Answer 1

你也可以做 tf.reduce_mean（x，axis = [1,2]），特别是如果没有定义你的身高和宽度。

通常，在CNN中，张量的形状为EmailAddresses，其中b, h, w, c是批量大小，b和w对应于宽度和高度尺寸，以及h是渠道/过滤器的数量。

沿着轴c缩小时，减少张量的第一维和第二维（保持批量大小和通道/过滤器的数量）

Answer 2

你不需要迈步。填充=＆＃39;有效＆＃39; （默认值），如果合并过滤器的空间范围与图像的空间范围相同，则会获得1x1图像。除此之外，你就是这样做的，是的。