Average pooling in CA

[FranYi]

Hello, why don't you use the average sum of absolute values of the feature map to compute the channel-wise statistic z_c? For example, if one feature map is simplistically represented as [10,-10], then it's weight is 0; if the feature map is [10,10], then it's weight is 10. If we use the sum of absolute values, both feature maps have the same weight of 10. Besides, for two feature maps [-10,10] and [-1,1], which one should be given a higher weight? Thank you for your response!