mxnet.ndarray.Correlation(data1=None, data2=None, kernel_size=_Null, max_displacement=_Null, stride1=_Null, stride2=_Null, pad_size=_Null, is_multiply=_Null, out=None, name=None, **kwargs)
data1: [n, c, h, w]
data2: [n, c, h, w]
out: [n, \(c_{out}, h_{out}, w_{out}\)]
其中：
\(w_{out}=\lceil \frac{w+2\times pad\_size-2\times (max\_displacement+k)}{stride1}\rceil,\quad k=\frac{kernel\_size-1}{2}\)
\(h_{out}=\lceil \frac{h+2\times pad\_size-2\times (max\_displacement+k)}{stride1}\rceil,\quad k=\frac{kernel\_size-1}{2}\)
\(c_{out}=(2\times \frac{max\_displacement}{stride2}+1)\times(2\times \frac{max\_displacement}{stride2}+1)\)
mxnet.ndarray.Correlation输出参数size计算
原文：https://www.cnblogs.com/hf19950918/p/11805053.html