- The size of second channel should be size[2] of output tensor, - The Scalar should be {variance[0], variance[0], variance[0], variance[0]} for _variance.size() == 1 case. Signed-off-by: Wu Zhiwen <zhiwen.wu@intel.com>