How to deploy detection model using C++ API

I found all the C++ examples are about the classification model. The output is only one and DLTensor* y is int out_ndim = 2; int64_t out_shape[2] = {1, 1000, }; and get_output(0, y);, But in detection model, for example SSD, the output is three, the shapes of cores and class are {1,100,1},the shape of boxs is {1,100,4}, In this case, how do I define the out_ndim and out_shape of the output DLTensor?