Integer template parameters and subfunction calls

In this short snippet, I want to create an Eigen::Tensor (in the unsupported module), with aribtrary dimensions.

template
<typename T>
Tensor<T, 2> convertNPToEigen2D(np::ndarray const & arr)
{
  //Some checking...
  T* raw_arr_data = reinterpret_cast<T*>(arr.get_data());
  TensorMap<Tensor<T, 2>> arr_eigen(raw_arr_data,
            arr.shape(0), arr.shape(1));
  //...
  return arr_eigen;
}

You can of course see that without variadic templates, I have to duplicate this function for every possible number of dimensions. This seems like a pretty basic examples, where variadic templates can avoid a lot of code duplication:

template
<typename T, uint64_t dims>
Tensor<T, dims> convertNPToEigenND(np::ndarray const & arr)
{
  //Some checking...
  T* raw_arr_data = reinterpret_cast<T*>(arr.get_data());
  TensorMap<Tensor<T, dims>> arr_eigen(raw_arr_data,
            /*arr.shape(0), ..., arr.shape(dims-1)*/);
  //...
  return arr_eigen;
}

I have several problems translating this into variadic template code,since I don't have an argument pack. I guess it would also be possible to let the caller do the work, i.e.

convertNPToEigenND<float, arr.shape(0), arr.shape(1), arr.shape(2)>(arr)

but I would prefer the above solution if it is possible. I thought it would be easy to look this problem up, but most of the questions I have found deal with an existing argument pack instead of creating one.

Thanks in advance!