The issue of using torch::upsample in C++ while using the libtorch

C++
1

hello everyone, recently I’m using the libtorch to coding my CNN model on C++. I have met a problem which you can see from the figure1.

image
image706×410 12.1 KB

the code has pasted below’

at::Tensor test = torch::rand({1,1,2, 10 }).to(torch::kCUDA);

std::cout << test << std::endl;

//torch::upsample_nearest1d(test, { 20 });

torch::upsample_bilinear2d(test, {1,1,2,20 }, true);//this line cause that problem

I have tried another functions such as maxpool, convolution, they all worked well. however all upsampling function will break down. I have no idea about how to solve this problem

environment：
win10
Visual Studio 2015
RTX2070
CUDA 10.0
Cudnn7.4
libtorch nightly version

2

Do you have the backtrace for this error?

3

thanks for your attention,I have done this bug, the error cause by the coding, the right way to call the function is this way:
std::cout << test << std::endl;

//torch::upsample_nearest1d(test, { 20 });

std::cout << torch::upsample_bilinear2d(test, { 2,20 }, true) << std::endl;