#include <torch/torch.h> #include <torch/script.h> // One-stop header. #include <iostream> #include <memory> #include <opencv2/core/core.hpp> #include <opencv2/imgproc/imgproc.hpp> #include <opencv2/highgui/highgui.hpp> using namespace cv; using namespace std; int main() { std::cout << st…

If I’m not mistaken, OpenCV reads the image as [height, width, channels], while PIL.Image returns the array as [channels, height, width]. The torch::from_blob call might thus interleave the pixels. Have you checked the outputs? Also, resize should use the linear interpolation by default in OpenCV …

Values after softmax in torchscript in c++ is not same as python torch

C++

dambo (Shlomo) January 29, 2020, 7:44am 5

Try reading your image like this:

For the sake of testing, DO NOT use normalisation either in Python nor C++ and see if you get teh same results.

2 Likes