Consolidate common CV functions

[atar13]

There's chunks of code in the CV files that are repeated.

These are the functions we should have in src/cv/utilities.cpp:

• openCVToTensor
  • Should be based on these lines in matching.cpp. @AchuthKrishna found a bug that he fixed so we should use his version https://github.com/tritonuas/obcpp/blob/main/src/cv/matching.cpp#L5-L28.
    

    obcpp/src/cv/matching.cpp

    Lines 29 to 35 in 3232fbd

    	auto toInput(cv::Mat img, bool show_output = false, bool unsqueeze = false, int unsqueeze_dim = 0) {
    	cv::Mat img_dst;
    	// scale the image and turn into Tensor, then input vector
    	cv::resize(img, img_dst, cv::Size(128, 128), 0, 0, cv::INTER_AREA);
    	img_dst.convertTo(img_dst, CV_32FC3);
    	torch::Tensor tensor_image = torch::ones({1, 3, img_dst.rows, img_dst.cols}, torch::kF32);
    	std::memcpy(tensor_image.data_ptr(), img_dst.data, sizeof(float)*tensor_image.numel());

• tensorToModelInput
  • Basically this https://github.com/tritonuas/obcpp/blob/main/src/cv/segmentation.cpp#L96-L99
• tensorToOpenCV
  • https://github.com/tritonuas/obcpp/blob/main/src/cv/segmentation.cpp#L101-L114
• loadPytorchModel
  • Take a model path string and return a torch::jit::script::Module
  • Basically like this code segment: https://github.com/tritonuas/obcpp/blob/cv-saliency-integration/src/cv/segmentation.cpp#L6-L12