So I finally got my model working well enough to consider moving to a prototype and realized it isn’t as easy as I had initially thought. I just want to make sure I understand the steps of the process before I put in a lot of time learning TorchScript and C++. The model is a binary(possibly more classes later) semantic segmentation of cracked roads. Also I know OpenCV has C++ capabilities but are there similar libraries like numpy, pandas. Any recommendations on hardware to use?
- Use TorchScript to map/compile just the Pytorch model to C++
- Then use CodeBlocks to write a C++ program that will call the model
- Use OpenCV and other addons in program for video filtering and matrix operations
- Install the program on a hardware device. Any recommendations?