We often encounter a problem when deploying deep learning models: after packaging our own model into a library and giving it to Party A, Party A integrates the model into its own software, and needs to check the Generate Dependencies when compiling ( right-click project- > Generate Dependencies -> Generate Custom -> Corresponding CUDA ), as shown in the following figure:
If the developer of Party A chooses VS as the IDE environment, then there is no big problem. If you choose Qt Creator or other IDE environment, then you have to be familiar with CUDA programming, the whole process is very cumbersome.
So how to quickly import algorithm modules?
- First of all, you need to understand the role of generating dependencies in the VS compiler: select some libraries that need to be introduced and compiled in advance when compiling the current code.
- Whether we are doing algorithm source code compilation or CUDA compilation or library call, we need to check CUDA11.1