major rework on the detect.py module, to encapsulate the main functions.
This allows to create hooks in the main loop, where those functions can be substituted with custom implementations to support different models or different inference engines like tensorRT.
I also provided those function for tensorRT.
Moreover, there are a couple of modifications in the installation of TF-cuda, where I inserted the possibility to select tensorRT and I had to modify the install script to insert a boundary on the register count when compiling from source, which was otherwise failing.
major rework on the detect.py module, to encapsulate the main functions. This allows to create hooks in the main loop, where those functions can be substituted with custom implementations to support different models or different inference engines like tensorRT.
I also provided those function for tensorRT.
Moreover, there are a couple of modifications in the installation of TF-cuda, where I inserted the possibility to select tensorRT and I had to modify the install script to insert a boundary on the register count when compiling from source, which was otherwise failing.