Is it possible to change some code in YOLO object detection (GPU Version)?

Question

I working on an object detection project in these days but I got stuck in changing some code of YOLOV3 GPU version.

I have tried to implement the YOLOV3 on CPU however, it did not satisfy me about FPS values. I want to implement it on GPU. So my question is "Can I change some code of YOLOV3 GPU version ?". What I mean by that is, when I open the darknet folder, I saw some files that were written in C++. I need to get the bounding box coordinates and modify some part of the codes. Is it possible to do that ?

In addition, I made some project with YOLOV3 about object detection on Python but as I said, I want to implement on GPU and also I am working on Ubuntu 18.04.

Answer 1

There is nothing special about the GPU code versus the CPU code.

You are asking about ML, so I'll take a tangent. Apologies if this is obvious. Compared to Python, C++ is compiled. So, you can't just expect to change some (C++) source and run again and obtain something different. This is true in most Python-based ML project, but not in C++.

You'll need to rebuild. Are you currently rebuilding the project?

You also don't make clear what changes you want to make. If it changes the network layers, you may have to retrain from scratch, instead of using pre-trained models. For most vision ML projects, the pretraining is very expensive in terms of computation. Consider this carefully.