Is that possible that, darknet running on Intel GPU by using OpenCL?

[leekwunfung817]

I have a lot of computer are using Intel GPU on motherboard, How to run darknet using them?

Any software support or alternative tools to use them, to make darknet are possible running in most computer?

[AlexeyAB]

@leekwunfung817

• Intel Integrated GPU is very slow for training. So it makes little sense. But you can try to use these OpenCL versions on Intel GPU or AMD GPU, but it is much slower than nVidia GPU cuDNN.
  • https://github.com/sowson/darknet
  • https://github.com/ganyc717/Darknet-On-OpenCL

---

• This repository supports Detection on CPU. It supports multi-threading OPNEMP=1, Vectorization AVX=1 and BIT1-XNOR-inference, so you can achive 20-25 FPS on CPU for https://github.com/AlexeyAB/darknet/blob/master/cfg/yolov3-tiny_xnor.cfg model trained by yourself. More about it: https://github.com/AlexeyAB/darknet/issues/1472 So it makes darknet is possible running in most computers.

---

• Also you can try to use Detection on Darknet Yolov2 and Yolov3 that are builtin inside OpenCV:
  • https://github.com/opencv/opencv/pull/9705
  • https://github.com/opencv/opencv/pull/11322 (As you can see, OpenCL slower than CPU on Intel)

DNN_BACKEND_DEFAULT, DNN_TARGET_CPU	216.11ms
DNN_BACKEND_DEFAULT, DNN_TARGET_OPENCL	353.77ms

Use one of these examples

• C++: https://github.com/opencv/opencv/blob/master/samples/dnn/object_detection.cpp
• Python: https://github.com/opencv/opencv/blob/master/samples/dnn/object_detection.py

[leekwunfung817]

Thank you very much @AlexeyAB

Issue solved