Deploy the trained model in production

[DeeeepNet]

Hi @AlexeyAB, Please guide me if I want to train yolo3 for custom my own dataset on ubuntu 16.04 and deploy the trained model on server or GPU or embedded system that have Linux based OS. I want to load the trained model on GPU or RAM of the system only once and then every time the camera captured the image feed to the loaded model. Is enough the one 1080-ti GPU for training yolo3?

thanks in advance.

[AlexeyAB]

@DeeeepNet Hi,

Please guide me if I want to train yolo3 for custom my own dataset on ubuntu 16.04 and deploy the trained model on server or GPU or embedded system that have Linux based OS.

So, just compile Darknet with GPU=1 CUDNN=1 OPENCV=1 LIBSO=1 and use https://github.com/AlexeyAB/darknet#how-to-compile-on-linux

• ./darknet detector demo..
• ./uselib ...
• or use compiled libdarknet.so by your own C/C++/Python application

Is enough the one 1080-ti GPU for training yolo3?

Usually yes. Just train while mAP is increasing: https://github.com/AlexeyAB/darknet#when-should-i-stop-training