AlexeyAB / darknet

YOLOv4 / Scaled-YOLOv4 / YOLO - Neural Networks for Object Detection (Windows and Linux version of Darknet )

http://pjreddie.com/darknet/

Other

21.59k stars 7.95k forks source link

What AWS instance to run inference for yolov4-tiny #7461

Open george-karma opened 3 years ago

george-karma commented 3 years ago

Hi, I have a trained yolov4-tiny model and I am looking for the best and cheapest way to deploy to AWS while maintaining 30fps. I read that t2.micro should be used for training #1380, however is this good enough for inference in this case? Since I'm using the tiny model is it possibile to choose a cheaper instance?

versavel commented 3 years ago

1380 says to use t2.micro to create the AMI, not to train a model. Trial and error method should allow you to quickly figure out the FPS-cost curve. I would start with cheapest gpu instance. Results will also depend on whether you use darknet or other package to infer.

george-karma commented 3 years ago

@versavel That is a good sugestion. I'l keep this open for now and update in a few days if anyone needs in the future.