Object Detection in 30 Seconds

Published in

SugarKubes

2 min readMar 19, 2019

SugarKubes is a container marketplace. Simply pull the containers to add amazing capabilities to your product! Learn more at https://sugarkubes.io

Introducing a new SugarKube: CPU / GPU Object Detection!

This pod includes the following models:

Fast

Very quick object detection with lower accuracy
can be run with or without a GPU
average CPU inference time is 80 ms per frame, that's roughly 12 FPS!
average GPU inference time is 23 ms per frame, roughly 45 FPS!
80 object classes

Accurate-600

Accurate object detection with slower detection time
can be run with or without a GPU
average CPU inference time is 500 ms, roughly 2 FPS
average GPU inference time is 50ms, roughly 20 FPS
600 object classes though with lower accuracy than the “Accurate” model

Accurate

most accurate object detection with slower detection time
can be run with or without a GPU
average CPU inference time is 500 ms, roughly 2 FPS
average GPU inference time is 50ms, roughly 20 FPS
80 object classes

We highly recommend running this system with an NVIDIA GPU. Tested on NVIDIA GTX 1080Ti for GPU and 2.7 GHz Intel Core i7 for CPU

Objects

80 objects, the full list is here. Includes, people, car, truck, bus, and other common items
600 objects, the full list is here. This thing’s got everything including guns, knives, weapons, man, woman, face, clothing!

Running

is as simple as pulling a docker container.

# sign up at https://sugarkubes.io first to get access!docker run --rm -ti \
-p 8080:8080 \
registry.sugarkubes.io/sugar-cv/object-detection:cpu# Then visit http://localhost:8080/tester/index.html to upload a photo and see the results!

Deploying

This pod comes ready to launch using kubernetes and helm. If you need help getting set up feel free to reach out.

cd chart
helm install --name object-detection -f values.yaml . --tls

System Requirements

Ubuntu 16 / 18
NVIDIA GPU or CPU (slower)

Want to learn more or need help getting set up! Chat us up at https://sugarkubes.io