Skip to content

Yolov2-tiny Tensorflow implementation for Movidius Neural Compute Stick

License

Notifications You must be signed in to change notification settings

bastiaanv/Yolov2-tiny-tf-NCS

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

26 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Yolov2 tiny tensorflow for Movidius Neural Compute Stick (NCS)

This is a Tensorflow/NCSdk implementation of Yolov2. It was inpired by simo23 repo, see: https://github.com/simo23/tinyYOLOv2.

Credits to him for the functions to load the .weights file and functions to process the outcome of the Network.

I made this repo to make a tensorflow implementation of Yolov2 possible for the Movidius NCS, because until now I've only seen caffe implementations of Yolo for the NCS.

How to use it on the NCS

  • clone the repo on a linux 16.04 system
  • install ncsdk2
  • run the following command in the terminal:
python3 Script.py
make compile
python3 run.py -m [video/image] [-i PathToImage] [-v PathToVideo] [-v 0] #Last one is for using your webcam
  • and done! The output of the network will look something like this:

alt text

How to use it normally / in debug mode

  • clone the repo
  • run the following script with python3:
Script.py
  • and done! The output should be the same as on the NCS

Note!

The processing power of the NCS is not as good as 4 titan X (pascal) GPU's, ofcourse. That is why you will get a processing time of ~6Hz / ~170ms, instead of 220Hz like advertised on the Yolov2 page. The biggest lost in preformance is made in conv7 and conv8:

alt text

If anybody can help me improve this lost, it would be much appreciated!! Any other sugestions and questions are welcome as well! You can contact me on this repo or at: [email protected]

About

Yolov2-tiny Tensorflow implementation for Movidius Neural Compute Stick

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published