Intel® Distribution of OpenVINO™ Toolkit
Community assistance about the Intel® Distribution of OpenVINO™ toolkit, OpenCV, and all aspects of computer vision-related on Intel® platforms.

Keras+ tensorflow in NCS

idata
員工
1,588 檢視

Am gonna to do real time object recognition process in keras+ tensorflow (backend) using raspberry pi 3 with help of Movidius neural stick. So I would like to know that can I do real time object recognition in keras+tf in movidius stick and raspberry pi..

 

Thanks in advance

0 積分
4 回應
idata
員工
1,263 檢視

@Hashir If you want to use TensorFlow and NCS on the RPi, the only solution I know of that works is using Tiny Yolo V2. Although you will have to generate the graph file on a host machine and then transfer it over to the Pi if you are using the NCAPPZOO sample code: https://github.com/movidius/ncappzoo/tree/ncsdk2/tensorflow/tiny_yolo_v2. Then do a python3 run.py.

 

The NCSDK doesn't have support for Keras at the moment.

idata
員工
1,263 檢視

@Tome_at_intel, Thanks for your valuable help. I ran my own tiny yolo model with no of class = 1. But it seems to be very slow processing. So do you know how to speed up detections (actually object counting only no recognition) process. One method I found out that use c++ instead of python.

 

Thanks in advance

 

Hashir
idata
員工
1,263 檢視

Hi @Hashir

 

It probably is a little slower because you are running on a Raspberry Pi and Tiny Yolo requires some post processing done on the CPU. But both Tiny Yolo V1 & V2 have an native input size that is large compared to other networks (like SSD MobileNet). Since these networks were designed with an overall higher resolution for better accuracy, it means reduced speeds overall.

 

Best Regards,

 

Sahira
idata
員工
1,263 檢視

@Sahira_at_Intel , but many of the article says that by eliminating tweaking and unwanted layers , we can boost the process speed in tiny yolo. How can i achieve this ?

 

Thanks in advance

回覆