Deep Learning Deployment Toolkit OpenVINO 2020.1

We’re happy to announce new package for OpenWRT and Turris OS. Please welcome OpenVINO™ Toolkit - Deep Learning Deployment Toolkit. Package features latest version 2020.1.

If you’d like to experiment with deep learning on Intel Neural Compute Stick, you’re now able to do so directly on OpenWRT-based devices. You can find more info about NCS and OpenVINO™ Toolkit on Intel NCS 2 website. The package is also compatible with older Movidius Neural Compute Stick.

The dldt package is for developers and contains libraries and samples. OpenCV v4 package is required as a dependency.

Package is available in our package feed repository located at:
https://git.octagon.cz/openwrt.git (available only via git clone)

You can consume this feed by adding the following line to your feeds.conf file:
src-git octagon https://git.octagon.cz/openwrt.git

After updating your feeds, you can find the package located in menuconfig group Libraries -> dldt . Samples and Includes installation options are available for the package. We’re not providing binary version of the package at the moment.

You can build the package by running:
make menuconfig (optional to select samples or includes)
make package/feeds/octagon/dldt/compile

Enjoy the package and feel free to share any feedback with us!

:wine_glass:

If you’re curious, we ran one example from Intel NCS2 website on Movidius NCS USB stick plugged to Turris Omnia running Turris OS 4.0.5.

The command was:

benchmark_app -i ./president_reagan-62x62.bmp -m ./age-gender-recognition-retail-0013.xml -api async -d MYRIAD

with the following result:

[Step 11/11] Dumping statistics report
Count: 9120 iterations
Duration: 60020.99 ms
Latency: 13.11 ms
Throughput: 151.95 FPS
Peak Virtual Memory (VmPeak) Size, kBytes: 37004
Peak Resident Memory (VmHWM) Size, kBytes: 25608

Feel free to post your performance stats for comparison.