Understanding the execution in depthai, openvino in VPU

sake

Hi I'm bit confused how all of this works.

For example with oak-1 that has myriad X, I run the feature_tracker.py example.
There is mention of HW_MOTION_ESTIMATION or lucas kanade.
I took a quick look into the sdk code both of those are executed in the VPU pipeline so they are both accelerated?

Is the FeatureTracker class based on neural network model or is it some other algorithm?
It's executed on the myriad shave cores or in the myriad cpu?

erik

Hi sake ,
Yes both of these features are HW accelerated. For FeatureTracker, please see docs here. It's run on the SHAVE cores. Thoughts?
Thanks, Erik

sake

HI erik,
That's cool.
At first I thought Oak-1 would only accelerate neural networks.
First search results seems to show that Myriad X is flashed with OpenVINO which seems to only deal with deep learning models.
But there are also some Intel SDKs so I assume the Oak firmware utilizes those.

erik

Hi sake ,
Yes OAK firmware (called depthai) utilized openvino. Besides deep learning models you can also run simple logic/calculations by "tricking it" it's an AI model (examples here). Thoughts?
Thanks, Erik