Hello!

We have recently started to work with the OAK-D PRO POE devices, we have been able to implement some example using DL models, but now we are facing a new goal.

In one of our cases we are exploring the possibility of incorporating an ML model into one of our OAK-D devices. The model is based on HOG (Histogram of Oriented Gradients) feature extraction techniques, coupled with an SVM (Support Vector Machine) algorithm for classification prediction. Specifically, we are interested in using this model within an OAK-D PRO camera.

However, we are currently facing uncertainties concerning the process of converting the ML model to the .bobl format, or even whether it is feasible to use such models with the OAK-D device.

We appreciate any advice or tips regarding this matter.

Thank you very much in advance!

  • erik replied to this.

    Hi Irena ,
    Whether it would work really depends on the layers you are using - the hardware doesn't support certain layers, but that rarely happens. I think the best way would be to try to convert the model to the blob, and you can see whether it throws and error. We have plenty of docs on how to do the conversion here:

        Hi erik,

        Thank you very much for your prompt response.

        OK! We will proceed with the conversion of our ML model and conduct testing. If any doubts arise, we will reach out to you for further clarification.

        Thanks 🙂