Deploy training to Oak-D on Raspberry Pi

Community Help
1

The Roboflow annotation and training went well! You have a great system here! Now I want to deploy this model to my Oak-D on a Raspberry Pi. The tutorial says this “is coming.” What can I do in the meantime? This guide says “easy” but it’s not easy for a beginner!

I’m stuck with trained data and not knowing how to get it into the Oak-D.
Is there another tutorial that I can use?

2

Hi,

There is a First Steps with DepthAI guide here that walks through getting DepthAI installed, so you can use the Linux instructions to get it up and running on the Raspberry Pi. From there, add the custom model blob so you can use your custom model: First steps with DepthAI — DepthAI documentation | Luxonis

You can train a MobileNetSSDv2 model to run on the OAK device via the connection to the Raspberry Pi. After training the model, save the model weights for use in the Raspberry Pi after conversion to a blob (within the DepthAI repository): MobileNetSSDv2 Tensorflow 1.5 Object Detection Model

https://docs.luxonis.com/en/latest/pages/training/#custom-training

https://docs.luxonis.com/en/latest/pages/model_conversion/

3

To run models directly on the Raspberry Pi itself, you can also use the Inference Engine (IE) which will utilize the IR model files: Use the Model Downloader and Model Optimizer for the Intel®....

A TensorFlow-trained model’s frozen graph can be converted to the pb-file necessary to create the bin and XML files for the IR format.

https://docs.openvino.ai/latest/openvino_docs_MO_DG_prepare_model_convert_model_Convert_Model_From_TensorFlow.html

4

The Docker load works for coco but not for my own trained model: It ends with Error 500, instead of the 200

“A TensorFlow-trained model’s frozen graph can be converted to the pb-file necessary to create the bin and XML files for the IR format.”
This is way too complicated.

Your responses are no doubt correct, but too involved for a noobie like myself. Sorry. I don’t have years of AI experience.
Yet.
But I have been using computers and programming since 1985.

5

Hi @Russ76 !

Can you share what you did understand from the responses, and we can go from there filling in any gaps? Or was it soley the part below we can rephrase or add more info?:

6

The entire process I don’t understand. Besides, the page you send me to has another link that I must use that goes to 404 not found!

I have to install “DL Workbench” and “Jupyter notebook” and “Blob converter”?

It’s so complicated.

Russ