tersekmatija / depthai-ml-training Goto Github PK

We at Luxonis encourage our users to try out RoboFlow for annotating, training and deploying their ML models, especially if you are new to the ML world. They have created a course for the DepthAI platform that is a great starting point for everyone. They also have two blog posts, quickstart guide and training custom object detection model with their platform, which are both tailored for DepthAI devices.

We want you to be able to train your own neural models (object detectors) for DepthAI.

This repository contains some pre-trained models under model-zoo and the Jupyter Notebooks we used to produce these neural models in colab notebooks.

Use the model-zoo models if we happened to already train something you need (and see here for models that Intel already trained that run on DepthAI already) and use the colab notebooks if you'd like to train something new or improve upon the training we've already done.

It's the platform for Embedded Spatial AI and Embedded AI. Four trillion operations on a tiny device which can ouput structured data of what is in the physical world - what it is, and where it is in meters - in real-time.

Google Colaboratory is a great tool to do so! It allows you to train neural models yourself using their fast GPU instances, and in some cases (depending on the dependencies of the training frameworks), even using the Google TPU - all for free!

It is exactly intended for this proof of concept and initial research. And if you hit the limits of the free account, you can upgrade to a Pro version of Google Colab for only $10/month.

You can browse (and open/use) all the Jupyter Notebooks in this repo by clicking on .