Using tensorflow machine learning

RoboCupJunior Rescue RoboCupJunior Rescue Maze
#1

Hi.

I read the thread (Use of Huskylens to Detect Victims - #8 by Matej), so let me ask you a question that has made me uneasy.

We use machine-learning (ML) to identify H/S/U victims. The ML model is made by Python and Tensorflow library. Tensorflow is universally used and it is NOT dedicated to the cameras we use (UnitV AI camera).

These are pictures of the ML model on training.

スクリーンショット 2023-03-15 21.04.32
スクリーンショット 2023-03-15 21.04.321134×792 191 KB

スクリーンショット 2023-03-15 21.04.46
スクリーンショット 2023-03-15 21.04.461314×712 77.2 KB

Of course, all images to train the machine ML model are taken by us. We also did NOT use the ML model generation site of M5 Stack (the company that manufactures the UnitV cameras).

I believe what we do is permitted because we’ve created our ML model and written our code by ourselves. Is this actually allowed?

I would like to know the official answer of the committee as soon as possible.

#2

Hi @shirokuma89,

First of all thank you for this question!

This is exactly what we expect from teams to do when trying out ML solutions! Primarily:

  1. Create own dataset.
  2. Come up with own NN architecture.
  3. Train the network and tune hyper-parameters.

So, all I can say is great job, this absolutely is allowed.

Best,
Matej
2023 committee

P.S. don’t forget to include this graph and whole proccess in your TDP as well! It is very nice to see :slight_smile:

#3

Thanks for your fast reply!!!