While we were following the instructions to host a local FTC-ml toolchain (so we do not have to use the google servers), we met a paywall. Is there any way to by it?
Thanks!
You do have to use Google Cloud services, the source code is not compatible with other services.
You do not specify the paywall that you encountered, nor the process you followed to get there. More information is necessary to help you.
-Danny
We have met the paywall when setting up the google cloud project, you have to set up a billing account and input credit card info.
That’s right. Google does not allow you to create Google Cloud projects without a billing credit card on file and the ftc-ml software is heavily dependent upon Google Cloud services. You can not get around that.
Then is there a way to create and generate our own .tflite model and add the metadata to be compatible with FTC? So far we are generating files like this, but it does not seem to work. When we are comparing it to the .tflite files created by the toolchain they are almost identical.
Yes, every FTC team has access to the ftc-ml Toolchain. We do not provide support for using other tools.
-Danny
@Toby , I have been able to create a TFlite model outside of the FML Toolchain…leveraging the awesome work the FML Toolchain team has shared in Github repositories. The key is the tensorflow version used and adding the metadata after exported.
Note it does take a PC with a graphics card and it is not an easy setup.
With three teams I have yet had time to post an understandable set of scripts to be repeatable by others. I was creating a tflite model outside of the FML tool so I would not use up the time for my team’s members. I would highly recommend teams use the FML toolchain.
Coach Breton
1 Like
Can you provide any more details about how you were able to train a model locally? I’ve been trying for a few weeks with no success because the site has been broken for our team.