FYI: How to build tensorflow-lib (libtf.a) for yourself

Just putting this out here, because the building instructions on the github page did not work for me, yet this did.

Requirements:

  • Ubuntu 18.04, 32-bit
    • I tried 20.04, but the “GNU Arm Embedded Toolchain” ppa does not support it.
    • I got very strange compilation errors on Ubuntu 18.04 64-bit, but I think it could be because I built it with python 3 instead of python 2.
  • also make, curl, g++, git

After installing these, follow the instructions on the github page (and run ./make.py twice).

We will be updating the lib soon. Just gotta focus on IDE updates.

Cool, what will be in the update? I’m working on implementing the tflite_detection_postprocess op registration.

Edge Impulse has bounding box detection. So, we will have the ability to run models with that.

Any updates / ETA on that?
I’ve seen some activity on the tensorflow-git github

I’m almost done making the new library. Then I have to just tidy things up and integrate into main.

1 Like

Edge Impulse doesn’t have the pipeline on their website just yet but I hope to have the library available for what we need.

Any news regarding this? I haven’t seen anything related in the release notes

The library actually compiled if you follow the build instructions on the repo. However, the version of tensor flow we linked to is the latest and it has segfault bugs in it. I have to fix that… but, I had another higher priority to tend to.

1 Like

These are amazing news for the new runtime! We really hope that it could overcome all our present limitations by using our custom tensorflow models!