Hello, I would like to train and run a custom network on my open mv camera but just following the basic tensorflow lite converter tutorial gives me a model incompatible with openmv. Is there are tutorial somewhere for building models outside of edge impulse with tensorflow?
Hi, we just support Edge Impulse as Google never actually provided any support. As such, they are the vendor to use.
Frankly, TensorFlow Lite for Microcontrollers was broken for Image Classification until Edge Impulse folks fixed it. I would not trust Google material on how to use their own library…
Ok, with that said, edge impulse thus far only supports classification problems. I would like to implement the project from https://github.com/yinguobing/facial-landmark-detection-hrnet.git, just generating and transmitting facial feature landmarks. Is this not yet possible with openmv or any other embedded platform yet?
Part of that is coming soon from Edge Impulse. But, Face Landmark is not here yet. Expect some announcements very soon.
We’re also going to be updating our TensorFlow library to support some new features.
Hi there, we had a similar issue as we wanted to use TFlite with the original Tensorflow runtime, as advertised in the product page. Any updates on that on when TFlite 2.4 could be implemented in the codebase? We appreciate your support!
I’ll be updating the TensorFlow package on the camera to support object detection per Edge Impulses updates. However, I’m working on frame buffer speed ups right now.