RJ Ascani

> Does the CI require a rebase instead of a merge? No, merge is fine.

@ericheim - Just as an FYI, I'm interested in merging this, just working out the ramifications of pinning against a specific version of the tensorflow.

> The pinning of the tensorflow version can probably be done somehwere in a central place. Understood, but we have to consider that not all users will be using the...

@vp-cad - The formatting errors can be fixed with: ``` tensorflow/lite/micro/tools/ci_build/test_code_style.sh --fix_formatting ```

> Should I used this existing function? Should I remove it and implement a new one based on the SumOrMean implementation? I would suggest re-using the QuantizedReduceProd in the reduce.h...

You are correct that TFLM does not currently support dynamic range quantization. This has primarily been because the microcontrollers we have supported lacked hardware floating point acceleration, so inference performance...

@AdrianLundell - is this fixed now? If so, please close.

This is by design. The `GreedyMemoryPlanner`, like several other TFLM objects, is designed to be allocated from the arena provided to the MicroInterpreter. It is typically allocated here: https://github.com/tensorflow/tflite-micro/blob/main/tensorflow/lite/micro/micro_allocator.cc#L88-L90. Notice...

> I would prefer option 1 as its very easy to skip the disclaimer and look straight at the numbers Do we have a pointer for the person detect model...

Thanks for filing the issue. This is indeed an interesting problem. The typical approach for using accelerators with tflite-micro has been to use a custom op. I'd recommend taking a...