I could sucessfully install armnn and pyarmnn for 32 bit raspberry pi os and could run a tflite model. Is there a pyarmnn and armnn version that could be installed on 64 bit raspberry pi os. Please provide the link
Hi! Raspberry Pi OS 64-bit is still in beta, but you can try to use this guide to cross-compile ArmNN for it:
https://developer.arm.com/solutions/machine-learning-on-arm/developer-material/how-to-guides/cross-compiling-arm-nn-for-the-raspberry-pi-and-tensorflow
This guide involves building a few dependencies (boost, protobuf, Arm Compute Library), so at each stage it's necessary to pay attention to which compiler is used and which compiler flags are specified (for example, armv7a should be changed to aarch64).
Hi,
To add to what Pavel said - currently we are only releasing binaries for ArmNN, but no PyArmNN. We are planning to start releasing those in future as well.
The guide for building for 32-bit is the same as the one for 64-bit: https://github.com/ARM-software/armnn/blob/master/python/pyarmnn/README.md
Best,
Mina
There are pre-built binaries available for bullseye version of Debian, which you can download and install with apt-get... Unfortunately the latest Raspbian is still Buster, so if you want to try out armnn and pyarmnn, you could add bullseye to sources.list like that
deb http://deb.debian.org/debian bullseye main contrib non-freeand then install everything you need with
deb http://deb.debian.org/debian bullseye main contrib non-free
sudo apt-get update && sudo apt-get install python3-pyarmnn libarmnn-cpuacc-backend22 libarmnn-cpuref-backend22Obviously it's a hack and stability of a system is not guaranteed - so this "method" is only good for testing.
sudo apt-get update && sudo apt-get install python3-pyarmnn libarmnn-cpuacc-backend22 libarmnn-cpuref-backend22
Hi Mina,
a follow-up question about binaries. I see that pyarmnn, armnn and backends binaries are available now for install with apt-get on debian bullseye and ubuntu 20.04. Unfortunately, latest Rasperry Pi OS is still Buster, so installation procedure is either a)hacky, such as adding bullseye to sources.list b)drudging, cross-compiling from source.
I'm currently developing AI applications for Raspberry Pi CM based reTerminal. e.g.
https://twitter.com/HardwareAi/status/1397859833357889537
https://twitter.com/HardwareAi/status/1394929843674882048
and would like to know if you plan to release binaries for Buster as well - that would simplify installation procedure for Raspberry Pi users greatly. Do you need any help with that?