Hey Embedded Machine Learning Folks,
So on thing we've kind of struggled to find is a full suite of tools for doing embedded machine learning directly on microcontrollers. We've been mainly focused on leveraging neural inference accelerators, so we haven't tried these yet. So we'd love any/all input here, and without further ado, here's what we've found so far:
- uTensor - This seems one of the more complete. It looks to work on a TON of boards (here)
- Microsoft Embedded Learning Library - Seems to be mostly focused on the Raspberry Pi. Also haven't tried it yet.
- EMDL - A consolidation of notes/capabilities/software suites for machine learning, with the target being aiding embedded application.
- SOD - Looks pretty feature-complete, but can't yet figure out which boards/architectures/etc. are natively/easily supported. It looks relatively straightforward to use, however.
- OpenMV - Always a popular choice for embedded computer vision - and I think they'll continue to be. They leveraged and made it easy to use the CMSIS kernel from ARM. Probably a great (the best) way to start WRT microncontroller-only - as its the most straightforward to get all the way to a working (and cool) solution (see here and here and here.
Any others? Feel free to add!