Tensorflow Lite on Omega2+
Joan Iglesias last edited by
Do you know if it's possible to use Tensorflow Lite on Omega2+? If not possible, any alternative?
I have seen two questions in the forum, but no clear answer.
@Joan-Iglesias as there is no other response, I'll give you my 2c worth, based upon my experience with tensorflow and my separate experience with Omega2.
While TensorFlow Lite (TFL) is targeted at low resource devices, ie IoT/Mobile devices, it is likely possible to run on Omega2, however I see a few challenges. 1) The list of supported platforms does not include MIPS, upon which the Oega2 is based. 2) Python3, required by TFL already eceeds Omega2 capacity so you need to extend the storage capacity via USB/SDCard, or use Python Light, but Python Light is "light", so it's missing some funtionality. from what I have seen, some of the missing libraries are written in C/C++ so you'd need to cross-compile this as required. 3) Since TFL has no wheel for MIPS, you'll need to build your own, this is documented on the TFL site but I've not attempted it thus far.
I played with TensorFlow on a Pi 4 (B) which is significantly more powerful than an Omega2, it gave me grey hairs. So my general conclusion is that TFL is not well suited to the Omega2, since Omega2 is designed as a low resource, low cost IoT device utilising a Linux like platform.