Training for non-image based models


#1

Hi Everyone,

Just wanted to know if there is a way for developers to use models that do not rely on computer vision; i.e. they relay on a set of inputs (number, arbitary ascii data) and return the probability of a match. I wasnt sure if there was an equivalent question in the community so thought ill start a thread on this

Currently most the function exposed by the KPU module in MAIXPy rely on image as an input.


#2

Similar discussion:


Yes, Currently only C based sdk supports non image data. Since Maixpy is apparently built on kendryte sdk, this function should be available soon.
For now, may be you should try the C sdk if you can.
example, https://github.com/kendryte/nncase/tree/master/examples/iris