Hi, I’m a noob struggling to use segmentation model which isn’t supported by edge-ai.
The edge-AI model maker only supports limited amount of classification and detection models.
Is it possible to use DL models which are not officially supported ?
Yes, it is. I haven’t used their modelmaker tools but they aren’t necessary for DL inference.
TI has an index of officially supported models here: GitHub - TexasInstruments/edgeai-modelzoo: AI / Deep Neural Network Models and Examples.
You’ll need to scrape together scripts and samples, but if your chosen architecture is one of the ones they officially support for the runtime it shouldn’t be too bad.
If you are looking for an infrastructure that isn’t officially supported by the edgeai tools at all, it’s still possible to compile arbitrary models but it’ll require more work.
Thank you for the comment, kaelinl. I’ll guess I have to put more efforts into compiling external models.