Loading a custom tflite model having image input and list of coordinates as output

I’m trying to deploy a custom tflite model that accepts facial images as input and returns 106 coordinates of facial landmarks. the existing tf_model object has the following functions
classify(), detect(), segment(), regression().
None of this takes image as input and returns nd array as output.
Is there any way in using this tflite model like how we use tf interpreter or how to do this with existing functions.

We don’t currently support that type of model. Does it even fit in RAM?

Classify would be the closest to what you want. However, it expects a linear output. It’s pretty trivial to add an ND array output to the code base.

If you can show the model would fit in RAM and is runnable I can extend the library to do what you want.