Running FOMO AD build using TIDL-RT Library (AM62A)

Help
#1

Hello,

I trained a FOMO AD model and chose the TIDL-RT Library (AM62A) build, Tflite classifier(float). Now, I have the files but how do I use/run the classification using a python application on my am62a board using the raspberry pi camera attached to it as the input which uses gstreamer as input and output streaming.

Also, is the inference time of 902ms too high for the board?

Thanks :slight_smile:

#2

Hi @tkapopr

The Linux Python SDK enables you to use GStreamer pipelines for capturing data from cameras.

You can define a GStreamer pipeline that captures frames from a Raspberry Pi camera or any other camera module and processes them using your model.

That does seem quite long, you can review the examples here to see if you have better performance with the examples:

Best

Eoin

#3

Hey, thank you for your reply. I also want to make some tests and use the tflite model for other datasets. I wanted to enquire if the scoring mechanism(patchcore/gmm) is a part of the tflite model that we download from edge impulse or is it applied externally during inference?

In short, if I am just using the tflite model and no other files, would it still have the scoring included or would I have to add it externally during inference?

Thanks and regards