Inference problem AM68A

[kira9k]

Question/Issue:
Hello! I have a problem.
I’m using an SK AM68a board from Texas Instruments(PSDK version is 08_06_00_10). On the edge impulse website I can save the model in the tidlrt for am68a format. To compile tidl rt into an executable file I use instructions from your github (GitHub - edgeimpulse/example-standalone-inferencing-linux: Builds and runs an exported impulse locally (Linux)). For model inference I use Python sdk (GitHub - edgeimpulse/linux-sdk-python: Use Edge Impulse for Linux models from Python). When I try to output the result I get nothing. I’m attaching a screenshot. After the inscription MODEL: /home/root/example-standalone-inferencing-linux/build/model.eim nothing happens. (here I used nasnetmobile for image classification).

Project ID:
339533
Context/Use case:

[Eoin]

Hi @kira9k

You are passing the incorrect data to the eim runner:

See: linux-sdk-python/examples/custom/classify.py at f1d7fa8944bbd35e6458bea228f225109a93f604 · edgeimpulse/linux-sdk-python · GitHub

Read through the following guide on eim - Linux EIM Executable | Edge Impulse Documentation
Best

Eoin

[kira9k]

Hi @Eoin
I made a text file with the effects as you said. for a model that was compiled into aarch64 format, everything works. But for models using tidl rt nothing has changed. The screenshot shows how I first used the mobilenet model in .eim format without using an accelerator, and then I did the same for the model using tidl rt.