Camera hardware confusion

[labaxisAI2]

I am curious about my Q1615-LE Mk3 box camera.
The product should have a ARTPEC chip with a DLPU, this is similar to my Q3628-VE Dome Camera (perhaps a different ARTPEC version)
https://www.axis.com/products/axis-q1615-le-mk-iii#technical-specifications

I was able to run the pose-estimator with flask on both cameras, however I noticed that my box camera received an error:
The requested image's platform (linux/arm64/v8) does not match the detected host platform (linux/arm/v7) and no specific platform was requested
When I change my scripts to the ARCH=armv7hf, and env.armv7hf.tpu config, the box camera runs the pose-estimator well. I thought this was odd since the chip and unit was artpec and DLPU.

My main issue is now I am receiving a fatal error attempting to run my Custom .tflite model (yolov5) Could not load model: Model is not compiled for EdgeTPU. I was able to adapt the pose estimator setup/scripts to run a yolov5 model almost flawlessly on the dome camera.

Have I overlooked something while setting up my camera or environment?

Edit: The box camera with env.armv7hf.cpu config "works", however inference speed is very slow at 1600ms. So does not take advantage of DLPU and artpec chip

[Corallo]

Hi @labaxisAI2

You can find more information about the products here
The difference is that Q1615-LE Mk3 is an Artpec-7, while Q3628-VE is an Artpec-8.
Those two families have both different architecture and DLPU capability.
I see you already solved the architecture problem by rebuilding for armv7hf.

Regarding the model compatibility, you can see here and here for more details on those DLPUs.
Yolov5 has an export flag to obtain a edgeTPU model (compatible with Artpec-7) but expect worse performance compared to Artpec-8.

[labaxisAI2]

Thank you for the information, I was using this as reference and that page was not as clear as the link you provided.