I have an OpenMV cam H7 (non plus) and I am trying to deploy a model trained with Edge Impulse. I have tried different models and got a model
small enough that can fit in the device. I have used a small image input (48x48) and a mobilenet V2 0.1.
I copied the tensorflow lite and labels files to the SD card and run the python code, but I always get OS error and no other info:
Traceback (most recent call last):
File “”, line 21, in
MicroPython: v1.15-r57 OpenMV: v4.1.1 HAL: v1.9.0 BOARD: OPENMV4-STM32H743
Type “help()” for more information.
Hi @mvitswd this could be a tricky one for two reasons:
We don’t officially support the non plus version (our documentation calls for the plus version)
The transfer learning models that we support work best with 96x96 images
I have a digits recognition model that I’ve downscaled to 48x48 MobileNetV2 0.1 and tested it on the Portenta H7 with Vision Shield, and real live performance is poor compared to the 96x96 MobileNetV2 0.35 version of the model which works really well. So, I’m concerned that even if you’re able to get it working on the H7 (non-plus) version that you’ll still not be happy with the end result.
I don’t have an H7 (non plus version) to test it with but I can ask around around to see if others from our team have experience with this issue.