Object Detection .tflite model on STM32H747-DISCO

TensorFlow General Discussion
, ,
1

Hi,

I’m trying to create an object detection application on the STM32H747-DISCO board. I’m trying to use the X-CUBE-AI tool to convert a given object detection model ‘ssd_mobielnet_v1_1_metadata_1.tflite’ so it’s usable by STM32CubeIDE, but that fails with the error message: 'INTERNAL ERROR: ‘TFLite_Detection_PostProcess’.

My question is, is it even possible to load an object detection model on to the STM32H747-DISCO board? Or is it worth it in the first place? I don’t need more than 2 FPS.

Thanks in advance!

1 Like
3

Hi @mvanwoensel,

As per the error message, the X-CUBE-AI tool is not supporting the Post Processing op ‘TFLite_Detection_PostProcess’. So you need to implement the CustomOp for ‘TFLite_Detection_PostProcess’ to run on MCU. The implementation of Custom OP guide you can find here(Please observe TFLite now become LiteRT). Let us know your response and issues on edge-AI-torch.

Thank You

1 Like