performance evaluation of yolo models on different edge platforms

[Ritesh1]

How to do benchmarking of YOLO model on stm32 platforms.

Ritesh

[Julian E. · Accepted solution]

Hello @Ritesh1 ,

You can benchmark YOLO models with the ST Edge AI Developer Cloud.

You first need to save your model as Keras, ONNX or TFLite (.h5, .hdf5, .keras, .onnx or tflite) and then follow this tutorial:

https://wiki.st.com/stm32mcu/wiki/AI:Getting_started_with_STM32Cube.AI_Developer_Cloud

I believe that you can find a .h5 for a yolo model in object detection in our ST Model Zoo: https://github.com/STMicroelectronics/stm32ai-modelzoo 

There, you can find it the pretrained_model folders, download it and use the ST Developer cloud to do the benchmarking.

Best Regards, 

Julian

[Ritesh1]

Thanks for the solution, I would also like to know how to change the resolution. 

Ritesh

[Julian E.]

Hello @Ritesh1 ,

In model zoo you can find 2 yolo architectures, with different resolutions:

If you want a resolution that is not listed, you will have to learn how to change the input size of a model, in a few words, it will consist in:

• Download a yolo model zoo model with a given input size
• Create a new input layer and replace it in the new model
• Fine tune the model
• Save the model
• Use model zoo as usual with your new model.

Depending on if the model was done with tensorflow or pytoch, the python code to do that differs.

You can find tutorials online to do that.

Have a good day,

Julian