Running AI Models on Embedded Systems -- 2

AI/ML/CV/OCR Project:

We have two models built separately one is for OCR/opencv in python , second is CNN Model:
• MobileNetV2 96x96 0.35: The final layer consists of 16 neurons with a dropout rate of 0.1.
• Another MobileNetV2 96x96 0.35 Variant:

CNN Model: Accuracy - 72%, Loss - 0.56
• MobileNetV2 96x96 0.35: Accuracy - 74%, Weighted F1-score - 0.74






What we need you to do :



We need to run these models on a embedded system (adruino/Raspberrypi etc) ,

The flow of the program is :

One frame is captured via camera

Frame is pre processed— we have this done, our model does a little pre processing but we need you to enhance it to make it better.

OCR is run easyOCR , alphabets and numbers are grouped to represent values (our model does not locate ROI) around 5-6 regions of interest — need to add a validation step here as well.


Then the second program runs a set of pre processi On the same image then runs by those models (you need to increase accuracy of the model).

All this to be stored locally on embedded device and updated on cloud db as well.

You need to optimise/merge both models together to get best efficiency and model should be optimized for embedded systems like embedded c or c+ etc.


Need someone to handle this end to end.