/TinyML-Contest-Solution

Code for the ACM/IEEE TinyML Contest at ICCAD 2022

File Structure

  • Folder for model training design: ./Training
  • Folder for C source code (including the whole project): ./Deployment
  • Trained model weights in tflite: ./Weight/best_weight.tflite

How to run?

Training

Set up environment

pip install tensorflow

and running

cd Training
python select_model.py

Deploy

For CubeMX

1.Import best_weight.tflite, use STM32RunTIME and low level compression  
2.In advanced setting, click on ONLY "Use activation buffer for input buffer" & "Use activation buffer for the output buffer"

For Keil Compiling Settings

1.Option for Target->Target->Code generation：Use default compiler version 6 & Use Micro LIB;
2.Option for Target->C/C++(AC6)->Optimization:-Oz
3.Option for Target->C/C++(AC6)->Click on "One ELF Section per Function" & "Link Time Optimization" & "Execute-only Code" & "Short enums/wchar"