EfficientDet

You can read more about the EfficientDet model in automl's repo.

The NN model is taken from PINTOs model-zoo. In this experiment we have used EfficientDet-lite0, which is the most lightweight one.

Instructions on how to compile the model yourself:

Download the lite0 zip from PINTO's model-zoo
Navigate to FP16/myriad where you will find model in IR format (.bin and .xml)
Compile the IR model into blob (instructions here). I have used the online converter. Note here that the model's input layer is of type FP16 and you have to specify that as MyriadX compile params: -ip FP16

Demo

python3 -m pip install -r requirements.txt

If you install requirements, you will use the NN performance improvements branch. This helps with boosting the NN FPS for ~10%.

Run the application

python3 main.py