This directory contains scripts that generate NN Archives compatible with Luxonis devices for the custom models in this folder. The scripts utilize ModelConverter for the NN Archive generation and model conversion.
Uses torch.cat operation to concatenate multiple frames.
Uses kornia.filters.GaussianBlur2d filter to create blurring model.
Uses kornia.filters.Laplacian filter to create edge detection model.
Uses torch.sub operation to compute the difference between two frames. Can be used to detect motion in consecutive frames.
The model generation scripts require Docker to be installed. You can install Docker by following the instructions here. You must also install the required packages specified in requirements.txt file:
pip install -r requirements.txtYou can generate each model by running the corresponding script. The scripts accept the following arguments:
positional arguments:
{rvc2,rvc3,rvc4} Platform to generate NN archive for.
The generated NN Archives will be available in the out/shared_with_container/models directory.
python3 kornia_blur.py rvc2This will generate the NN Archive, that can be used on the RVC2 platform.
python3 pytorch_concat.py rvc4This will generate the NN Archive, that can be used on the RVC4 platform.