Getting started with the code

The training code lives at https://git.uwaterloo.ca/WATonomous/road-dataset, well actually the real training code lives in subrepos of that repo. Once road-dataset is cloned, we can get the submodules by running git submodule update --init --recursive

• In some instances the repository 'acar-net' may not have been pulled.
  • This can be manually manually cloned inside the road-dataset directory: https://git.uwaterloo.ca/WATonomous/acar-net

Working with ACAR-Net

Developing:

docker-compose up -d acar

Then you can connect to the container as usual via VSCode SSH Remote Containers

Just run training (from outside the container):

docker-compose run acar python3 main.py \
    --config configs/ROAD/SLOWFAST_R50_ACAR_HR2O.yaml \
    --nproc_per_node 1

(or from inside the container after doing an up):

python3 main.py \
    --config configs/ROAD/SLOWFAST_R50_ACAR_HR2O.yaml \
    --nproc_per_node 1

The data is mounted at /road in the container. You can use VSCode to add this dir to the file explorer File -> Add folder to workspace

SETUP NOTES

If working on one of the WATonomous machines, the road-data has already been prepared in /mnt/wato-drive/road