- Clone the repository with all the submodules:
git clone --recurse-submodules git@github.com:lolipopshock/Detectron2_AL.git
- Install dependencies:
- Installing object detection environment with according to your environment
- The tested version of pytorch is 1.4.0 with CUDA 10
- And you must install Detectron2 with version 0.1.1. Newer versions has different APIs.
pip install detectron2==0.1.1 \ -f https://dl.fbaipublicfiles.com/detectron2/wheels/cu100/torch1.4/index.html
- Installing other necessary dependencies:
pip install -r requirements.txt
- Installing UI components
cd src/label-studio pip install -e .
- Installing object detection environment with according to your environment
- Setting up the label-studio server and modeling backend
- Initialize the labeling server (If your image folder is
./data
)And you can start the server vialabel-studio init labeling/tk-labeling \ --input-path=./data \ --input-format=image-dir \ --allow-serving-local-files --force \ --label-config=extra/config.xml \ --ml-backends http://localhost:9090
label-studio start labeling/tk-labeling
- Initialize the model backend server
And similarly, you can start the backend server by
label-studio-ml init labeling/backend_model --script extra/backend_model.py
label-studio-ml start labeling/backend_model # There's a relative import of the libraries # So you have to run this command in the project project # root path to avoid import errors
- Initialize the labeling server (If your image folder is
- Start using active learning for annotation