Detectron2 for Active Learning in Object Detection

Usage

1. Clone the repository with all the submodules:

   git clone --recurse-submodules [email protected]:lolipopshock/Detectron2_AL.git

2. Install dependencies:
   1. 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

   2. Installing other necessary dependencies:

          pip install -r requirements.txt

   3. Installing UI components

      cd src/label-studio
      pip install -e .

3. Setting up the label-studio server and modeling backend
   1. Initialize the labeling server (If your image folder is ./data)

      label-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

      And you can start the server via

      label-studio start labeling/tk-labeling
      

   2. Initialize the model backend server

      label-studio-ml init labeling/backend_model --script extra/backend_model.py

      And similarly, you can start the backend server by

      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

4. Start using active learning for annotation