How to import YOLO OBB data into Labelstudio?

My objective is to import YOLO OBB data to check them and eventually correct them with labelstudio.

I tried the tutorial approach here but it does not work with my YOLO OBB data.

Have some tips to make it work to import YOLO OBB data?

Version of labelstudio : Label Studio version: 1.14.0.post0
Version of label-studio-converter: 1.0.7

Hello, right now LS doesn’t support YOLO OBB import in a native way.

  1. The main entry point for simple yolo import is here:
    label-studio-sdk/src/label_studio_sdk/converter/imports/yolo.py at master · HumanSignal/label-studio-sdk · GitHub
    (label-studio-converter was moved to LS SDK).

  2. However, we have this function to convert YOLO ML predictions to Label Studio:
    label-studio-sdk/src/label_studio_sdk/converter/utils.py at master · HumanSignal/label-studio-sdk · GitHub

  3. It’s used in Label Studio ML Backend for YOLO:
    label-studio-ml-backend/label_studio_ml/examples/yolo/control_models/rectangle_labels_obb.py at master · HumanSignal/label-studio-ml-backend · GitHub

So, you can try to combine SDK Converter from (1) and function from (2) to write your converter from YOLO OBB to Label Studio RectangleLabels with rotation.

1 Like

Thanks you.

I’ll code one adaptation to import YOLO OBB.

1 Like

I made the code and tested it.
It works fine.

See the branch
It works like :

python main.py import yolo_obb -i path_to_my_data -o path_to_my_output/output.json --image-root-url "/data/local-files/?d=my_datasets/images"

I did not add test function. It is in the agenda.

However, on which branch should I make the PR?
I did not see a develop branch.

1 Like