Label Studio annotation exporter

For labeling text regions and their contents we used Label Studio. AnnotationExport is an CLI script that builds datasets for TrOCR, YOLO and CRAFT out of LS annotations.

• Images are taken from an S3 bucket
• Annotations from Label Studio can be taken from both s3 and .json file
• Resulting dataset can be saved to both s3 and a local folder

Installation

1. Create an virtual environment (Optional)

python -m venv venv
# for windows:
./venv/Scripts/activate.ps1
# for linux:
source ./venv/bin/activate

Either:

2. Install package from GitHub

pip install git+https://github.com/DialecticalHTR/AnnotationExporter.git

3. Create .env in the folder you'll use the app, copy .env.template contents into it, add necessary data

or

2. Clone repository

git clone https://github.com/DialecticalHTR/AnnotationExporter.git

3. Install as editable package

pip install -e .

4. Rename env.template to .env, add necessary data

Usage

Parameters:

• --from source_type path: an annotation type to use and the path to data. Source type can be s3 or export (Local Label Studio JSON file). You can supply multiple annotations!
• --to output_type path: an output type and path to a place where dataset would be saved. Output type can be s3 or folder. You can have multiple outputs at the same time!
• --data model_type: dataset to generate. Dataset type can be trocr, yolo or craft. The default dataset is TrOCR.

Example:

anno-exporter --from s3 dialectichtr-data --from export 13.json --to folder output --data yolo