Skip to content

Build a Custom rapidocr Wheel

RapidOCR now uses pyproject.toml + setuptools-scm to build wheels. The build environment requires Python >=3.9; the resulting wheel can be installed on Python >=3.8.

1. Enter the Python package directory

1
2
git clone https://github.com/RapidAI/RapidOCR.git
cd RapidOCR/python

2. Install the build tools and asset preparation dependencies

1
2
python -m pip install --upgrade pip
python -m pip install build setuptools wheel setuptools-scm PyYAML

If you also need to run tests or prepare the default models, you can install the full dependencies:

1
python -m pip install -r requirements.txt

3. Prepare the default model assets

1
python tools/prepare_wheel_assets.py

This step will:

  • Derive the default Det / Cls / Rec models from rapidocr/config.yaml
  • Read the download URLs and SHA256 values from rapidocr/default_models.yaml
  • Download or verify the model files
  • Generate MANIFEST.in to make sure these models are included in the wheel

If you only want to check whether the models are ready:

1
python tools/prepare_wheel_assets.py --check

4. Specify the version and build the wheel

1
SETUPTOOLS_SCM_PRETEND_VERSION_FOR_RAPIDOCR=3.1.0 python -m build --wheel

The build output is written to:

dist/

For example:

dist/rapidocr-3.1.0-py3-none-any.whl

5. Verify the wheel version

1
unzip -p dist/rapidocr-3.1.0-py3-none-any.whl "*/METADATA" | grep "^Version:"

Expected output:

Version: 3.1.0

6. Verify that the models are included in the wheel

1
python -m zipfile -l dist/rapidocr-3.1.0-py3-none-any.whl | grep "rapidocr/models"

You should see the default model files.

If you do not specify a version manually, you can also build from a Git tag:

1
2
3
git tag v3.1.0
python tools/prepare_wheel_assets.py
python -m build --wheel

When the current commit is exactly on the v3.1.0 tag, setuptools-scm derives version 3.1.0. If the current commit is not on a tag, a development version is generated.

Comments