Installation

Install the package

pip install ofiqpy

or from source:

git clone https://github.com/AVHBAC/ofiqpy
cd ofiqpy
pip install -e .

Dependencies are pinned to OFIQ v1.1.0's toolchain for bit-faithful model inference:

Package	Version
numpy	>=1.23,<2
opencv-python-headless	==4.5.5.64
onnxruntime	==1.18.1

Provide OFIQ's models

ofiqpy reproduces OFIQ's algorithms but does not bundle OFIQ's model files (they ship with OFIQ and may be licensed separately). Obtain them from an OFIQ install and point ofiqpy at the data/ directory:

git clone https://github.com/BSI-OFIQ/OFIQ-Project
# download OFIQ's models per its instructions, then:
export OFIQPY_OFIQ_DATA=/path/to/OFIQ-Project/data

The data/ directory must contain ofiq_config.jaxn and the models/ tree (face_detection/, face_landmark_estimation/, head_pose_estimation/, face_parsing/, face_occlusion_segmentation/, sharpness/, no_compression_artifacts/, expression_neutrality/, unified_quality_score/).

(Optional) The conformance gate

To reproduce the ±1 conformance validation you also need a built OFIQSampleApp and pandas:

pip install "ofiqpy[verify]"
export OFIQPY_OFIQ_ROOT=/path/to/OFIQ-Project   # dir containing install_x86_64_linux/
python tests/gate_slice.py 100

Verify the install

python -c "import ofiqpy; print(ofiqpy.__version__)"
ofiqpy -i face.jpg -o out.csv