MKoumans/pidu-optical-parameter-extraction

PIDU: Physics-Infused Deep Unfolding for Optical Parameter Extraction

Trained checkpoints for Physics-infused Deep Unfolding for Automated Material Parameter Extraction of Optical Data (Koumans, Stevens, van Sloun, van Mechelen; Eindhoven University of Technology). The models extract Drude-Lorentz parameters (layer permittivity and, per oscillator, w0, wp, g) from infrared spectra of thin films.

Code: https://github.com/MKoumans/PIDU · Data: MKoumans/pidu-data

Checkpoints

One checkpoint per model type and use case (the released seed of the 8 trained):

Use case	CNN	DU (no physics)	PIDU
1	models/usecase1/CNN-t01-00.pt	models/usecase1/DU-t01-00.pt	models/usecase1/PIDU-t01-07.pt
2	models/usecase2/CNN-t01-00.pt	models/usecase2/DU-t01-00.pt	models/usecase2/PIDU-t01-03.pt

The matching training logs are in outputs/case<N>/logs/train_<TYPE>_<i>.log.

Usage

git clone https://github.com/MKoumans/PIDU.git && cd PIDU
pidu-hub download --what models logs --usecase 1
python examples/quickstart.py

Files keep their repository paths, so a download into the repository root puts them where the code expects them.

Citation

To be added.