Implementation of DL models (Tandem Neural Networks and Conditional Generative Adversarial Networks) for photonic surface inverse design used for comparison with the Multi-fidelity (MF) ensemble framework detailed in https://github.com/lukagrbcic/MFEnsemblePhotonicSurfaces and https://arxiv.org/abs/2406.01471.

---

Dataset Details

---

The dataset used to train the models can be downloaded at: https://osf.io/dwgtf/

The dataset should be put in the main directory in a folder named inconel_data.

Further dataset details are given in the MF ensemble github repository: https://github.com/lukagrbcic/MFEnsemblePhotonicSurfaces

---

Tandem Neural Network Details

---

Tandem Neural Network (TNN) implementation can be found in the TNN folder.

To reproduce the training, testing and the postprocessing of the TNN model, it is neccessary to run train_test_tnn.py script.

The forward Deep Neural Network (DNN) training is given in the script train_forward.py.

The DNN architecutre details for both the inverse and forward DNN are given in the inverse_forward.py file.

---

Conditional Generative Adversarial Networks Details

---

Conditional Generative Adversarial Networks (cGAN) implementation can be found in the CGAN folder.

To reproduce the training, testing and the postprocessing of the CGAN model, it is neccessary to run inconel_gan.py script.

The Generator training is given in the script train_generator.py.

The Generator and Discrimnator architecutre details are given in the generator_discriminator.py file.