The proposed hybrid model aims to deliver both aberration-free in-focus amplitude and phase reconstructions, while accurately predicting in-focus distances, from out-of-focus holograms. The tasks were handled independently with the aim of later merging them.
The development began with a 7-category ResNet model for hologram processing, which includes a Fourier spectra amplitude layer. This was expanded to a 21-category regression model, followed by transfer learning to create a final regression model with added dense layers.
For the image-to-image (hologram-to-reconstruction) task, a U-Net architecture was developed, leveraging the CNN from the regression model.
Trained models are available at: https://www.kaggle.com/models/mariareyb/autofocusing-model-for-dhm-holograms.
For more details on these models, please refer to the following publications. These are also the recommended citations if you use this tool in your work. (Pending Publication)
Copyright 2024 Universidad EAFIT Licensed under the MIT License; you may not use this file except in compliance with the License.
Applied Sciences and Engineering School, EAFIT University (Applied Optics Group)
- Maria Paula Rey (mpreyb@eafit.edu.co)
- Raul Castañeda (racastaneq@eafit.edu.co)