Code for using an Autoencoder network for reconstructing FTIR spectra. The repository comes with the following datasets:
- ATR Spectra from the siMPle library, found in the folder "ATR Spectra"
- µFTIR Spectra acquired in lab, found in the folder "MicroFTIRSpectra"
There are four different benchmark protocols that can be used to explore the autoencoder's performance.
- benchmarkATR.py -> Benchmarking on the ATR spectra
- benchmarkMicroFTIR.py -> Benchmarking on the µFTIR Spectra
- benchmarMicroFTIRFromATR.py -> Forcing poor selection of training spectra. Training on ATR Spectra, inference on µFTIR spectra
- benchmarkQuantitative.py -> Running peak area restoration test
The "ManuscriptImages" - Folder includes the code to generate the figures that were used in the publication.
In addition, the "Spectra Generation" folder includes code for training and using a Generative Adversarial Network for training spectra synthesis.