gokhankocmarli/holopaswin-v1

HoloPASWIN

A deep learning project for eliminating the twin-image problem in in-line holography using a physics-aware Swin-UNet architecture trained with synthetic holograms generated via the Angular Spectrum Method.

Find the source on GitHub: HoloPASWIN

Training Metadata

Find the training source code using the HoloPASWIN project's GitHub page.

Dataset: gokhankocmarli/inline-digital-holography
Data Files:
- Training: https://huggingface.co/datasets/gokhankocmarli/inline-digital-holography/resolve/main/data/000000-014999.tar.gz
- Testing: https://huggingface.co/datasets/gokhankocmarli/inline-digital-holography/resolve/main/data/060000-074999.tar.gz 
Base Model: microsoft/swin-tiny-patch4-window7-224
Samples:
- Training: 12000
- Validation: 3000
- Test: 2000
Epochs: 5
Batch Size: 8

Accuracy

========================================
   FINAL MODEL ACCURACY REPORT   
========================================
Evaluated on: 2000 samples
----------------------------------------
MSE (Mean Squared Error):     0.000642 (±0.001029)
SSIM (Structural Similarity): 0.9933   (±0.0059)
PSNR (Peak Signal-Noise):     34.01 dB  (±3.66)
----------------------------------------