Integrated-Photonic-Encoder-for-Terapixel-Image-Processing

Offical repository for "Wang, Xiao, Brandon Redding, Nicholas Karl, Christopher Long, Zheyuan Zhu, Shuo Pang, David Brady, and Raktim Sarma. "Integrated Photonic Encoder for Terapixel Image Processing." arXiv preprint arXiv:2306.04554 (2023)"

Jupyter Notebooks

train_decoder.ipynb: Construction and training of a neural network optimized for a compressive ratio of 8:1. The network takes compressed data as input and outputs original images.
demo_decoder.ipynb: A guide on using a pre-trained model to reconstruct the original images from their compressed versions.

Files in data/

File formats

• Ground truth data
  • dimension: 512 $×$ 512 (H $\times$ W)
  • format: PNG
  • total number: 19
• Compressed data
  • dimension: 64 $×$ 64 $×$ 8 (H $\times$ W $\times$ C)
  • format: .npy
  • total number: 19

The corresponding compressed data of "000005_10.png" is "000005_10.npy", and the rest files follow the same pattern.