Code for paper 'From Disorder to Design: Physical Mechanisms Governing Generalization and Hallucination in Deep Learning for Imaging Through Scattering Media'

This capsule contains the source code and simulation environment for research on imaging through scattering media, integrating a physics-based optical simulation with deep learning reconstruction frameworks. It features a MATLAB-based module that models coherent light propagation using Fresnel diffraction to generate large-scale training datasets, alongside TensorFlow implementations of two neural network architectures: a classic U-Net and a Conditional GAN (cGAN) equipped with a custom hybrid loss function (Adversarial + PCC) for high-fidelity image recovery. The repository includes complete pipelines for data generation and model training, as well as interactive GUI tools for inferencing and visualizing reconstruction results.