"ml_training_data_full"

"Charge sharing between adjacent pixels in CdZnTe (CZT) photon-counting detectors degrades energy resolution and reduces spectral imaging performance. We present a complete machine learning framework for charge-sharing correction that recovers true photon energy from measured pixel charges. Our physics-informed multi-task neural network simultaneously predicts corrected energy, sub-pixel interaction position, and event classification. Using synthetic training data generated with Allpix$^2$ Monte Carlo simulation, we demonstrate a 52.9\\% reduction in energy RMSE compared to simple charge summing, achieving sub-100~$\\mu$m position resolution and 98.5\\% event classification accuracy. The framework provides an end-to-end pipeline from simulation to deployment, addressing a critical gap in photon-counting detector development. "