CPU-Net – Cyclic Positional U-Net model, training data, and model weights for HPGe pulse-shape translation

CPU-Net is a CycleGAN-based framework that learns an ad-hoc translation between simulated and measured high-purity germanium (HPGe) detector pulses. Through a pair of Positional U-Net generators (REN & IREN) and attention-augmented RNN discriminators, the network replaces hand-tuned electronics-response models with a data-driven mapping that preserves detector physics. 

This repository contains the model weights (REN / IREN generators & discriminators) plus all pulse datasets used to train and validate CPU-Net. Real pulses come from an HPGe characterization run at ORNL, while synthetic pulses were generated with a matching Geant4 geometry and post-processed through siggen.

Full training scripts, notebooks, and updated README live on GitHub:
https://github.com/aobol/CPU-Net