Evaluation of the CaloINN generative network - Fast detector simulation

These are the sampled used in the paper "Normalizing Flows for High-Dimensional Detector Simulations" for the evaluation of the CaloINN and the CaloVAE+INN models. Each zip file contains the generated showers stored in the CaloChallenge format together with figures of high-level observables. We include the layer energy deposition, the center of energy, the width of the center of energy, the layer sparsity, the ratio E_tot/E_inc, and the full voxel energy distribution. For these features we provide: energy inclusive histograms for all the datasets; for dataset 1, three incident energies: 256 MeV, 8.192GeV, 262.144 GeV; for dataset 2 and 3, three incident energy windows, (1-10), (10-100), (100-1000) GeV; separation power between Geant4 and our samples calculatd from the histograms.本数据集为论文《归一化流用于高维探测器模拟》（Normalizing Flows for High-Dimensional Detector Simulations）中用于评估CaloINN与CaloVAE+INN模型的采样数据。 每个压缩包均包含以CaloChallenge格式存储的生成粒子簇射，并附带高阶物理可观测量的可视化图形。 本次数据集涵盖以下物理特征：层状能量沉积、能量质心位置、能量质心展宽、层稀疏度、总能量与入射能量比值（E_tot/E_inc），以及完整体素的能量分布。 针对上述特征，我们提供如下内容： 1. 所有数据集的全能量范围直方图； 2. 针对数据集1，包含三种入射能量：256 MeV、8.192 GeV、262.144 GeV； 3. 针对数据集2与3，包含三个入射能量区间：(1-10) GeV、(10-100) GeV、(100-1000) GeV； 4. 基于直方图计算得到的Geant4模拟样本与本次生成样本的分离能力。