Data and Reproducibility Materials for HMF-PC-MTP: Multi-Fidelity Pareto Prioritization for Deep Neural Network Testing

The experiments cover CIFAR-10 and ISIC 2019 using ResNet-50 and ConvNeXt-Base under multiple class-balanced Top-K budgets and repeated experimental seeds.</b></p><p dir="ltr"><b>The CIFAR-10 experiments use the standard 50,000-image training set.

The revised ISIC 2019 experiments use the complete official 25,331-image training pool under an eight-class closed-set protocol. A separate held-out known-class test set of 6,191 images is used only for checkpoint generalization auditing and is not included in the primary HMF-PC-MTP prioritization statistics.</b></p><p dir="ltr"><b>The deposited materials support the main quantitative analyses, external baseline comparisons, ablation studies, robustness analyses, efficiency evaluation, and metamorphic-relation allocation analyses reported in the manuscript.