Reproducibility of Results for ResNet-50

[scarrasco-az]

Dear authors,

After generating explanations for the four models presented in your results, I obtain similar results with VGG-16 for all eddits as well as with ResNet50 for the baseline. However I only managed to generate 625 counterfactual explanations with your method and ResNet50, with an average number of edits of 3.44. The evaluation results are:
Eval results single edit: {'Near-KP': 0.7672, 'Same-KP': 0.4288}
Eval results all edits: {'Near-KP': 0.6101537028411738, 'Same-KP': 0.32836516068933397}

Do you have any suggestions for how to resolve this issue? Thank you.