Reproducing the paper results

[TerezaSikova]

Hello, thank you again for your amazing work!

I've been working on reproducing the results from your paper and was able to achieve similar performance using fragments from the Breaking Bad dataset. However, I'm having trouble obtaining good results on the RePAIR fresco fragments.

Would it be possible for you to share a recommended set of parameters that work well for this type of artifact? I’ve tried parameter tuning, but so far without much success.

Additionally, I have a question about the evaluation metrics. In your paper, you mention using Relative RMSE (R) and RMSE (T), but I'm unclear on whether these were computed using rms.py from the evaluation_modules, or if custom evaluation code was used. Any guidance you could provide on this (how to get this evaluation) would be greatly appreciated.

Thank you again for your time and support!

[freerafiki]

Hi Tereza, the parameters should be the one included, but we can try to check if something changed, since the repair project data was post-processed a bit to refine it and this might also have changed some of the characteristics.
When you refer to the RePAIR fresco fragments, which one are you using? From the zenodo dataset? which group?

About the metrics, I need to check since it has been some time, but I think yes, the metric should be the one in this codebase in evaluation_modules (rms.py). Is there a specific problem with that evaluation?