Questions about settings in visual localization task on InLoc

Hello! Performance of DKM and ROMA reported in the papers on InLoc are admirable. However, settings are not detailed for me to reproduce the results. So, I want to ask for some suggestions for the reproduction. Are the matching and the sampling parameters kept as default? What are the recommended settings for quantization?