Protocol for running Colmap

[btoussai]

Hello,
I have some questions regarding the generation of the input point cloud.

The paper makes it clear that the method takes as input a dense point cloud generated by Colmap, which is claimed to take approximately 2 minutes.
However, I tried running colmap's dense reconstruction on DTU scan 24 and it took more than 30 minutes (using a cuda-capable colmap version and a powerful workstation). Estimating per-camera depthmaps seems to take much longer than what is claimed in the paper.

Could you share the exact command line /steps that you used to generate the input point clouds?

I would also be interested in knowing the rendering speed and quality achieved by the method.
Could you provide some details regarding this matter?

Thanks for your time.