Question about the required depth maps precision

I ran some data I had through BadSlam, and the results were mediocre.
The reason is that I used predicted depth maps, which means they are far from perfect.
I so wanted to know if there is a -known or estimated- threshold about depth maps metric precision under which the algorithm performs really badly.
Thanks in advance, and great work!