problem with odometry in 0.96 step1

[yuckinus]

Running step1 0.96 on a cave scan (tiny spaces preset) I get widely different results then what I get with same settings in 0.95

0.95 looks closer to reality but there are still issues because of dataset. Will test on reference data set.

[bloom256]

Could you please clarify whether the behavior is stable in both versions?
Specifically: if you run step1 multiple times with the same settings and data, do the results look consistent run-to-run for 0.95 and for 0.96, or do you see variability between runs?

That would help understand whether this is a deterministic difference between versions or potential non-deterministic behavior.

[yuckinus]

Sure, will also get to that but it will take me days to go through the iterations. Don't have so much computing power.

Tested 0.95/0.96 on another "good" dataset and results looked almost identical. Have another dataset under test on 0.96 to compare. Will update when I have more info

[JanuszBedkowski]

I am running 10x on 095 and 10x on 096, give info ASAP

[JanuszBedkowski]

We do not support cave + forest scenario, please split data

[yuckinus]

finished another comparison 0.95/0.96 on reference dataset and results are very similar so I'm a little bit lost. will try your proposal next by processing multiple times the problematic dataset with both 0.95/0.96