Markus Vieth

> I wonder if it's due to float? 1000 times might result in some small errors not resulting in any difference in float representation. Maybe, although it doesn't sound like...

I took a quick look at the program and activated more logging with `pcl::console::setVerbosityLevel(pcl::console::L_VERBOSE);`. The number of correspondences is the same for meter and mm (3912), but for the meter...

My current state of the investigation is that the function which the BFGS algorithm tries to minimize might be the problem. The transformation is described with 6 values, 3 values...

@keineahnung2345 Thank you for investigating! I haven't looked much into `IterativeClosestPointWithNormals`, only GICP, but it is good to know that there does not seem to be a problem there. Although...

@keineahnung2345 Sorry for the late reply! It is good to hear that `pcl::IterativeClosestPointWithNormals` works fine if supplied with adequate normals (and parameters). > @mvieth I've added debug message of loss...

> I use `PCL_DEBUG` not `PCL_VERBOSE`, do we need another function `pcl::console::isDebugLevelEnabled()` or just using `pcl::console::isVerbosityLevelEnabled()` would suffice? Sorry for the confusion, here is the doc of that function: https://pointclouds.org/documentation/namespacepcl_1_1console.html#a59064b5f6807552c0bac04377b1d1d0c...

> For point cloud of size 3912, it takes about 137.8 ms for one iteration averagely. That's quite long, but if we check the verbosity level first, that should be...

I find it difficult to image how this problem would occur in practice. It sounds like it could theoretically happen, but I am not sure which inputs specifically would lead...

Sorry for the delay, and thank you for your investigations. I am still looking into the problem, but I am not so familiar with NDT, so it is taking some...

@christian-rauch Could you explain what this pull request is intended to fix? For example, do you experience a build error without this change? I don't really see why `QUIET` and...