while rerunning RC on an identical dataset with an identical CLI Script the results are always slightly different and sometimes the difference is even so large that one result is absolutely great and the second one shows craters in the objects surface.
I clear the cache and reload the global settings at the beginning of each run to be sure to start on an identical initial point.
Is there a possibility to reduce this randmness, e.g. by giving a fixed seed or something similar?
Can I somehow surpress the randomness in any other way?
Each alignment is slightly different.
But the craters? Can you show what exactly do you mean?
If this is the project from camera rig and the camera positions are still the same you can use XMP workflow using locked position. In that case will be the each alignment same as the first one.
To do so follow this tutorial: Coordinate System Preservation Using XMPs or GCPs | Tutorial
It depends what you mean by that randomness.
Number of sparse points, orientation, precision, position?
Can you share some images of different results you are getting?
As your rig is not perfectly stable, that could be also the reason of your results.
By identical dataset do you mean the same images are used or the dataset captured by the same way?
No I’m really talking about binary identical images. The same code, running on the same data.
Here is an example image of the same part of an object which are reuslts of two runs on the same images. On the left you can see some craters, the object depicted on the right is nearly perfect.
The difference is sometimes even larger. In some cases even some cameras are not aligned and for the next rund, all cameras can be added to the component.
(im clearing the cache and the metadata in between)
Can I force RC somehow to use the same initial point. It would be much easier to compare different scans if their result would be repoducible.
You can try to use locked XMP settings if you are processing the same dataset more times. In that case the same alignment will be used also for your next project.
I had the impression that the “bad” results appeared more often on a weak PC that takes much longer for calculation compared to a run on better hardware.
Is the quality of the result somehow dependend on the hardware?
