Since we can now easily impot and export custom settings I thought I’d share my settings for a more detailed preview reconstruction.
What you will get:
Preview Reconstructions between 250k and 1.2M tris. Typical between 500k and 800k.
Much better than the usual ones.
These take not much longer to calculate than the normal preview and the quality is somewhere inbetween default preview and normal reconstruction.
With these settings, normal reconstruction is identical to high.
Don’t forget to back up your settings before applying these.
And better do not touch all of them until you will completely understand all mathematical models underlayed.
What i can say. For perfect align and undeistortion 100 tie points per image is enough for perfect data. 1000 for most good stuio works. And its count connected to Preselectors, 10K default can have enough range for 98% of all scans.
But you set preselectors 1000 times more. For non professional cameras like iPhone or smartphone, preselector count more than 10K faster way for bad alignment, ghosting etc.
400MLN Features!!! From perfect texture rich real world image you rarely can find more than 400K features with Ultry detector sencitivity. But you set 1000 times higher! Do you think all pixels on your image are features? Per MPX same…
So your result for most newbies will required tons of time and tons of memory for bad result from first steps.
DistortionModel… Even non professional iPhone camera did not required such complicated Distortion model. THis models for really bad, probably hand made from glue lenses
I just included whatever settings I had at the time when I saved them.
The primary tweaks are for reconstruction anyway
Alignment settings for me change all the time depending on the source material I have available.
For some of the source material I had massively better alignment results when cranking up those values you complain about.
Even if the settings are controversial, at least it gets a discussion going and can give other users some idea of what they can try!
I would not be surprised if there are much better results for some projects with unusual settings.
Not all images can be perfect and taken in studio conditions - real worls sometimes throws weird problems at you.
I have not tried it yet but I was going to experiment with much higher features but much lower max reprojection error.
I find that 2 is quite high, especially for “not ideal” images, 0.5 is what I am usually aiming for if I can influence it properly.