TextureQuality improvement

Comments

13 comments

  • Avatar
    Götz Echtenacher

    Hi Paul,

    the term texture quality can be a bit misleading - there is a post where I elaborated on this subject.

    Bottom lin is that it only means that your actual texel size is 100% the same as the ESTIMATED average texel size of the whole model. So it is mostly size related and also an average value.

    If you mean by quality rather blurriness or such, then you need to improve your alignment rather than the unwrap.

    0
    Comment actions Permalink
  • Avatar
    Paul Krist

    okay thanks for letting me know. I was referring to the blurriness of the texture. Since I am using over 500 4k images I was hoping for a better reult of my scan texture. How would I be able to improve on my alignment. Also I ran another unwrap with fixed texel size and created 6 8k tiles the quality now beeing 180%. This however according to you explanation doesnt necessary mean the result is better correct?

    0
    Comment actions Permalink
  • Avatar
    Götz Echtenacher

    Yes, it just means that your texture resolution is about twice as much as RC calculated it's optimum. The more complex the model, the more it can deviate from reality.

    You need to improve your alignment to avoid blurriness. There are plenty of threads here in the forum covering exactly that (possible key word: exif grouping). If you have any other specific question after reading those, I am more than happy to help!  :-)

    0
    Comment actions Permalink
  • Avatar
    Paul Krist

    thank you!

    0
    Comment actions Permalink
  • Avatar
    Paul Krist

    so my alignment report gives me quite good numbers according to the help page

    total projections : 12776755

    average track length: 5.070273

    Max error:1.999922

    Median error: 0.287917

    mean error: 0.373249

     

    I got 413 registered cameras with a lot of tiepoints (most of them over 10k tiepoints) see picture

    From your experience is there a chance to improve the blurriness of the texture with some tool

    0
    Comment actions Permalink
  • Avatar
    Götz Echtenacher

    Hmm, the numbers seem good indeed. Although they are not everything. After all the software is there to create the best numbers possible - whether it is the "real" solution or not...

    The fact, that basically all your tie point numbers are below 40k (the standard maximum) tells me that there aren't enough "good" ones to max the number out. Some are even below 9k, which is below the standard number of the top 10k that are selected from the 40k. So that means you can try to raise the preselector to 20k (which means 20k will be picked) and also raise the detector sensitivity (which means that weaker points will be used).

    Did I mention exif grouping already? I really recommend it, since your images seem to have large patches with no points.

    0
    Comment actions Permalink
  • Avatar
    Paul Krist

    Thanks for the advice. I will raise the preselector and detector sensitivity and run another test. Regarding the exif groupings everything was shot on the same camera and focal length. I also undistorted the footage already with the distortion grid before importing it into RC with the focal length in the metadata of the images. As far as I understand the exif groups its more about footage shot on different cameras and focal length

    0
    Comment actions Permalink
  • Avatar
    Götz Echtenacher

    Hi Paul,

    what do you mean by undistorting by distortion grid??? In an image processiong software?

    That can be a cause for quite a bit of problems, so maybe that is behind all of it. Usually, you are supposed to use out-of-the-camera distorted images and let RC do all that. If you change something, RC can get confused.

    Also, exif grouping is precisely for your case. It puts all images from ONE camera and ONE focal length into one group and undistorts them with the same model. That ensures a better first alignment, since there are no possibilities for stray errors due to bad coverage of tie points. You can ungroup them at the end and will get an individual model for each image. But that is not neccessary if you only used a fixed focal length and did not change focus or aperture. Then the undistort models will be identical anyway.

    0
    Comment actions Permalink
  • Avatar
    ShadowTail

    If your source images are extracted from a video source (4k resolution is somewhat an indicator for that) then you need to take a look at every single one of them as some are very likely going to be blurry. Those may have a bad effect on the final texture. You want to exclude the blurry ones from the texturing process.

    Also I have found that especially with videos as source material oftentimes using less extracted images yields a much better result (preview reconstruction around a million tris versus only a few 100k with lots of images). Results may vary though.

     

    0
    Comment actions Permalink
  • Avatar
    Paul Krist

    yes I am using drone footage. The drone was flying very slow and there is hardly any motion blur in these images since they were made specifically for photogrammetry. I shot distortion grids with the specific lens camera combination to calculate a precise undistortion for the images before importing them into RC. I will make a second test using the original images and exif groups. I tried using less images but it seems to work best for me having around 400-500 pictures per scan.

    0
    Comment actions Permalink
  • Avatar
    Götz Echtenacher

    A drone can be used for stills   a n d   video!  :-D

    What Shadowtail describes is due to the distance of the individual shots, the slower the flying speed, the closer the images. And although the shots should not be taken too far apart from one anoter, a certain distance is required for the 3D information to be extracted.

    However, that should not affect the number of detected features. In fact, a low number is a very good inicator for blurry shot, especially amongst many more images with a significantly higher number of features. But that is also not neccessarily the problem here.

    I'd be curious if you get any improvements and keep my fingers crossed!

    0
    Comment actions Permalink
  • Avatar
    Götz Echtenacher

    I just though of another possible issue with videos:

    The notorious Rolling-Shutter-Effect ! The faster, the worse, obviously.

    Again, this would not explain the low number of features...

    0
    Comment actions Permalink
  • Avatar
    Paul Krist

    right I will run the scan and let you know if it got any better.

    0
    Comment actions Permalink

Please sign in to leave a comment.