XMP reconstruction Problem

Comments

8 comments

  • Avatar
    Ondrej Trhan CR

    Hi Urdwexd, what are your steps here? What kind of dataset are you using? How were the XMPs created? Did you follow our tutorial on YouTube channel?

    0
    Comment actions Permalink
  • Avatar
    Urdwexd

    Hi, 
    I followed the instructions on the YouTube Channel. 

    I created the XMP's with the same Image Set in Reality Capture and exported it as locked. 
    So I first aligned all images(it's a Full body Scan system). Exported the XMP's. Restarted RC and cleared the Cache, imported the Images with the XMP's aligned it and then reconstructed in High Quality. And as you can see, the reconstruction looks terrible.
    Something I noticed was that after importing the images with XMP, the distortion model changes from Brown3 to K+Brown3. I used Brown3 for the initial alignment.
    I already tried using different XMP export setting, and changed the applications import settings. But so far nothing seemed to fix it. 

    0
    Comment actions Permalink
  • Avatar
    Ondrej Trhan CR

    Can you try this sample (Full body scan) if it will work for you correctly: https://www.capturingreality.com/SampleDatasets

    Did something change, when you change the settings back to Brown3?

    Are you able to see the georeferenced sign after import images with XMPs in 1Ds view? 

    0
    Comment actions Permalink
  • Avatar
    Urdwexd

    Looks like the sample is working.
    https://imgur.com/gZiGMMS

    I couldn't change it back to brown3, maybe because of the georeferenced properties ? 
    Yes, the georeferenced sign is visible after importing the images. 

    0
    Comment actions Permalink
  • Avatar
    Ondrej Trhan CR

    You should be able to do so. Are you trying to do so in Prior lens distortion?

    When you are able to see the georeference sign, what is the prior pose of the images? Is it possible, that the XMPs are wrong? Is it possible to share your data?

    0
    Comment actions Permalink
  • Avatar
    Urdwexd

    I am able to change the camera model, I did try again to align the images with the XMP's and it used Brown3. 
    But still the reconstruction looks bad. https://imgur.com/tpYFqxS 

    How can I check if the XMP's are wrong? I mean, it definitely could be possible.

    I can't share the data in public, but I could send it too you. 

    0
    Comment actions Permalink
  • Avatar
    Ondrej Trhan CR

    When you align these data without XMPs, then it is good. With XMPs not. But the results should be the same, if everything is OK.

    You can send it to trhond@capturingreality.com

    0
    Comment actions Permalink
  • Avatar
    Urdwexd

    Without XMP's it's really hard to align all images, because some cams have a steep angle. But the reconstruction looks good. But I need to manually align images with control points.  

    My plan was to use a good alignment image set, where I can easily align all images. To use for other image sets where it is difficult to align the cams.

    I will send you the image set and the XMP's.

     
     

    0
    Comment actions Permalink

Please sign in to leave a comment.