Photo adjustment for face modeling
Answered
Greetings!
We have a problem with kids face photogrammetry. In some cases 3d model assembles correctly (smoothly):
and sometimes it adds a lot of noise:
We can't retake this photos, so we are looking for other ways to reduce this noise. What parameters should we adjust to correct this issue? Is this even possible?
Thanks a lot!
We have a problem with kids face photogrammetry. In some cases 3d model assembles correctly (smoothly):
kid1.jpg
and sometimes it adds a lot of noise:
kid2.jpg
We can't retake this photos, so we are looking for other ways to reduce this noise. What parameters should we adjust to correct this issue? Is this even possible?
Thanks a lot!
-
kid1.jpg
-
kid2.jpg
-
Wishgranter wrote:Hi Arseniy
Are you using single cam scanning or multi camera head scanner ?
We using multi camera studio-
stud.jpg
-
-
Hi Arseniy
create a mock-up with newspaper on them, approximately of the size of the subject head. the newspaper will help align it very well.
then select ALL images and use METADATA (XMPs) and export camera positions, set the LOCKED option in the export settings of XMPs.
But keep in mind the images MUST have the same filenames !!!! or it will not work..
then just copy the XMPs next to the data you need to reconstruct and hit align. the XMPs will be used and no additional settings need to be used... -
Hello, Wishgranter!
Here the results of my test.. The mockup camera calibrations is really good. But the same XMPs is not working on child photos. We have locked camera positions on the studio, but the angles coult be changed on 5-10%. Is that cause failure like that or I`d make some mistakes?
Thanks a lot!-
695Без имени-1 копия.jpghttps://www.capturingreality.com/forum/download/file.php?id=6952097281imageFile jpgБез имени-1 копия.jpg
-
696Без имени-1 копия2.jpghttps://www.capturingreality.com/forum/download/file.php?id=6962097281imageFile jpgБез имени-1 копия2.jpg
-
-
I`ve send it via https://wetransfer.com/, you might have a notification
-
Hi Arseniy
So I have inspected the dataset and:
1. images are NOT in the same positions as the reference dataset for creating XMPs = moved, adjusted camera positions, angles ??
or image names are not correct ? ( image numbering issue ? )
that is why you get BAD reconstruction that is far away from ideal results, as your cameras have different angles from the reference data and it is not possible to use them....
2. photos of kids are in JPEGs with pretty bad quality ? you did not send me the RAW files.... -
Arsenyi, some more tips:
try setting in alignment settings image downscale factor to 1
I am in a similar situation, reconstructing some data from a not much rich camera array(only 11 cams).This improved the calibration for me. But I have similar problem, some models are very noisy, while others smooth. It's a lot about the models skin shininess I think....
Another important parameter is Adaptive blending start in advanced reconstruction settings. It can, If I understand it correctly, determine influence of depth maps, and in case of not having 360 degrees covered, it might improve the coverage. Sadly, by now the parameter can't go over 0.5. Default is 0.45. Wishgranter, would it be possible to allow higher values for this parameter? Not sure if I understood it right, but by my tests it behaves this way( direction of depthmap vs normal of suface -> depth map is used or not on a certain area).
I am also curious if you find another ways to improve the result.
Please sign in to leave a comment.
Comments
22 comments