Texture not ideal
Recently, I noticed that some of my results were not es good as I expected. Because even though the geometry was very good and precise (high detail with some careful simplification), I still ended up with a texture that was considerably less crisp than the closest images provided. I am not talking about problems caused by alignment or too low texel size, just the resulting texture showing way less detail than should be possible. And also I am talking about the real texture as seen on an orthophoto and not the sometimes problematic reduced texture on the mesh as seen in 3D sweet view.
So I left most of the images with a greater distance to the mesh surface away and voila, the result was much crisper. That means that RCs algorythms are not ideal in that respect, as opposed to Zuzana'S statement in this thread: https://support.capturingreality.com/hc/en-us/community/posts/360003794351-what-photos-are-used-for-texturing-
Did anybody else have similar observations and/or does anybody have a solution how to deal with it other than switching off certain cameras? Ideally RC should weigh the texture of close cameras higher than further ones (or rather prefer the ones with a higher resolution on the surface).
-
I have not tested this out but Vlad talks about texture weighting in this tutorial (under the texture section). He just selects all the outer cameras and turns down their texture weight value. I suppose this is about similar to turning off cameras, like you're already doing. But yeah it would be helpful if RC could do this for us by default (with the option to override--maybe a texture weight bounding box with falloff outside the box or something like that), considering closer images are always likely to provide more detail
https://80.lv/articles/full-photogrammetry-guide-for-3d-artists/
-
Hi Goetz,
I recently had a similar Issue but I concluded this was just due to some occlusion. Maybe this is not what you meant but let me share it with you anyway. I need to notice that the images werent shot ideal so I am not suprised RC couldnt come up with a sharp result in my case.
First I played around with lowering the weighting of the far most distant cameras (pic #2). If look closely to the bottom of the image a bar starts to ghost into the image
My second attempt was to disable the far distant cameras and ended up with a much better result (pic #3). The bar in the lower frame of the image starts to become more visible.
If you compare the texture from my first attempt (pic #1) with the last attempt (pic #3) its a massive improvement but comes with the price of showing some ghosting. To be honest I am still not happy with the texture even though the geometry is nice and smooth.
Even though the bar makes just a tiny part of the big wall, RC still takes the pictures from far away for texturing for a big chunk of geometry, or at least it seems so.
I guess this wont help much because this all sounds like we need to generate different textured versions of the same model and combine them in a 3rd party package.
-
Hi Uwe,
thanks a lot for the replies!
What you both say confirms my approach.
Uwe, I think the 3rd sample has almost as crisp a texture than your sample image, not?
About the bar: you can only avoit this if the bar is modelled correctly. If it isn't present in the geometry of the model, then RC will just map it on the next surface behind it. Or phrased the other way around, if the bar was there in the model, the texture would be caught by it and not the wall. Does that make sense? :-)
Please sign in to leave a comment.
Comments
3 comments