I’m trying to work out the best approach for reprojection of textures in RC with geometry that’s been optimized in external softwares. I’ve tried UDIMs with one combined OBJ made up of multiple geos which as I’ve read is not yet supported and overlapping UVs also aren’t generally a good bet either as they often don’t import. 16k textures do work with basic settings but when you get into more advanced image groupings or optimizations of the settings to get sharper or smoother or more consistent results performance goes way down and definitely makes rendering in other softwares afterwards even less performant.
I’ve noticed that all the textures exported have a u1_v1 trailing descriptor in the name and wondering if there is a method to use a UV space greater than 1 in UDIM fashion? OR is there a way to properly generate geometry with overlapping UVs that RC will accept? OR will we eventually be able to select multiple models within RC and project to them as a group?
The big issue I’m having is when I split up my geo into smaller component parts I end up with alot of extraneous texture projection occlusions and “splotches” from hidden geo. Eg. I’ve done eyes separate from a human head and not only do the back of the eyes (inside of the skull) pick up the hair from the back of the head, I even have the clothing texture from the lower legs on the bottom of the eyes. (200 camera scan rig)
My best solution at present is to cut the geo up and use the projection weighting to narrow down/refine the result but this is a time consuming process and often requires doing multiple projections and exports to have a set of images that can be further cleaned up in 2D paint packages.
For now, post cleanup in Mari works but even that is less than ideal given there’s no FBX export and the Maya export suffers from Maya rounding of small numbers and completely incorrect orientation of cameras. I’ve resorted to exporting dense point clouds and using PFTrack manually creating cameras to use in Mari. This process could be automatic with a better Alembic, FBX, or Maya export.
Hi Robert Moggach
Overlapping UVs will be always an issue in photogrammetry… As they need proper projection parameters so the UVs need to be done in a proper way, too…
I’ve noticed that all the textures exported have a u1_v1 trailing descriptor in the name and wondering if there is a method to use a UV space greater than 1 in UDIM fashion? OR is there a way to properly generate geometry with overlapping UVs that RC will accept? OR will we eventually be able to select multiple models within RC and project to them as a group?
take a look at a very good article about UDIM https://www.fxguide.com/featured/udim-uv-mapping/
The big issue I’m having is when I split up my geo into smaller component parts I end up with alot of extraneous texture projection occlusions and “splotches” from hidden geo. Eg. I’ve done eyes separate from a human head and not only do the back of the eyes (inside of the skull) pick up the hair from the back of the head, I even have the clothing texture from the lower legs on the bottom of the eyes. (200 camera scan rig)
on a 200 camera rig this should not happen, is it possible to get one dataset so that we can take a look at what is the issue and how to solve it ??
My best solution at present is to cut the geo up and use the projection weighting to narrow down/refine the result but this is a time consuming process and often requires doing multiple projections and exports to have a set of images that can be further cleaned up in 2D paint packages.
this should not be necessary, it looks like a not ideal workflow…
Please contact me directly so that we can solve this…
Hey Milos -
I understand UDIMs but great article nonetheless on the history. I was wondering if you actually had an implementation of UDIMs working that isn’t immediately apparent as to how to use given the UDIM naming suffix. I’ve tried using models with correct layouts without success and noticed in the forums that it’s a feature that is coming soon.
As for dataset, the EXR dataset I sent you last week should work just fine although that one’s not as complicated as what I’m working on now. I’m sure some of the issue is focus and the zoom lenses in Tony’s rig hence the high levels of noise. As for workflow, if you want to relight photogrammetry and especially human scans, a complicated workflow is the solution. Honestly, a working maya or fbx export with floating point image support and ability to swap in different image filetypes would be ideal so we can do the heavy lifting of texture projection in Mari by hand.
Let’s talk on email when you want to coordinate a further discussion.
Hi Robert Moggach
UDIM is “on the way” but, as every solution, it needs its time to mature.
Yup, drop me an email for further discussion…
Look forward to it… while abstract the attached image is not really ideal… nor is cutting up the model into separate pieces and multiple projections to achieve better texture coverage and usability. (currently the only way to do this well)
Screenshot 2016-07-14 14.28.09.jpg