I’m trying to develop a workflow that provides for seamless render meshes from RC dropping into a common scale and coordinate system in UE4. A Reconstruction Region of 10x10x10 size and centered at 0,0,0 is used to reconstruct a new component which is exported as Mesh 1. After exporting/importing this Reconstruction Region and moving it to 10,0,0 I then reconstruct Mesh 2. I see the adjoining meshes in UE4 don’t quite adjoin, there’s a slight gap, which points to the need for a bit of overlap, so moved to 9.95,0,0 I see the gap close, and it’s not terrible, but the vertical seam is still visible, and it seems that’s because of both geometry and textures. My subject matter is highly organic-looking cave passage, so straight seams tend to jump out. Thanks for a tip on what might alleviate the issue with seams.
Hi Benjamin
And why you are not reconstructing it as a single model and just then alternatively cut it into pieces ??
Happy New Year, Wishgranter. I thought that’s what I was doing, but maybe I’m unclear about Reconstruction Regions. I had reconstructed a larger set, so how then is this cut into pieces, if not by a series of smaller adjoining or slightly overlapping Reconstruction Regions? Are large models segmented into pieces outside of RC, and if so, what tools and workflow provide for that?
Hey Benjamin,
I’m working on the same problem but in Unity ( see: Efficient (automatic) tiling of large point clouds)
Built in option:
- In the reconstruction settings you can set a maximum vert-count for partcreation. Whenever the reconstruction gets to that limit (or sometimes earlier) it creates a new part. These parts connect seamlessly. In the export-options (when you click “Mesh”) you can decide to export every part as an individual model.
These models
=> share one texture
=> import into Unity with the exact, seamless coordinates (didn’t test that in Unreal, but should work as well)
If you want more control over the models:
(My landscape is made of very different LODs, so I need to specify different resolutions for some parts and might need to redo parts later on)
- basically create a model and then use the filtering tool to cut the corners off.
- Use the reconstruction region to create a model.
- Export the reconstruction region to some file
- Open that file with a texteditor
- Take off a few centimeters of the size on all sides and save
- reimport that new file
- use the filtering tool to delete all vertices outside the new reconstruction region.
- export the mesh
To add an adjoining mesh just modify the first reconstruction region file so that it’s smaller version aligns.
This is a pretty awful manual workflow for large landscapes but it’ll give you more control than the automatically created parts
Hello Linus,
Many thanks for sharing.
That tip on using parts based on max vert count seems ideal, no seams, but the limitation on a single texture obviously renders that a non-starter for larger sets.
As for your workaround, if I’m following you, you’re creating say a 10x10x10 reconstruction region as chunk 1 of a larger set, exporting the box for that region, editing its size in text editor, say 9.5x9.5x9.5, reimporting the smaller box, use filtering to delete everything outside of that smaller box, yes? If so, why not edit the box within RC’s info panel and then filter? Maybe, I’m missing something.
I’m trying to understand what this visible seam is, whether in the geometry, the texture, or both. In UE4 I moved the light around and it seems I’m seeing the seam at least in the geometry, so that would support my understanding of your approach to shave off these edge polys a little from two adjoining meshes, leaving a bit of overlapping polys behind on each face of the cube for neighboring meshes. Yes? If so, I’d be tempted to do this shaving in another app with less to do. I can see how creating a construction zone to split a larger model requires this linear slice to break lots of faces, the edge verts having to be reconnected to close up with new faces, which when applied to both sides of the split in adjoining meshes predicts a visible seam. I’ll try trimming these edge polys and leaving a rattier edge without these verts closing to new faces, leave enough overlap to hopefully hide the seam.
Do let me know if I’m misconstruing your steps. Thanks.
Hi Benjamin
That tip on using parts based on max vert count seems ideal, no seams, but the limitation on a single texture obviously renders that a non-starter for larger sets.
Single texture ?? you can have multiple textures per part, it’s much more about settings on UNWRAP ( can you describe the workflow used? )
As for your workaround, if I’m following you, you’re creating say a 10x10x10 reconstruction region as chunk 1 of a larger set, exporting the box for that region, editing its size in text editor, say 9.5x9.5x9.5, reimporting the smaller box, use filtering to delete everything outside of that smaller box, yes? If so, why not edit the box within RC’s info panel and then filter? Maybe, I’m missing something.
In RECONSTRUCTION->INFO PANEL you can set the proper dimension, position, orientation with numeric input…
I’m trying to understand what this visible seam is, whether in the geometry, the texture, or both. In UE4 I moved the light around and it seems I’m seeing the seam at least in the geometry, so that would support my understanding of your approach to shave off these edge polys a little from two adjoining meshes, leaving a bit of overlapping polys behind on each face of the cube for neighboring meshes. Yes? If so, I’d be tempted to do this shaving in another app with less to do. I can see how creating a construction zone to split a larger model requires this linear slice to break lots of faces, the edge verts having to be reconnected to close up with new faces, which when applied to both sides of the split in adjoining meshes predicts a visible seam. I’ll try trimming these edge polys and leaving a rattier edge without these verts closing to new faces, leave enough overlap to hopefully hide the seam.
Any examples - screenshots ?
Single texture ?? you can have multiple textures per part, its much more about settings on UNWRAP ( can you describe used workflow )
On a previous post related to my question about how best to segment large sets, you suggested you broke large reconstructions into a sequence of slightly overlapping Reconstruction Regions.
- I created a Reconstruction Region, exported .rcbox, used this first region to reconstruct first mesh, exported, imported to UE4
- selected large set component, imported .rcbox, moved it along an axis by length of that axis minus .05 for overlap to previous region, export, imported to UE4.
In UE4, the two regions drop in to scale and alignment, but seam is slightly visible:
Seam.JPG
That’s great to hear about using parts with max vert count to produce seamless meshes, with single texture per mesh. So, regarding “it’s more about settings on UNWRAP”, where in the HELP can I learn about that workflow?
With an eye on the big picture, my understanding is the RealityCapture Promo license is limited to 2500 photos per project, am still needing clarity on how far I can push capability to support large sets with the most efficient workflow. I’ve worked through various workflows aimed at various strategies at how to build up to this 2500 limit, whether throwing in all photos at once and getting to work growing the largest component using tie-ins to the smaller ones, or working in separate projects to export point clouds as .xyz, then importing related point clouds and adding tie-ins. Either way, I can see there’s a path to chunking 2500 52 Mp images worth of chunked meshes to produce a seamless virtual environment. Is the difference then between RC Promo and CLI how efficiently one moves beyond this 2500 photo limit on much larger projects? I’m clear about other distinctions, devoted technical support, etc. simply asking about this one aspect. With RC CLI, would one use command line instructions to relate projects each limited to 2500 photos? What does CLI workflow make possible?
Many thanks.
Hello Benjamin,
there is no restriction to the number of photos and/or laser scans for any projects created in the RealityCapture CLI.
Yep, that was my workflow.
As for the info panel - that was something I didn’t find. Even though I was looking for somewhere to adjust things like that. Thanks!
I might still use my workaround to avoid typing errors by automatically generating the textfiles in batch.
The normal cut-offs seem to affect the geometry quite lot with rounded edges and artifacts. Using the filtering tool seems to get rid of that by simply deleting the artifacts and closing the predefined (correct) vertices.
I’m not sure whether overlap between reconstruction regions is necessary. In my last attempt I had some overlap but I can’t recreate and/or modify that right now (next week).
I didn’t want to use another software for editing as that would add more steps to the process and I have a really large dataset. But I’d be interested to hear about your final workflow!
Ich bin kein Berliner, aber wohnte in München Jahre her. Grüss Gott, Linus.
I’d welcome comparing workflows.
The normal cut-offs seem to affect the geometry quite lot with rounded edges and artifacts. Using the filtering tool seems to get rid of that by simply deleting the artifacts and closing the predefined (correct) vertices.
I’m yet new to much in CG, want to be sure I’m following you. Are you saying that when you generate a Reconstruction Region, whether via text files or within RC, that each plane of the box defines what you call “normal cut-offs”? And when you say those affect the geometry, is that because each vertice inside the box has to close and RC is closing it in ways that result in this visible seam I’m seeing? Once reconstruction is done, how do you use the filtering tool to target these artifacts and incorrect vertices, leaving just the correct vertices, and how are those then closed? I read the Filtering Tool Help and played a bit with marking, but it’s not making sense. The Help shows red marks showing up on the scene, I don’t see those in mine after marking and I don’t get how a Reconstruction Region (box) is used to filter tris that a box created to begin with.
I’d like to better understand how it works exporting a mesh in Parts with max vert count. I see in Settings under Mesh Calculation that only Normal quality meshes feature a Force Single Part option. On export I again see the Force Single Part option. I’ve not been able to try this because my system isn’t enough to support running large components in High, but will both Normal and High reconstructions export as separate meshes according to max vert count set in part size?
Thanks!
Linus, Benjamin, was there a good outcome from this? You worked out a nice method? Quite keen to do this myself.