Just wondering if there is a way to filter out points in the sparse point cloud or throw out tie points if not used in x number of images?
I am not sure exactly what is going on under the hood, but based on what I am seeing, RC doesn’t go from sparse cloud to dense cloud to make the mesh, but uses something called depth maps – any explanation on how this works? I have noticed that these depth maps seem to always produce way better meshing results.
As of now, I only know how to filter the points out using the box, but I am wondering if there is something more? For example, in Agisoft, it is possible to select the individual points to delete; or delete points based on confidence; etc.
I pretty happy with the results, but just wanted to see if this is possible in RC. Or if it is possible to export the points and edit them in another software and import them in, etc.?
Hi TiberiusTyrone,
it is not possible to filter out sparse point cloud.
Sparse point cloud is used only for Preview quality meshing. For each image are computed depth maps (like distances from the camera) and these are combined to create the mesh.
Thanks, @OndrejTrhan, good to know. Does this mean that the Alignment step doesn’t mean much for the Normal and High Detail meshing? Normal and High Detail compute their own denser point clouds from scratch in addition to depth maps to produce the mesh? Or is there some dependency on the sparse point cloud?
So, it would okay to 4x downscale on the Alignment step to perhaps produce a faster point cloud; and then run at 2x downscale for Normal mesh? Or is there better uplift to do 2x on Alignment?
But actually… finding it a bit strange that when I play with 1x, 2x, or 4x downscale on Alignment (leaving all other settings on default), I always end up with the same point count… referring back to my previous post, v1.2 gave me different point counts for the different downscales. However, in v1.3, I get the same point count whether I choose 1x, 2x, 4x downscale; medium or high sensitivity; etc. And runtime is about the same for all of them.
Okay, it is actually a bit finicky – it seems if I run a new project at 4x downscale, then no matter what I do, it will always give the same output in the Alignment setting. And the same happens with 2x; and 1x. If I however delete all the components and rerun Alignment with different settings, then I get major differences:
Example when first starting at 4x downscale… no matter what I do afterwards in the settings, it as if it does not matter:
And then if I remove everything and run at 2x, then I finally get a difference where the point cloud moves from 50k points to 500k points. However, running then at 1x still keeps it at 500k points even though it says more cameras have been matched:
A bit annoying that I have to make separate projects for different alignment settings, but not the worst thing. Bug or is there something I am doing wrong?
The alignment and the sparse point cloud are necessary for the reconstruction. The denser sparse point cloud means that there are more common features between images, and therefore the higher detail information in the reconstructed model.
When running the alignment, it will also take into consideration the cache files and the existing components. You can save your project, and remove the cache files from the application settings.
