I am developing a web-based jewelry design workflow that generates .3dm files from four reference images:
Front view
Back view
Side view
Top view
My current pipeline:
Removes the background and extracts silhouettes from each image.
Registers the orthographic views.
Uses a voxel-based multi-view visual-hull reconstruction.
Converts the reconstructed volume into a mesh.
Writes the mesh and related metadata into a .3dm file using rhino3dm.
Although all four views are supplied, the resulting model is not sufficiently accurate. Common issues include:
Flattened or incorrect depth
Misalignment between views
Loss of small jewelry details
Incorrect openings, stone settings, and undercuts
Mesh geometry rather than clean, editable NURBS/Brep geometry
Front silhouette looking reasonable while the side and back geometry remain inaccurate
I am considering Rhino.Compute, potentially with RhinoCommon or a Grasshopper definition.
Could you please clarify the following?
Can Rhino.Compute itself reconstruct accurate 3D geometry from multiple 2D reference images, or does it only execute a reconstruction algorithm that I must provide?
Are there RhinoCommon, Grasshopper, or recommended plugin workflows for calibrated multi-view reconstruction?
Can Compute help convert a visual-hull or mesh result into clean, editable jewellery-quality Breps, SubDs, or NURBS surfaces?
Would the source images need camera calibration, matching scale, known projection parameters, landmarks, or depth information?
Is reconstruction from AI-generated orthographic images practical when those views may not be geometrically consistent?
Would a better architecture be:
External photogrammetry/reconstruction service
Custom Grasshopper reconstruction definition
Rhino.Compute for geometry refinement, booleans, meshing, and .3dm creation
Are there any Rhino.Compute samples or commercial plugins particularly suited to jewellery reconstruction from multiple images?
My application is built with TypeScript/Next.js and currently uses rhino3dm. I am trying to determine whether moving geometry processing to Rhino.Compute would materially improve reconstruction quality or simply provide access to more geometry operations after reconstruction.
I can provide sample front/back/side/top images and the resulting .3dm file if that would help diagnose the limitation.
Rhino.Compute allows you run Rhino on servers giving you the ability to make REST calls that create or modify geometry, solve Grasshopper definitions or Python scripts. Rhino.Compute is merely the layer that manages the requests made to Rhino. In this mode, Rhino is a geometry calculator. It does nothing by itself. You must provide the algorithm, via a plugin, script, or Grasshopper definition.
Rhino itself is great for such modelling. Automating this via Rhino.Compute is a different task. If you have this automated via RhinoCommon or Grasshopper (on desktop Rhino), then it is very likely you could set that up via Rhino.Compute.
You can use tools like QuadRemesh and Shrinkwrap to help cleanup geometry. Again, this is something that I would work through and automate, probably via Grasshopper, before moving onto Rhino.Compute.
You probably know more about this than we do. From the work that I have done with photogrammetry, the answer would be, yes.
Your milage may vary, I suppose. I am not sure what the current state of AI assisted reconstruction. I’ve only used ML and CV tools, and these were all external to Rhino.
This sounds like an interesting pipeline. I would find a reconstruction service with an API that serves you the resulting mesh that then you could pipe to your GH definition that would refine the mesh and give you the 3dm. I would suggest you limit the kinds of things you want to reconstruct at the beginning in order to exercise the pipeline and work our any kinks. It seems that this is the kind of workflow that will have a lot of edge cases to work out.
You might want to check Food4Rhino for 3rd party plugins for reconstruction and photogrammetry.
Rhino.Compute will not improve quality by itself. You will have access to all of Rhino’s geometry operations that if used correctly, will give you more powerful tools for reconstruction.
I have use case where I get images from which I need automated way to generated 3dm files. To get reliable 3d view, I got suggestion to provide 360 degree view rhino3dm.
It did generate 3d view to some extent. However, there were accuracy issues, broken parts, missing grooves, etc.
Those photos would be too low resolution even if there were a magic plugin. I suggest you work from the original jewellery and explore the possibility of macro photogrammetry (although I don’t know whether such a thing even exists).
