Reproduce “InstructMesh: Selective Refinement of Generative 3D Models for Fabrication” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Faraz Faruqi, Ahmed Katary, Demircan Tas, Theresa Hradilak, Ning Zhang, Jiaji Li, Fabian Manhardt, Martin Nisser and 5 more
InstructMesh enables non-experts to repair fabrication-relevant defects in generative 3D models through localized latent-space edits, combining region selection, natural-language instructions, structured sliders, and visual previews. On reconstructed Thingiverse models, its operations corrected up to 96.3% of missing-opening cases, while novice users identified 90.4% of flaws and successfully repaired 89.7% of annotated issues.
Key insights
- Fabrication-related flaws are common in generated reconstructions: 78.3% of 120 evaluated models contained multiple flaw categories, averaging 2.4 issues per model.
- Localized, constrained edits can make generative 3D correction more predictable than unconstrained re-prompting while avoiding direct exposure to mesh topology.
- Language and sliders provide complementary interaction styles: language reduces perceived effort, sliders offer slightly greater precision, and all participants preferred access to both through a hybrid workflow.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.