Research model that splits a 3D mesh or single image into non-overlapping, gap-free parts from one 3D point prompt per part.
Updated EU status checked Last shown
Sources: the vendor's own pages (linked above) and creator videos at the linked timestamps. EU status is information, not legal advice; see how we check.
Point2Part is a research model for 3D part decomposition. It treats decomposition as a joint partitioning of the whole shape, so the resulting parts are exclusive (no overlap) and exhaustive (no gaps) by construction. Users choose the parts they want by giving one 3D point prompt per part.
A single promptable model covers three tasks: image-to-part generation, mesh-to-part generation, and part segmentation. Because the prompts set the granularity, the same mesh can be decomposed into coarser or finer parts. The project page reports better part quality than OmniPart, PartPacker, PartCrafter, X-Part, CubePart and HoloPart + P3-SAM on part Chamfer distance and F1, lower part penetration, and higher watertightness. It also reports higher mIoU than PartField, P3-SAM, S2AM3D, PartSAM and Point-SAM on segmentation. These figures are the authors' own.
The work is described in an arXiv preprint (arXiv:2609.38180) by Tsui, Tuan, Ma, Ugrinovic, Shiratori and Kitani. The provided material gives no vendor, code, weights or license. A video creator said models and training code are to be released, but no release links were found.
Watch the moment: the exact second top creators show it. Videos load only when you press play, via youtube-nocookie.com.
Takes an image or 3D mesh plus point prompts and decomposes the object into non-overlapping parts without gaps. It beats similar segmentation tools on benchmarks. Models and training code are to be released.
Closest neighbours by meaning, not just by category.
Model
Roblox research model that generates text-to-3D objects as separate part meshes from a user-defined part list.
Model
Research model that decomposes a single image or scene mesh into object-level 3D meshes with consistent spatial layout.
Model
Feed-forward 3D Gaussian Splatting model that reconstructs scenes from arbitrary input views, with or without camera poses.
Model
Research framework that turns posed images with instance masks and 3D boxes into per-object meshes for cluttered scenes.
Model
Feed-forward 3D scene reconstruction model that outputs triangle primitives directly usable as meshes for simulation.
Model
Research model that predicts layered, pixel-aligned 3D point clouds, including hidden geometry, from one image or video.