Research model that decomposes a single image or scene mesh into object-level 3D meshes with consistent spatial layout.
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.
CoLa3D (Composable Latent 3D Decomposition) is a research method accepted at NeurIPS 2026. It decomposes a single whole-scene 3D latent into object-level meshes that keep a consistent spatial layout. The input can be a single image or an existing scene mesh.
According to a YouTube creator summary, the model processes the whole scene at once rather than object by object, and each object is output as its own mesh. It reportedly supports automatic, point-prompt and mask-prompt modes. The same source says code, training code and a dataset have been released, but no repository links were available in the provided material, so this is unverified.
The project page offers interactive results and comparisons with state-of-the-art methods, plus downstream applications. It is aimed at researchers and developers working on scene-level 3D generation and editing. The authors are Xudong Cai, Yongcai Wang, Deying Li and Chuanxia Zheng.
Watch the moment: the exact second top creators show it. Videos load only when you press play, via youtube-nocookie.com.
Turns a single image into a 3D scene with each object as its own mesh, processing the whole scene at once. Supports automatic, point and mask modes; code, training code and dataset released.
Closest neighbours by meaning, not just by category.
Model
Feed-forward 3D scene reconstruction model that outputs triangle primitives directly usable as meshes for simulation.
Model
Feed-forward system that turns an RGB image or video into editable, textured per-object 3D meshes in under a minute.
Research · Preferred Networks
Generative method from Preferred Networks that reconstructs high-resolution 3D scene meshes from posed multi-view images.
Model
Research framework that turns posed images with instance masks and 3D boxes into per-object meshes for cluttered scenes.
Model
Research model generating 3D indoor scenes as sparse voxels, refined from 25 cm to 1.5625 cm by an autoregressive cascade.
Research
Research method that reconstructs PBR-ready indoor 3D scene meshes from multi-view images using a Trellis.2 generative prior.