Research method that reconstructs PBR-ready indoor 3D scene meshes from multi-view images using a Trellis.2 generative prior.
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.
GenRecon is a research method for multi-view 3D scene reconstruction from RGB images or casually captured smartphone videos. It couples reconstruction with a 3D generative prior, using Trellis.2 as the example shape model. The scene is treated as conditional 3D generation over spatially localized, overlapping chunks that together tile the scene, which lets the object-level prior scale to larger indoor environments.
A projection-based conditioning mechanism lifts features from posed input views into a shared 3D voxel grid. The grid is independent of view order and anchored to scene coordinates. At inference, camera poses are recovered, the scene is split into chunks, and all chunks are generated jointly in a shared latent space to keep boundaries consistent. The fused latent is decoded into a complete mesh with PBR materials, which can be relit and edited in standard rendering pipelines.
The authors report results that outperform existing reconstruction methods by 16%. Their comparisons use 8 input images, and they note that baselines tend to produce noisy or oversmooth surfaces and leave occluded areas incomplete. At the time of the source material, only the project page and an arXiv paper (2605.23888) are available. No code or weights are linked, and a YouTube review described the code as "coming soon".
Watch the moment: the exact second top creators show it. Videos load only when you press play, via youtube-nocookie.com.
Turns smartphone videos or image sets of a room into a PBR-ready 3D mesh scene using chunked generation and Trellis 2 as a shape prior. Supports relighting and editing; only a paper is out, code marked coming soon.
Closest neighbours by meaning, not just by category.
Research
Agentic framework that reconstructs sim-ready 3D scenes with rigid, articulated and deformable objects from multi-view RGB.
Research · Preferred Networks
Generative method from Preferred Networks that reconstructs high-resolution 3D scene meshes from posed multi-view images.
Research
Research framework that generates collision-free, physically stable 3D indoor rooms from diffusion-sampled layouts.
Model
Research model generating 3D indoor scenes as sparse voxels, refined from 25 cm to 1.5625 cm by an autoregressive cascade.
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.
