Agentic framework that reconstructs sim-ready 3D scenes with rigid, articulated and deformable objects from multi-view RGB.
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.
CoDimRecon is a research framework from Tsinghua University (SIGS), Carnegie Mellon University and Genesis AI. It takes multi-view RGB images of a real environment and produces an editable, simulation-ready scene containing rigid, articulated and deformable objects. Deformables are handled by dimensionality: curves (e.g. a telephone cord) become rods with radii, surfaces (e.g. a plastic bag, paper) become thin shells, and volumes (e.g. a chair cushion, beanbag) become watertight solids for volumetric meshing.
The pipeline has three stages. An agent first builds an editable scene in Blender using geometric context (VGGT-Omega), object masks (CropFormer) and generated reference meshes (SAM3D, REST3D). It then refines appearance, articulation (joints such as revolute and prismatic) and rigid-body stability in MuJoCo. Finally, separate agent sessions per deformable type reconstruct each object, apply reusable simulator skills on an IPC-family contact backend, and run robot manipulation tests whose failures trigger targeted revisions.
The project page shows demonstrations such as a Franka robot lifting a phone handset, folding paper, pulling a drawer, and a G1 humanoid walking through the reconstructed office. The authors state that none of the evaluated baselines output deformable assets, so deformable results are capability demonstrations rather than benchmark comparisons. Code is reported as coming soon; no paper or code links were found in the provided material.
Watch the moment: the exact second top creators show it. Videos load only when you press play, via youtube-nocookie.com.
Turns photos of a real environment into an editable, physics-ready 3D scene with deformable object properties (cables, sheets, solids). Code is coming soon.
Closest neighbours by meaning, not just by category.
Research
Research method that reconstructs PBR-ready indoor 3D scene meshes from multi-view images using a Trellis.2 generative prior.
Research
Research method that turns a single photo or video into a simulation-ready 3D scene for robot reinforcement learning.
Model · Meta
Meta research framework that aligns a frozen SAM 3D model to input images or video at inference time for 3D reconstruction.
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
Feed-forward system that turns an RGB image or video into editable, textured per-object 3D meshes in under a minute.