CoLa3D is a lightweight, query‑based transformer that performs composable 3D scene decomposition directly from a single whole‑scene latent, whether derived from an image or an existing mesh. It outputs a set of object meshes with a consistent spatial layout that can be reassembled into a coherent scene, avoiding fragile pose and scale estimation used by other methods. The model achieves state‑of‑the‑art performance on synthetic and real‑world datasets, improving global scene consistency and object quality while running twice as fast and using less than 1% of SAM 3D’s training data.
arXiv:2609.23796v2 Announce Type: replace Abstract: Single-image 3D object generation can now produce high-fidelity assets, yet accurately placing them into a coherent scene layout remains an open ch...
arXiv:2609.23796v1 Announce Type: new Abstract: Single-image 3D object generation can now produce high-fidelity assets, yet accurately placing them into a coherent scene layout remains an open challe...
WorldSculpt presents a method for generating compositional 3D representations of cluttered scenes with hundreds of objects by adapting a single-object 3D generative prior to multi-view observations. The approach, built on Pixal3D with a multi-view conditioning pathway, can generalize to highly occluded scenes without scene-level training. The authors also introduce the UE-MeshyScene benchmark and demonstrate that their method outperforms prior approaches across various evaluation settings, including converting existing 3DGS worlds into compositional mesh scenes.
LucidLyra is a framework that reconstructs 3D scenes from video latents in a feed‑forward manner while simultaneously providing open‑vocabulary semantics and instance cues. By distilling knowledge from a camera‑controlled video diffusion model and frozen 2D vision foundation models, it jointly infers 3D Gaussians that encode geometry, appearance, and semantic information without requiring per‑scene optimization or manual 3D annotations. The resulting scenes support novel‑view rendering, open‑vocabulary segmentation, and object grouping, and can be extended to monocular video for dynamic 3D reconstruction.
VolFill is a generative framework that reconstructs the complete 3D geometry of a scene from a single RGB image, addressing the challenge of inferring hidden structures. It uses a hybrid 3D VAE to encode sparse truncated unsigned distance function grids into a latent space, and a latent Diffusion Transformer to denoise and recover the full scene. By conditioning on a geometry foundation model, VolFill leverages spatial priors and provides a structured representation that enables direct surface extraction and scalable occupancy queries, outperforming existing amodal baselines on SCRREAM and NRGB-D datasets.
ControlFlow3D is a framework that upsamples point clouds by distilling knowledge from pretrained foundation models into a lightweight generation network. It performs conditional flow matching in a latent space defined by a frozen 3D encoder, using unsupervised soft KMeans to cluster multi‑view image tokens and 3D latent tokens into shared prototypes that guide the flow trajectory. A Mamba‑based geometry decoder then maps latent features to dense coordinates with linear complexity, and the authors introduce ShapeNetPU, a 35K‑object multi‑modal benchmark 31 times larger than previous datasets.
The paper explores how 3D Foundation Models (3DFMs) like VGGT can be leveraged for zero‑shot depth synthesis. By decoding hidden surfaces from the models’ internal representations, the authors introduce Z3D, a method that uses latent diffusion on 3DFM representations to estimate pointmaps in unseen views. Experiments demonstrate that Z3D can generate realistic depth maps across multiple datasets.
MessyKitchens introduces a new dataset of cluttered real-world kitchen scenes with detailed 3D object shapes, poses, and accurate contact information. The authors extend the SAM 3D single-object reconstruction method with a Multi-Object Decoder (MOD) to jointly reconstruct entire scenes, achieving better registration accuracy and reduced inter-object penetration compared to prior work. The dataset, benchmark, code, and pretrained models will be publicly released on the project website.
3D-SceneQ is a new 3D large language model that tackles two key limitations of existing 3D scene understanding pipelines: the inclusion of irrelevant object tokens that add noise, and the exclusive use of point cloud data that lacks rich semantic cues. It introduces a Query‑Guided Adaptive Pruning (QGAP) module that selectively keeps task‑relevant objects based on user instructions, and a Multi‑modal Object‑level Feature Enhancement (MOFE) module that fuses 2D feature embeddings into the 3D representation to enrich semantic information. Experiments show that 3D‑SceneQ surpasses state‑of‑the‑art methods on several benchmarks, achieving significant gains in metrics such as CIDEr and EM@1.
arXiv:2605.13018v2 Announce Type: replace Abstract: Object-centric scene understanding is a fundamental challenge in computer vision. Existing approaches often rely on multi-stage pipelines that firs...
UniQueR is a unified query‑based feedforward framework that reconstructs 3D scenes from unposed images by treating reconstruction as a sparse 3D query inference problem. It learns a compact set of 3D anchor points that serve as explicit geometric queries, allowing the network to infer scene structure—including occluded geometry—in a single forward pass. By encoding spatial and appearance priors directly in global 3D space and using a decoupled cross‑attention design, UniQueR achieves strong geometric expressiveness while reducing memory and computational cost, outperforming state‑of‑the‑art feedforward methods on Mip‑NeRF 360 and VR‑NeRF with far fewer primitives.
Part-level 3D generation has recently attracted increasing attention for producing structured and editable 3D assets. However, existing methods typically decompose objects according to functional semantics rather than the editable material boundaries (e.