3D Scene Understanding

3D grounding aims to localize target entities in complex scenes from natural language and plays a fundamental role in embodied perception and spatial reasoning. However, existing approaches mostly rely on feature similarity or direct matching, making it difficult to connect natural-language intent with the implicit semantic and geometric structures hidden in billion-scale urban point clouds. We reformulate city-scale 3D grounding as structured constraint reasoning, where description semantics are organized into computable cross-modal constraints over open-vocabulary 3D entities, attributes, and spatial relations. We present CitySTAR, a training-free framework for reasoning-driven urban 3D grounding. CitySTAR lifts raw billion-scale urban point clouds into a query-ready scene graph of open-vocabulary 3D instances, with CodeLLM-driven tools supplying multimodal evidence for node attributes and 3D spatial relations. It then models target-context topology with paired hypergraphs and performs bidirectional topology verification for structural disambiguation. Finally, a Reflective Cross-modal Grounding module integrates topology consistency and candidate-centered 2D visual evidence to make decisions over a metric-aware 3D context graph. To further support this setting, we introduce CitySTAR-3D, an enhanced benchmark that improves semantic coverage, instance completeness, bounding-box fidelity, and spatial-relation complexity in city-scale 3D grounding. Extensive experiments show that CitySTAR consistently improves open-world urban 3D grounding while maintaining strong interpretability and generalization.

We present ParticleSplat, a self-supervised object-centric learning method that decomposes scenes into a set of latent ''particles'' representing semantic entities through feedforward 3D Gaussian Splatting. Building on the Deep Latent Particles (DLP) framework, which represents images as a set of particles with attributes such as position, scale, and visual appearance, we address a key limitation of DLP: its inherently 2D nature, which prevents explicit 3D spatial and geometric reasoning that are critical for downstream tasks such as robotic manipulation. Leveraging the structural similarity between latent particles and 3D Gaussian primitives, we introduce a 3D latent particle space trained with a novel view synthesis objective. Our model jointly encodes multiple views with camera poses into a shared 3D object-centric latent space, then transforms particles into particle-aligned 3D Gaussians whose composition reconstructs the full scene. On simulated and real-world datasets, we show that this formulation inherently learns object masks without supervision and supports controllable 3D scene editing, such as moving objects by modifying particles in the latent space. We further establish that the learned 3D representation improves downstream performance on robotic manipulation tasks.

3D Gaussian language fields provide an explicit, spatially grounded representation for 3D visual question answering (VQA), but their dense semantic features can require tens of thousands of embeddings per scene, resulting in substantial storage, memory, and inference costs. We investigate how much of this representation is actually necessary for downstream reasoning. Starting from a full embedding representation, we systematically sparsify its semantic embeddings, including the previously underexplored regime below a single image-equivalent block down to 8 visual tokens. We compare random, geometric, semantic, and joint spatial-semantic selection strategies and introduce an object-based sparsification method that distributes the token budget across detected object instances while retaining background context. Experiments on ScanQA and MV-ScanQA reveal substantial redundancy in dense Gaussian language fields. Strong VQA performance is retained with only a few hundred semantic embeddings, corresponding to less than 1% of the original representation. Object-based selection performs well relative to others, with only modest observed changes down to 256 tokens. At this budget, SparseTalk retains 0.80% of SplatTalk's 32,076-token inference input and 0.332% of the mean 77,207-Gaussian dense field, increasing inference throughput while reducing decoded-feature memory 125-fold.

Scene-level point cloud understanding remains challenging due to diverse geometries and spatial layouts. While pre-trained 3D point cloud foundation models (PFMs) offer strong transferability, full fine-tuning (FFT) incurs substantial computational and storage costs. Parameter-efficient fine-tuning (PEFT) provides a promising alternative, but existing PEFT methods largely focus on object-level point clouds and overlook serialization-induced partition variations in large-scale scenes. To address this issue, we propose PointPiT, a partition-invariant tuning framework for scene-level point clouds. Specifically, a Scene-aware Structural Adapter (SSA) integrates local geometric patterns with global scene context to mitigate partition-induced representation shifts. Moreover, Gradient Subspace Optimization (GSO) selects informative and partition-stable update directions, suppressing partition-dependent variations during optimization. Extensive experiments across multiple scene-level benchmarks demonstrate that PointPiT achieves competitive or even superior performance to full fine-tuning with less than 1% of backbone's parameters, while achieving consistent state-of-the-art performance among representative PEFT methods.

SceneHI is a framework that lifts high-resolution, illumination-aware priors from 2D diffusion models to perform 3D texture synthesis. It is the first to demonstrate that high-resolution textures, previously limited to 2D synthesis, can be generated directly on 3D objects without model fine-tuning or optimization. Designed for complex, multi-object environments, SceneHI uniquely combines 3D-consistency, high-resolution fidelity, and physically plausible baked shadows within a single generative pipeline. To enforce strict geometric coherence, we introduce an exact analytical pixel-to-texel mapping that aligns diffusion trajectories across multiple viewpoints. We utilize High-Resolution Latent Textures (HRLTs) as a persistent canvas for gradually denoised textures, while camera views perform the denoising steps in latent pixel space. This ensures a shared base texture that can be subsequently refined to high resolution without compromising multi-view consistency. Finally, a light-aware generative pass embeds realistic geometry-consistent shadows directly into the atlases, bridging the gap to production workflows. SceneHI achieves high visual fidelity while reducing generation time by 80% compared to existing scene-level methods.

Open vocabulary 3D semantic segmentation methods typically lift CLIP features into 3D. This embeds points in a joint vision-language space known to behave like a bag-of-words on compositional tasks. Furthermore, even annotation free variants often require a large 3D training corpus and a dedicated 3D encoder per domain. Instead we use a vision-language model purely as a translator. It produces structured, entity-level descriptions of each posed image. These descriptions are grounded, projected, and aggregated directly in a general-purpose, language-only embedding space, with no 3D training corpus or encoder required. On ScanNet++, our pipeline is competitive with strong annotation free baselines trained on ScanNet. On a 5-building cultural heritage benchmark, raw scores initially favor a CLIP-based variant, but a single systematic vocabulary correction reverses this ranking. An effect confirmed by a second, independent correction on a different class, indicating that language-space embeddings track physical content more faithfully. This fidelity extends to genuinely out-of-vocabulary (OOV) objects on ScanNet++ proving that language-space embeddings separate presence from absence objects far more sharply than CLIP-based embeddings do. GoDeep also localize these OOV objects within the scene, all without any 2D-3D annotation. Because every representation remains discrete text, predictions are also explainable at the point level. Finally, exploiting both a heuristic weighting, that favors precise over merely frequent observations and GoDeep's explainability property, we propose an aggregation strategy, as a proof of concept, that favors finer elements localization.

Spherical observations provide global visual context for 3D scene understanding. However, visual information is encoded in an angular domain, whereas the physical world is represented in Cartesian coordinates. This cross-space representation gap complicates geometric correspondence and semantic evidence aggregation. To delve into this challenge, we introduce Spheriverse, comprising 64,400 temporally aligned spherical image-LiDAR pairs organized into 644 sequences. The dataset spans diverse scenes, illumination, and weather conditions, with fine-grained semantic classes. We further establish benchmarks for semantic occupancy prediction, semantic mapping, and 3D object detection, evaluating 30+ methods through overall and scene-wise comparisons. For dense prediction, we propose SphereOcc, an occupancy framework that couples spherical geometry modeling with semantic evidence retrieval. Cartesian-Spherical Representation Remodeling (CSRR) incorporates spherical range-azimuth geometry into Cartesian voxel features through region-wise modulation. Spherical Evidence Re-querying (SER) then conditions queries on voxel content and range-height-azimuth geometry to adaptively retrieve relevant semantic evidence from source spherical image features. SphereOcc achieves 13.91% mIoU and 24.65% GeoIoU, yielding relative improvements of 13.9% and 9.3% over the respective best-performing methods, TPVFormer and SurroundOcc. It also ranks first in both metrics across all five scene categories, with consistent advantages across the evaluated spatial partitions and reduced fields of view. The established benchmark and source code will be available at https://feit-feiteng.github.io/Spheriverse.

We present Voxel-Grounded Online Instance Manager (VOIM), a training-free voxel-grounded instance manager that builds open-vocabulary 3D instance maps from RGB-D or from monocular RGB alone, a regime no prior training-free system addresses. Online systems typically segment object instances and label them at first detection, committing when evidence is weakest. VOIM instead defers label and instance decisions until soft evidence from unmodified, off-the-shelf perception has accumulated per voxel across views. We show that the mapping stage, rather than the particular perception models, carries the result: across four perception configurations on ScanNet++, varying the region descriptor, the detector label prior and the mask source, the map exceeds the strongest online RGB-D system, OVO-SLAM, by between 4.8 and 11.7 mIoU. Perception is not neutral, and substituting that baseline's own descriptor family costs 4.1 of the margin, yet the baseline carries the marginally better 2D descriptor (33.7 vs. 31.5 mIoU over three scenes) and still realizes the weaker map. Under a like-for-like protocol VOIM reaches 44.07 mIoU on ScanNet++ against 32.37, winning all ten scenes and both aggregations (pooled 33.31 vs. 25.97), and the same system runs unchanged to fully monocular RGB, matching that baseline pooled on Replica (27.80 vs. 27.50). The advantage is regime-specific: under Replica's all-classes scoring, matched inputs give a split result, 28.60 vs. 27.50 pooled against 24.59 vs. 30.11 on the per-scene mean. Room scale is label-limited and building scale drift-limited. Labeling does not run in real time, dominated by per-class detection over the full vocabulary. The maps export occupancy grids and resolve free-form queries to object instances.

Constructing a unified 3D scene understanding model has long been hindered by the topological discrepancies across sensor modalities. While applying the Mixture-of-Experts (MoE) architecture is a flexible approach for multi-domain 3D understanding, we observe that conventional feature-only MoE routers may underrepresent local sampling topology under semantic supervision, making expert allocation difficult when semantic consistency coexists with geometric heterogeneity. To overcome this challenge, we propose STAR (Spatial-Topology Aware Routing Framework). Specifically, we introduce a multi-attribute self-supervised pre-training branch, covering topological and textural variations, to anchor cross-domain structural priors. Building upon this, we design a domain-aware expert branch with two mechanisms: Domain-Spatial-Guided Routing (DSR), which captures local topological variations from spatial context, and Entropy-controlled Dynamic Allocation (EDA), which adjusts the number of activated experts according to routing uncertainty. Together, these branches combine stable cross-domain representation learning with adaptive expert allocation. Extensive experiments across various tasks, encompassing both indoor and outdoor scenes, demonstrate the effectiveness of STAR. It achieves 80.1% mIoU on the ScanNet validation set and 77.2% mIoU on S3DIS, consistently improving over strong baselines. Code is available at our project page (https://xmw666.github.io/STAR/).

Free-viewpoint 3D scene media is increasingly important for immersive applications, yet practical capture often suffers from severe view sparsity and motion blur. Although neural rendering has advanced sparse-view synthesis, existing blur-aware methods typically require substantial multi-view redundancy, accurate camera poses, or costly per-scene optimization. We address a stringent yet practical setting: reconstructing a coherent 3D scene from only two motion-blurred images with known intrinsics, without input-view poses, auxiliary sharp images, or per-scene test-time optimization. To this end, we propose CasDeblurGS, a cascaded framework that progressively recovers reliable cross-view information from local 2D correspondences to global 3D guidance. Stage 1 constructs locally reliable guidance through occlusion-aware correspondence filtering, while Stage 2 aggregates the intermediate restorations into a provisional pose-free 3D Gaussian representation whose input-view re-renders provide dense global guidance for final restoration. The resulting views enable a more coherent 3D representation and higher-quality novel-view synthesis. Experiments on real-world and synthetic Deblur-NeRF scenes show consistent gains over strong baselines, improving PSNR by 1.19 dB and 2.11 dB, respectively. Progressive ablations, cross-view correspondence visualization, and camera reprojection analysis further demonstrate improvements in both rendering quality and multi-view geometric consistency.

We introduce LEGO for advanced open-vocabulary scene understanding. Beyond basic concept recognition, its core innovation lies in capturing the intrinsic semantic hierarchies within the scene, such as the "flowerpot -> bouquet -> bud -> petal" lineage. While foundation models like SAM can identify multi-granular structures in 2D, their partitions are strictly perspective-bound and lack cross-view consensus. LEGO self-adaptively re-grades volatile multi-view SAM granularities into a unified, 3D-consistent hierarchy. This provides precise supervision for the structurally coherent, multi-level segmentation of 3D scenes. By grounding these segments with CLIP embeddings, LEGO recovers open-vocabulary semantic logic across hierarchical levels. Furthermore, by incorporating spatial relationships, we elevate these segments into level-wise language scene graphs, effectively empowering Large Language Models to perform complex, context-aware spatial reasoning and precise visual grounding. Experimental results demonstrate that LEGO establishes new state-of-the-art performance across both promptable and open-vocabulary 3D segmentation benchmarks, exhibiting advanced hierarchical scene decomposition and context-aware spatial reasoning.

3D occupancy prediction is fundamental to scene understanding, yet existing 3D semantic occupancy methods are typically specialized to fixed scene types and occupancy protocols. We introduce Cross-Scene 3D Semantic Occupancy Prediction, a new task setting which requires a single model to handle heterogeneous indoor and outdoor scenes with varying cameras, spatial ranges, voxel specifications, and semantic taxonomies. This setting poses a fundamental challenge: achieving metric-consistent yet scene-adaptive image-to-3D lifting across varying camera configurations and scene scales. To address this challenge, we propose OccAnyScene, a pixel-frustum-centered Gaussian framework built upon a pretrained depth foundation model. Specifically, the framework employs Pixel-Aligned Frustum Feature Aggregation to construct a camera-aware frustum query for each feature pixel, and Frustum-Parameterized Gaussian Construction to decode each query into multiple Gaussians whose positions and sizes are constrained by the predicted pixel depth and corresponding frustum geometry. OccAnyScene sets new state-of-the-art results, achieving 59.92% mIoU on the indoor Occ-ScanNet and 23.06% mIoU on the outdoor SurroundOcc-nuScenes.

3D scene understanding is increasingly important in construction, yet most methods are developed on curated datasets that do not fully reflect real site sensing conditions. In many workflows, individual LiDAR scans provide rapid local updates rather than complete scene representations, producing limited surface coverage, acquisition-driven density variation, and severe imbalance between dominant planar surfaces and sparse construction elements. Because large point clouds must be downsampled, sampling resolution and point allocation directly affect the balance between geometric detail and spatial context. This study evaluates these effects under a fixed per-fragment point budget and introduces an incidence-aware sampling strategy for individual LiDAR scans. The method maps points to a geometry-normalized manifold space for voxel-based selection while preserving original Euclidean coordinates for downstream learning. It requires only point coordinates and normals and no backbone modification. Using the Site in Pieces (SIP) benchmark, experiments with Point Transformer and PointNeXt show improved resolution-averaged segmentation performance, especially for non-planar elements and ladders, while reducing sensitivity to sampling resolution. The results show that acquisition-aware sampling can provide a more stable geometric representation and should be treated as an active component of individual-scan 3D segmentation rather than generic preprocessing.

Feed-forward 3D Gaussian Splatting (3DGS) enables efficient and generalizable 3D reconstruction, but current feed-forward 3DGS methods for scene understanding remain largely category-oriented. In contrast, instance-aware 3DGS methods typically rely on per-scene optimization and often decouple reconstruction from instance and semantic learning, limiting reciprocal interactions among them. We present InstanceSplat, a unified feed-forward 3DGS framework for generalizable 3D reconstruction and instance-aware scene understanding from pose-free multi-view images. In a single forward pass, InstanceSplat constructs an instance-aware Gaussian representation that jointly encodes appearance, geometry, instance identity, and language-aligned semantics. Shared 3D Gaussians ground instance identities across views, producing renderable and cross-view-consistent instance features. To allow reconstruction and scene understanding to benefit from each other, we further design an instance-centric learning strategy that connects reconstruction, instance learning, and semantic learning through shared instance structure. Specifically, instance cues guide reconstruction, language-aligned semantics strengthen the discrimination of confusing same-category instances, and instance regions aggregate semantic evidence into coherent object-level predictions. Experiments on novel-view synthesis, instance segmentation, and open-vocabulary semantic understanding under varying input-view settings and on an unseen dataset demonstrate state-of-the-art performance, practical efficiency, and strong generalization.

Joint-embedding predictive architectures learn by predicting latent representations of missing observations, yet many masked JEPAs are evaluated primarily through the encoders they produce. We ask what a trained predictive pathway itself infers when an entire entity is absent from a native 3D scene. We introduce SR-JEPA, a point-native JEPA for scene-scale point clouds whose original frozen predictive pathway can be queried at a supplied location. At evaluation, every point of one object is removed before encoding and replaced by the same shape-free 32-point query at its centroid. Training uses only self-contained 3D EMA targets: no reconstruction, semantic labels, language, or lifted 2D features. On 5,953 held-out ARKitScenes objects, the imputed latent reaches 43.13% semantic-identity macro accuracy, 22.18 points above the strongest floor. Randomizing the prediction path removes 9.78 points, while substituting matched donor context removes 21.98 points. On 8,570 Sr3D support pairs, the full latent reaches 41.15 AP; identity decoded from the missing-object latent, combined with anchor identity and geometry, reaches 39.37 AP, leaving an unresolved 1.78-point residual. These results reveal a queryable, compositional 3D predictive state: the model completes context-dependent entity content, which downstream computation combines with metric geometry.

Understanding 3D scenes is fundamental to embodied intelligence, requiring joint reasoning over heterogeneous information from multiple modalities, including visual and geometric cues. However, the relevance of these modalities often varies across queries. Existing Multimodal Large Language Models (MLLMs) typically rely on fixed modality combinations, overlooking query-dependent modality needs. Such a rigid design can introduce semantic noise from irrelevant modalities while underutilizing more informative ones, leading to wasted computation and diluted reasoning. To address these challenges, this paper proposes SmartMage, a unified MLLM that dynamically orchestrates heterogeneous modalities for semantic-aware 3D scene understanding. Specifically, SmartMage incorporates: (1) a Semantic-guided Modality Adaptive RouTing (SMART) module that selects task-relevant modalities using semantic priors, text-modality alignment, and modality quality; and (2) a Modality-Aware Gating Expert (MAGE) module that leverages modality priors to guide expert activation, fostering adaptive specialization in multimodal reasoning. Empirically, SmartMage achieves state-of-the-art performance across five 3D scene understanding benchmarks, and attains competitive results on RGB-only video understanding benchmarks. In our diagnostic benchmark ScanFacet, tasks are divided into fine-grained semantic categories, enabling analysis of modality combinations preferred by each semantic type. The observed modality-semantic patterns provide further evidence of SmartMage's effectiveness. Project page: https://yuecheong.github.io/SmartMage/.

In this work, we explore an alternative paradigm for spatial reasoning by explicitly disentangling 3D perception from reasoning, rather than jointly acquiring implicit 3D perception and reasoning through large-scale training. Our key observation is that modern perception models excel at estimating continuous 3D geometry, whereas large language models (LLMs) are particularly effective at compositional and symbolic reasoning. Motivated by these complementary strengths, we propose the Disentangled Spatial Reasoner (DiSR), a simple yet effective framework that reconstructs the physical world into structured 3D evidence using off-the-shelf expert perception models and fine-tunes an LLM with LoRA to perform reasoning solely over this explicit geometric evidence. Without large-scale 3D VQA training or complex tool-use policies, DiSR achieves competitive performance on popular spatial reasoning benchmarks. Beyond its strong performance, DiSR offers improved interpretability, modularity, and computational efficiency, demonstrating that explicit separation of perception and reasoning is a scalable and effective alternative paradigm to end-to-end modeling for spatial intelligence.

3D vision-language models (3D VLMs) enable spatial reasoning over multi-view scenes but suffer from substantial token redundancy due to duplicated observations and large uninformative regions, leading to high computational cost. Although visual token compression has shown promise in accelerating 2D VLMs, it fails to capture the structured nature of 3D scenes and leads to incomplete spatial coverage and loss of fine-grained details. In this paper, we propose \textbf{HiSC}, a training-free framework for hierarchical spatial clustering token compression in 3D VLMs. HiSC lifts token compression from token-level selection to cluster-level processing by organizing tokens into spatially grounded clusters using joint geometric and semantic cues. Specifically, we first introduce a \textbf{spatial graph-based merging (SGraM) strategy} that models cross-view redundancy as spatial connectivity and consolidates physically consistent regions, effectively merging extremely similar redundant tokens prior to LLM inference. We then propose a \textbf{spatial clustering-based pruning (SCluP) paradigm} within LLM inference, which performs hierarchical compression across clusters and within clusters, preserving object instance completeness while retaining fine-grained details for important regions. Extensive experiments on diverse 3D reasoning benchmarks show validate the effectiveness of HiSC, particularly under high visual token pruning ratios. Besides, HiSC achieves over 90% token reduction with minimal performance degradation. Code is accessible at https://github.com/elecreak/HiSC.

3D Language Gaussian Splatting embeds open-vocabulary language features into 3D Gaussian Splatting, providing an efficient explicit representation for text-driven 3D scene understanding. However, existing methods are limited to indoor or small-scale scenes, and tend to fail in Unmanned Aerial Vehicle (UAV) outdoor scenes, where severe occlusions and long distance viewpoints often lead to incorrect semantic activations and missing target responses. In this paper, we present OutLangSplat which adapts language Gaussian representations to UAV outdoor scenes by improving feature representation and aggregation reliability. For the feature representation, a 2D-3D dual-branch representation with region-based alignment and fusion is designed to improve spatial consistency, reducing incomplete target responses and background misactivations. For the feature aggregation, we introduce a training-free contribution and consistency-aware Gaussian feature aggregation strategy that leverages pixel contribution reliability and cross-view semantic consistency to suppress unreliable responses from noisy viewpoints. A new dataset is provided by manually annotating various objects on four real-world public UAV outdoor scene datasets. To the best of our knowledge, it is the first accessible dataset of open-vocabulary 3D scene understanding for UAV outdoor scenes. Quantitative evaluations and ablation studies demonstrate that OutLangSplat outperforms SOTA methods on both open-vocabulary semantic segmentation and instance localization tasks. The datasets and codes will be open-sourced.

Detection transformer models, including DETR and its extensions, learn to output a set of object-level embeddings that can be simultaneously decoded into 2D bounding boxes and class distributions. In this paper, we investigate what pre-trained 2D detection transformers understand about the 3D properties of objects. Specifically, we investigate the extent to which properties including the depth of objects from the camera and the 3D location of objects relative to the camera can be recovered from object-level embeddings using linear and non-linear probes. Across a range of detection transformer models, our results show a surprisingly strong and previously unknown ability of 2D DETR models to represent useful information about the 3D properties of objects, despite the complete lack of 3D supervision during model pre-training.

Recent 3D vision-language models (3D VLMs) construct geometry aware tokens by projecting 2D visual features into world coordinates, enabling spatial reasoning for tasks such as 3D question answering. However, this design generates thousands of tokens per scene, resulting in substantial computational and memory overhead. While token compression has been extensively studied in 2D VLMs, existing approaches rely on semantic relevance or attention-based selection that overlook the structured spatial nature of 3D tokens. Moreover, redundancy in 3D representations cannot be resolved by spatial proximity alone, as object-level token imbalance persists even after spatial aggregation. To address this, we propose 3DZip, a three-stage token compression framework that first applies coarse voxelization to remove point-level redundancy, then selects anchor tokens based on feature-space diversity via a Determinantal Point Process, and finally merges remaining tokens under spatial constraints to preserve geometric coherence. Experiments on three 3D question answering benchmarks demonstrate that 3DZip consistently outperforms existing compression methods, retaining 94.7% of the original performance with only 128 tokens, achieving a $1.92\times$ faster inference speed.

3D scene understanding requires reasoning about entity existence, spatial layout, and object relations, yet RGB images alone often provide insufficient 3D cues. Existing 3D-VLMs commonly rely on depth or 3D-position-aware inputs at inference time, introducing additional acquisition, reconstruction, or annotation costs that limit RGB-only deployment. We therefore study how training-time 3D evidence can be converted into spatial reasoning capabilities retained under RGB-only inference. We propose a privileged-evidence distillation framework that constructs a distillable teacher through a unified evidence interface and controlled residual injection, and transfers its knowledge to a deployable student receiving only RGB images and questions through logit and structured representation distillation. To avoid imitating teacher signals unsupported by RGB, we further introduce evidence-sensitivity-guided distillation, which uses corrupted evidence to identify highly evidence-dependent targets and down-weight their supervision. We also define a recoverability decomposition based on the matched baseline, teacher, and student, separating privileged gains into RGB-recoverable improvements and residual teacher advantages. Across four benchmarks, the teacher achieves the best result on 7 of 11 reported metrics among the compared methods. The RGB-only student outperforms its matched baseline on all 11 metrics, including gains of 10.4 ScanQA CIDEr and 19.1 Scan2Cap CIDEr@0.5, without additional inference-time inputs. These results validate the effectiveness of training-time privileged 3D evidence distillation for both teacher performance and deployable RGB-only spatial reasoning. Separately, our matched baseline-teacher-student analysis characterizes privileged-gain transfer across evidence types and spatial skills.

Open vocabulary 3D scene understanding is essential for next-generation interactive systems, empowering users to intuitively query and navigate reconstructed environments using natural language. However, current 3D Gaussian frameworks are often bottlenecked by restrictive multiview capture requirements, costly scene-specific optimization, and the massive memory overhead of storing dense language features. We present a novel, training-free pipeline that fundamentally reimagines this paradigm by explicitly decoupling 3D geometric reconstruction from semantic integration. Given a standard monocular video sequence as input, our method efficiently outputs a compact, highly interpretable, and fully searchable object-level semantic Gaussian map. Rather than entangling heavy language embeddings within the mapping loop, we extract geometry independently and ground semantics through a lightweight, modular post-processing framework. Extensive evaluations on the Replica dataset demonstrate that this decoupled architecture preserves strong rendering fidelity and competitive segmentation accuracy. Crucially, by replacing dense per-Gaussian storage with modular, object-level semantic embeddings, our approach delivers an order-of-magnitude reduction in memory usage compared to SOTA baselines. This provides a highly efficient, scalable, and practical solution for open-vocabulary 3D retrieval and question answering directly from everyday monocular video.

3D Gaussian Splatting (3DGS) has emerged as an effective representation for novel view synthesis and 3D scene reconstruction, creating an increasing demand for reliable quality assessment. Unlike conventional image quality assessment (IQA), the quality of a 3DGS scene depends not only on the perceptual fidelity of rendered views, but also on scene-level factors such as spatial structure and cross-view consistency. Existing IQA methods are limited by their reliance on 2D perceptual cues, whereas general multimodal large language models (MLLMs) are not designed for stable quality regression and may produce unreliable judgments. To address these limitations, a multimodal quality assessment framework is developed for 3DGS scene understanding. First, a 3D-aware quality representation learning framework is introduced by augmenting a VGGT-based encoder with a dedicated quality head. Multi-view images are encoded into view-specific features and aggregated to capture cross-view consistency, while geometric cues are incorporated through joint modeling of depth and point-cloud-related structural information, enabling the learning of structure-aware quality representations beyond appearance-driven features. Second, a grounded multimodal reasoning mechanism is constructed by jointly feeding original images, depth maps, point cloud renderings, and camera parameters into a Qwen-based MLLM.

Camera-only 4D occupancy forecasting enables autonomous vehicles to predict future 3D semantic scenes solely from historical multi-view images, which is critical for driving safety. Even though current methods have achieved good performance, the strong spatial-temporal modeling between the input multi-view frames is still underexplored, which limits the performance of those methods in future 4D forecasting. To address this gap, we introduce a novel framework, InterOCF, for 4D occupancy forecasting that jointly models temporal dynamics in both 3D voxel-based representations and multi-view segmentation sequences, while explicitly incorporating feature interaction between the 2D and 3D branches. Our framework incorporates three core components: 1) A 3D Spatio-Temporal (3DST) module that learns volumetric dynamics from historical voxel states to predict future voxel states; 2) A 2D Spatio-Temporal (2DST) module employing an auxiliary multi-view temporal segmentation forecasting task to enhance temporal semantic dynamics; 3) A Spatio-Temporal Interaction Modeling (STIM) module that enables feature interaction between 2D and 3D representations. Experiments on the nuScenes, Lyft-Level5, and nuScenes-Occupancy datasets show that InterOCF consistently outperforms existing baseline approaches.

Recent advances in Multimodal Large Language Models (MLLMs) have triggered the development of end-to-end MLLMs for autonomous driving. However, the main emphasis to date has been for MLLMs using 2D images and videos. In contrast, this paper considers MLLM effectiveness using 3D sensors, particularly LiDAR and stereo cameras. LiDAR presents unique challenges to integration within an MLLM, largely because of data sparsity and lack of a grid structure for the data. For similar reasons, fusion of camera and LiDAR data within an MLLM pipeline is also uncommon. However, most autonomous systems rely on LiDAR-based sensing, and incorporating 3D data has been proven to improve performance in traditional 3D scene perception tasks. This paper presents D3VL, a novel MLLM framework that integrates 2D and 3D time-series data in a single but simple architecture. The model aims to answer questions involving traffic scene understanding and safety. D3VL shows an 11% improvement in the KITTI Question-Answering (QA) dataset compared to baseline methods in processing 2D and 3D time-series data. This paper further introduces the Waymo QA dataset extension, which assesses models' capabilities in processing 3D and time-series data under diverse driving conditions. D3VL implementation code and WaymoQA extension can be found on our supplemental website: https://automotivesafety-lvlm.github.io

Edit-conditioned 3D scene retrieval pairs a reference 3D room with a natural-language modification and retrieves rooms from a corpus that satisfy the edit. Three lines of prior work each fall short on this task. 2D composed image retrieval reasons over pixel-level edits and has no primitive for 3D object sets. 3D foundation encoders embed individual objects but cannot compose at the scene level. 3D scene-grounding methods localize references inside a static scene rather than rank modified rooms across a corpus. We present CR-Refiner, a training-free reranker that wraps any base retriever's top-K candidates with three components. A frozen LLM parses the edit into a structured query entity, and each candidate is scored by an unbalanced optimal-transport problem over a 1xG cost matrix coupling category, style, material, and geometry. The unbalanced solver lets the single-entity query drop mass on irrelevant objects, modelling the asymmetry directly. An axis-conditional structural prior adds size-keyword cues for geometric edits and subject-anchor direction cues for spatial edits. An LLM verifier refines the top three candidates with continuous confidence. Because no benchmark evaluates compositional matching over 3D object sets, we additionally release 3D-CER, 4,963 edit-conditioned queries over a 23,381-room indoor corpus across five edit axes, with multi-positive ground truth, CIRR-style hard subsets, and zero-target adversarials. Across three qualitatively distinct base retrievers, CR-Refiner consistently improves hard-subset R@1 and mAP@10 on every edit axis.

Open-vocabulary 3D maps let robots answer language queries about what and where, but they assume a static world and cannot answer queries about how scene elements behave. We introduce Vision-Language-Motion Maps (VLMM), an open-vocabulary, language-queryable 3D map - queried through a rule-based intent router over open-vocabulary object nouns, not a general natural-language interface - in which each element carries a fused motion attribute: a VLM/LLM semantic movability prior combined with geometrically observed cross-frame motion, together with a per-element uncertainty. Queries reduce to attribute filters that distinguish what has been seen to move, what could move but has not, and what stays still. On a controlled simulator benchmark with exact ground truth (AI2-THOR, three scene types) we show through ablation that the schema fields are non-substitutable: a semantic-only baseline fails motion queries even with strong features, and neither motion field substitutes for the other (the prior cannot answer "what is moving," observed motion cannot answer "what could move"). On real dynamic RGB-D (TUM and Bonn, six sequences) we show the uncertainty channel - our key difference from prior fused-motion work - consistently improves moving-vs-static average precision and reduces false motion flags, and that it is robust to estimated (noisy) poses. The raw confidence is not calibrated, but post-hoc isotonic calibration reaches an expected calibration error of 0.10. VLMM is a representation contribution: the closest prior maps each lack at least one of the four properties - open-vocabulary, language-queryable, fused prior-and-observed motion, and per-element uncertainty - that our combination provides.

In goal-directed embodied navigation, where an agent must locate a specified target in an unseen environment, 3D scene understanding and navigation reasoning must work in concert. Current approaches transmit 3D scene information to vision-language models (VLMs) through text, suggesting a representation gap in our tested configurations; a controlled ablation confirms that direct embedding-level transfer significantly outperforms the evaluated text serialization formats. We introduce SoftNav, which injects entity-level 3D continuous representations -- one token per detected object or frontier -- into a VLM's hidden space as soft tokens through a lightweight projector. With the 3D encoder and VLM frozen, only ~1,200 samples and ~17M trainable parameters are needed. On HM3D-OVON, SoftNav achieves 74.2%/68.3%/66.7% SR across three splits, surpassing all prior methods in both SR and SPL; the same navigation policy transfers zero-shot to GOAT-Bench (67.2% SR), SG3D (47.2% s-SR), and real-world robot deployment without retraining or architectural modification. Injecting 3D scene tokens directly into VLMs bridges the representation gap, enabling transferable navigation with minimal training.

Vision-Language-Action (VLA) models have emerged as a powerful end-to-end paradigm for robotic manipulation by mapping language instructions and 2D visual inputs directly to actions. However, these models lack an explicit, scene-level 3D representation, limiting their ability to reason over spatial layouts and geometric constraints. While recent efforts incorporate explicit 3D cues, such as depth maps or point clouds, to improve geometric awareness, they primarily capture low-level structures and lack high-level semantic grounding in 3D space. In human cognition, interaction with the physical world relies on a 3D semantic cognitive map - an internal mental model that integrates spatial layouts with semantic context to enable persistent, viewpoint-invariant reasoning. In light of this, we present VistaVLA, a novel two-stage framework that constructs a geometry- and semantics-aware 3D cognitive representation from 3D Gaussian primitives and grounds it as compact context tokens for VLA policy learning. Specifically, VistaVLA lifts multi-view vision-language features into 3D Gaussian primitives, forming geometry-anchored semantic tokens that align view-consistent spatial grounding with 2D visual feature spaces. To make this 3D representation computationally tractable for effective VLA control, we introduce Merge-then-Query (MtQ), a token summarization mechanism. MtQ compresses dense Gaussian primitives into a highly compact set of spatially informative tokens, achieving a 99% token reduction while preserving action-relevant 3D layouts and semantic context. Extensive evaluations in both simulated and real-world environments demonstrate the effectiveness of VistaVLA. Notably, in real-world scenarios, VistaVLA improves success rates by 22.8% across seven real-world tasks and by 30.0% over the VLA-Adapter baseline on challenging out-of-distribution tasks.

Open-vocabulary 3D scene understanding is commonly achieved by embedding 2D vision-language features such as CLIP into a 3D Gaussian Splatting scene, turning it into a text-queryable semantic field. However, attaching a high-dimensional feature to each of millions of Gaussians inflates a single scene to gigabytes, which makes storage and deployment the real bottleneck of these fields. Existing compact methods each learn and ship a per-scene codec, an autoencoder, a quantized codebook, or a distilled feature field, entangling field construction with field storage and never compressing the per-Gaussian assignment that holds the bulk of the cost. We argue that construction and storage should be decoupled, and that storage is a rate-distortion problem over the per-Gaussian binding to a small anchor table, a structure no prior open-vocabulary method compresses. We present CoSAG, which constructs the field without any per-scene training through a closed-form transmittance-weighted lift, spatially grounded semantic anchors, and multi-view denoising, and stores it with a spatially predictive entropy coder that ships no decoder. Because the anchors are spatially grounded, the binding is predictable and therefore highly compressible. The transmittance-weighted lift and multi-view denoising yield a clean, view-consistent assignment, so the entropy coder spends almost no rate on correcting noise and instead codes only the residual against its spatial prediction. CoSAG reaches sub-megabyte storage while matching or exceeding the state of the art across the 2D-rendered, 3D-selection, and dense-LSeg protocols, reducing field size by 37 to 76x relative to LangSplatV2 at higher accuracy.

Recent benchmarks for VLMs largely assess single- or limited-view perception, leaving untested the core cognitive ability to integrate observations across viewpoints into a coherent, world-centric (allocentric) 3D mental model. We introduce MultiView-Bench, a diagnostic benchmark expressly designed to evaluate multi-view integration for holistic 3D scene comprehension. Unlike existing datasets that focus on pixel-level mapping or camera-relative navigation, MultiView-Bench requires models to decouple object positioning from transient perspectives and ground them in a fixed global coordinate system. This capability serves as a prerequisite for VLMs before being deployed for downstream tasks such as mechanical part assembly. Our systematic evaluation of frontier VLMs reveals consistent failure modes: strong performance on 2D planar relations from a single image, but marked difficulty with 3D spatial relations and with aggregating information across views. We further identify biases in VLMs, such as struggles with unconventional axis directions and sensitivity to object colorways and texture variations. Acknowledging these limitations, we propose ViewNavigator, a multi-agent framework that actively selects informative viewpoints, perceives, and fuses multi-view evidence, improving diverse base models on MultiView-Bench even under a strict budget-matched comparison (and by 3-5x for the full agent).

Existing 3D scene-grounded Large Language Models (3D-LLMs) focus on answering questions grounded in simplified single-room 3D scenes, lacking the ability to reason over real-world household environments containing multiple interconnected rooms and diverse object categories. We introduce CAIRN, a topology-aware 3D-LLM for multi-room 3D scene understanding. CAIRN aligns transformer attention with scene hierarchy, giving the model explicit awareness of object-level relations and room-level connectivity. It enriches object tokens with room-local relational context via a graph neural network, introduces learned room tokens for room-level abstraction, and applies a hierarchical attention mask with geometric bias to route information according to scene topology. CAIRN is developed on CAIRN-MR, a benchmark we introduce on HM3D for multi-room 3D scene understanding, covering grounding, captioning, and four question-answering tasks that progressively evaluate from intra-room perception to cross-room reasoning. Experiments show that CAIRN outperforms prior 3D-LLMs by a large margin across all CAIRN-MR tasks while remaining competitive on five single-room benchmarks.

3D dense captioning, an emerging vision-language task, aims to generate descriptive sentences for each object in the 3D scene. Despite the impressive results achieved by previous methods, they suffer from two limitations. First, current research often employs global rigid transformations, such as rotation, to augment scenes without changing their spatial layouts. However, diverse spatial layouts are crucial for training a 3D dense captioning model to describe spatial relations between objects. Second, previous works mainly focus on the design of the caption generation pipeline while utilizing a simple network architecture for other components, i.e., backbone and detection head, which is crucial for extracting rich semantic information for captioning. In this paper, we propose PVCap to alleviate the aforementioned problems. Our PVCap consists of PseudoCap and VoxelCapNet. Specifically, PseudoCap employs a random mixing technique on instances within the dataset, generating numerous pseudo frames with diverse spatial layouts at the instance level. By utilizing a teacher-student framework, PseudoCap obtains pseudo caption labels for these pseudo frames. This data augmentation approach significantly increases the number of training samples and enhances the model's ability to describe the environment effectively. Regarding VoxelCapNet, we introduce a robust caption network that utilizes voxel features and adapts the caption head to the voxel-based network architecture. Our VoxelCapNet can serve as a competitive baseline for future research on 3D dense captioning. Extensive experiments are conducted on two prevalent benchmarks, i.e., ScanRefer and Nr3D. Notably, our method surpasses current state-of-the-art by 11.41% and 13.99% in CIDEr@0.5IoU, respectively. Codes will be made publicly available.

Natural-language queries about 3D environments become actionable when responses are verifiable and metric. Verifiability requires explicit grounding to the referred 3D region, while metric answers report physical measurements in real-world units (e.g., size, thickness, clearance, and distance). Existing 3D large multimodal models (LMMs) approaches remain limited: conversational systems typically respond without explicit 3D grounding, while 3D grounding models are not designed for interactive, metric-aware dialogue. In this paper, we present Ground3D-LMM, a unified model that takes a point cloud and an optional RGB image as input and supports 3D spatial conversation with (i) point-grounded responses and (ii) metric numeric outputs at both object and part granularity, including multi-object queries. To evaluate this intersection of grounding and measurement, we define the 3D Grounded Measurement task, which requires predicting the referred 3D region and the corresponding metric quantities in real-world units. We introduce a large-scale dataset built on ScanNet and ScanNet++ datasets with dense object and part annotations and roughly 2.5M question-answer pairs spanning eight tasks, along with a manually verified test set. Extensive experiments on multiple datasets and tasks show that our proposed Ground3D-LMM model provides a strong baseline for grounded, metric-aware 3D conversational understanding. Our dataset and model are publicly available.

Open-vocabulary 3D scene understanding aims to segment 3D scenes beyond predefined categories by transferring semantic knowledge from vision-language models. Existing methods have advanced this task by lifting language-aligned 2D features into 3D, yet they often rely on context-independent semantic representations, leaving object relationships underexplored for contextual refinement. We propose RelGraphOV, a relationship-aware framework that uses 3D scene graphs to enhance open-vocabulary 3D understanding. Our method constructs relational scene graphs from multi-view observations by leveraging vision-language reasoning to infer object relationships and prune geometrically implausible connections, without manual relationship annotations. To aggregate relational context while avoiding feature interference, we introduce an Adaptive Gated Dual-Stream Contextual GAT that separates dense geometric features and semantic CLIP embeddings, performs edge-guided message passing, and adaptively fuses complementary semantics. A hierarchical contrastive objective further promotes instance-level consistency and category-level discrimination. Experiments on ScanNetV2, ScanNet200, ScanNet$++$, and Replica demonstrate strong performance and generalization ability. Project Page: https://cxavireh.github.io/relgraphov-projectpage

Recent Multi-modal Large Language Models (MLLMs) have demonstrated remarkable performance on 2D question answering tasks. However, extending these models to the 3D question answering remains challenging, as they typically require multiple views of the scene, which incurs substantial computational cost at inference. To mitigate this issue, existing solutions rely on strategic frame selection or token-merging algorithms that require preprocessing in advance all frames of the scene, i.e., an offline fashion. In contrast, we propose the first online token-pruning method that can be integrated seamlessly with current MLLM models for 3D question answering tasks, without additional training and with lower memory usage.Our key insight is to project each input frame into a shared voxel space using depth information and camera pose, identifying spatially-overlapped regions across frames and selectively pruning redundant image tokens before they enter the language model. Our method enables efficient online processing while reducing up to 50% of token usage. We apply this approach to Qwen2.5-VL-7B and Qwen3-VL-8B, demonstrating improved performance on the ScanQA, SQA3D, and OpenEQA-HM3D benchmarks.

Feed-forward 3D models are commonly trained using either expensive geometric supervision or self-supervised photometric objectives, both of which provide incomplete learning signals. We introduce Vision-Language Reprojection Consistency (VLRC), a scalable auxiliary objective that exploits frozen vision-language representations as semantic multi-view supervision. Given a predicted 3D reconstruction, VLRC reprojects dense vision-language features across views and enforces feature consistency between corresponding image locations, requiring no additional 3D annotations. The objective integrates seamlessly with both self-supervised monocular reconstruction and supervised-pretrained feed-forward 3D models during unlabeled adaptation. By aligning geometry with language-grounded features, VLRC not only improves depth and camera estimation but also enables more coherent multi-view semantic fusion for open-vocabulary 3D scene understanding. Experiments on indoor and outdoor benchmarks demonstrate consistent gains in 3D reconstruction accuracy and zero-shot open-vocabulary 3D semantic segmentation.

Generative diffusion models excel at synthesizing high-quality images, videos, and 3D content under multimodal control. However, arbitrary user-defined modality-to-4D (X-to-4D) generation remains challenging due to the high cost of constructing diverse datasets and the limited scalability of existing methods. This paper presents Align4D, a flexible framework that translates any-modal input into coherent video-3D pairs, using video to guide 4D motion and 3D data to shape 4D geometry. Align4D introduces three key techniques: (1) Object Distance Alignment, which searches Video-Aligned and Multiview-Aligned Object Distances (VAOD/MAOD), respectively, to reconcile 4D renderings with video and the priors of multiview diffusion models; (2) Motion-Geometry Joint Alignment, which constrains known and unknown views through synchronized video and 3D inputs, ensuring consistent 4D generation; and (3) Asynchronous Optimization, which decouples Gaussian attribute and deformation network training to enhance motion and geometry fidelity. We further propose the X4D dataset, which integrates prompt, image, video, and 3D data for benchmarking. Experiments on X4D and Consistent4D demonstrate that Align4D achieves state-of-the-art quality and consistency in X-to-4D generation. Project page: https://miaoqiaowei.github.io/Align4D/.

Comprehensive 3D scene understanding from sparse, unposed images requires a model to recover renderable geometry, open-vocabulary semantics, and free/occupied 3D space without relying on external camera calibration. Recent feed-forward Gaussian methods improve pose-free reconstruction and semantic rendering, but their Gaussian primitives are mainly optimized through image-space objectives and remain weakly constrained in unobserved regions. We propose \textit{COVScene}, a pose-free semantic Gaussian framework that couples renderable Gaussian primitives with a dense semantic occupancy field through differentiable volumetric lifting. Instead of converting Gaussians to voxels only at evaluation time, COVScene lifts the predicted semantic Gaussians inside the training computation graph, so volumetric regularization provides gradients to Gaussian opacity, geometry, and semantic features. The framework combines a semantic-aware Geometry Transformer, multi-task Gaussian decoding, geometric foundation distillation, and occupancy entropy regularization to support novel view synthesis, open-vocabulary semantic querying, and semantic occupancy prediction within a single representation. Experiments on ScanNet and ScanNet++ show that COVScene maintains competitive rendering quality, improves open-vocabulary segmentation, and achieves stronger semantic occupancy prediction than the self-supervised baseline without direct voxel-level supervision.

Video predictive models are emerging as a powerful paradigm in robotics, offering a promising path toward task generalization, long-horizon planning, and flexible decision-making. However, prevailing approaches often operate on 2D video sequences, inherently lacking the 3D geometric understanding necessary for precise spatial reasoning and physical consistency. We introduce a Structured 4D Latent Predictive Model, which predicts the evolution of a scene's 3D structure in a structured latent space conditioned on observations and textual instructions. Our representation encodes the scene holistically and can be decoded into diverse 3D formats, enabling a more complete and 3D consistent scene understanding. This structured 4D latent predictive model serves as a planner, generating future scenes that are translated into executable actions by a goal-conditioned inverse dynamics module. Experiments demonstrate that our model generates futures with strong visual quality, substantially better 3D consistency and multi-view coherence compared to state-of-the-art video-based planners. Consequently, our full planning pipeline achieves superior performance on complex manipulation tasks, exhibits robust generalization to novel visual conditions, and proves effective on real-world robotic platforms. Our website is available at https://structured-4d-model.github.io/.

Recent advances in deep learning have notably improved steganographic message hiding. However, designing a generalizable steganographic approach for 3D Gaussian Splatting (3DGS) that can embed meaningful 3D scene content remains challenging. In this paper, we propose IBRSteG, a generalizable framework for 3DGS steganography that enables undetectable concealment of secret scenes within a steganographic scene. Unlike existing approaches whose parameter generation is rigidly coupled with the specific scene, we formulate 3D steganography as a feed-forward 3D Gaussian embedding process that generalizes across different 3DGS scenes. To realize this, we introduce GAS (Gaussian Attributes Steganographer), a network that learns a scene-independent embedding function by injecting the attributes of secret 3D Gaussian points into a cover scene, thereby directly reconstructing the steganographic scenes without per-scene finetuning or optimization. By transforming 3D Gaussian into these structured attributes, these attributes are compatible with 2D learning paradigms and benefit from their structured nature, thereby enhancing generalization to unseen 3DGS scenes. Extensive experiments on established datasets demonstrate that IBRSteG can effectively conceal different scenes with high visual quality, and achieves superior capacity and security. Code is available at https://github.com/LingXiang2023/IBRSteG.

3D Gaussian Splatting~(3DGS) has emerged as a promising paradigm for reconstructing streamable free-viewpoint video~(FVV) from multi-view videos. However, 3DGS-based FVVs typically lack user interaction and editing capabilities, which diminishes the immersive experience. Recent research has integrated language features from CLIP into 3DGS via distillation, enabling open-vocabulary queries and supporting many downstream applications. Nevertheless, the stringent requirements of FVV, low frame size and high FPS, make current language Gaussian representations unsuitable for language-embedded FVV. In this paper, we propose DLGStream, a novel language-embedded FVV representation that streams time-varying language features alongside Gaussian attributes to support 4D environment interaction, scene editing, and spatial intelligence. Specifically, we propose a dual-opacity dynamic language Gaussian representation, which maintains two opacity attributes for color and language features to deal with performance degradation that occurs when colors and features are jointly optimized. Furthermore, we introduce an interpolation-based deformation field to reduce temporal redundancy. This deformation field can also be used for 4D frame interpolation, boosting FVV sequences from low to high FPS. Experimental results demonstrate that DLGStream achieves superior performance in both on open-vocabulary segmentation and reconstruction quality with an average frame size of merely 43 KB. The code is available on \href{https://github.com/kkkzh/DLGStream}{https://github.com/kkkzh/DLGStream}.

Recent advancements have explored agentic zero-shot 3D understanding by reformulating it as video keyframe understanding with Multimodal Large Language Models (MLLMs). However, existing methods face an intrinsic bottleneck due to the finite observation perspectives inherent in videos and the implicit perception of 3D scenes. In this paper, we propose a collaborative multi-agent framework that assigns a Planning Agent to handle high-level viewpoint planning and supplement novel perspectives, and a Perception Agent to explicitly summarize the 3D scene into a structured holistic cognitive map. Specifically, Planning Agent first analyzes this cognitive map to determine query-relevant viewpoints and supplements missing critical perspectives to ensure comprehensive observation. Subsequently, Perception Agent documents object-level attributes from these views by assigning consistent instance identifiers across viewpoints, thereby integrating fragmented observations into the holistic cognitive map. In parallel, it provides feedback to filter out mismatched candidate objects and guide subsequent viewpoint planning. Through this closed-loop iterative process, two agents collaboratively figure out candidates until Perception Agent determines that sufficient information has been captured to complete the task. Extensive experiments demonstrate that our method achieves state-of-the-art performance on 6 benchmarks, with improvements of 11.1% Acc@0.5 on ScanRefer, 14.6 BLEU-1 on 3D-assisted dialog, and 2.1 EM on SQA3D.

Meshes are among the most common 3D scene representations, but directly generating meshes is challenging because the representation contains important symmetries, including permutation invariance of faces and vertices. MeshFlow learns to generate triangle meshes directly as triangle soups, avoiding the need to serialize meshes into long autoregressive sequences. We adopt equivariant optimal-transport flow matching models that respect the key symmetries of triangle soups: arbitrary permutations of faces and permutations of the vertices within each face. Toward this goal, we propose a simple yet effective modification to the Diffusion Transformer architecture, resulting in a scalable network capable of modeling a velocity field while maintaining the desired equivariance. We further introduce an optimal-transport-based training objective that improves convergence by eliminating supervision signals that violate these symmetries. MeshFlow achieves mesh quality comparable to state-of-the-art autoregressive mesh generators while providing about an 18$\times$ speedup during inference. Project page is at https://qiisun.github.io/MeshFlow/.

We present Casper3D, a lightweight probabilistic framework for converting noisy multi-view 2D foundation-model embeddings into a latent 3D semantic representation. We model view-level semantic features as noisy observations of an underlying 3D semantic state and infer this state with a set-based variational model that incorporates relative pose during multi-view reasoning. Casper3D is trained by predicting held-out semantic observations from novel viewpoints, while remaining aligned with visual and text semantic spaces for open-vocabulary 3D understanding. The framework is backbone-agnostic and applies to both language-aligned and self-supervised embeddings. Experiments show that Casper3D produces more stable 3D semantics than simple multi-view pooling, especially in ambiguous and noisy settings.

Recently, vision-language models (VLMs) have made significant progress in 3D scene understanding, driving advances in applications such as embodied intelligence and robotic vision. However, existing approaches typically either rely directly on explicit 3D inputs (e.g., point clouds or RGB-D sequences), or introduce an additional 3D geometry encoder to derive 3D-aware visual tokens from 2D images. Such designs structurally decouple 3D geometric perception from the rich 2D semantics learned via vision-language pre-training, hindering the development of a unified 3D vision-language representation. In this work, we propose Occ-VLM, a novel framework for 3D scene understanding that operates purely on posed RGB images and employs a single 2D vision encoder. Specifically, Occ-VLM reconstructs 3D scene occupancy as an auxiliary geometric prior, which is utilized to spatially associate foreground 2D tokens with 3D space. These tokens are then decoded by a Large Language Model (LLM) for unified scene understanding. Extensive experiments demonstrate that Occ-VLM achieves both accurate geometric perception and robust vision-language reasoning: it attains state-of-the-art performance on multi-view occupancy prediction, while performing on par with 3D-input VLMs on 3D Visual Question Answering (VQA) and 3D dense captioning benchmarks.

Efficiently retrieving specific 3D instances from large-scale scenes via natural language prompts remains a formidable challenge in multimedia analysis. Existing approaches predominantly follow a "scene-level embedding" paradigm, which requires distilling high-dimensional semantic features into every 3D primitive. This strategy suffers from a fundamental architectural bottleneck: memory and computational costs scale linearly with scene complexity, inevitably triggering out-of-memory (OOM) failures in city-scale environments. To address this barrier, we propose QueryGaussian, a training-free framework for expeditious and scalable open-vocabulary 3D instance retrieval. Unlike holistic semantic distillation, QueryGaussian employs an instance-level query mechanism that decouples semantic understanding from geometric representation. Specifically, we leverage pre-trained 2D vision models to interpret user prompts and lift segmentation masks into 3D via a concurrent maximum-weight association strategy, ensuring semantic-visual consistency. To mitigate projection ambiguity, we introduce a temporal fusion module with multi-stage adaptive density clustering. Experimental results demonstrate that QueryGaussian not only matches the accuracy of state-of-the-art methods but also delivers a decisive efficiency leap, reducing GPU memory usage by over 70% and accelerating inference by 180x. Crucially, QueryGaussian enables expeditious instance retrieval on city-scale scenes containing tens of millions of Gaussians using consumer-grade hardware.

We introduce 3D-DLP, a self-supervised object-centric representation learning model that decomposes scene-level RGB-D or voxel observations into a set of 3D latent particles. Building on the Deep Latent Particles (DLP) framework, each particle encodes disentangled attributes, including 3D keypoint position, bounding box dimensions, and appearance features, and represents a distinct entity in the scene. The model learns interpretable per-particle segmentation maps through an end-to-end self-supervised reconstruction objective. We demonstrate on both simulated and real-world datasets that the learned latent space is interpretable and controllable: by manipulating particle positions and decoding, we can generate novel scene configurations. Furthermore, we show that leveraging these compact 3D latent particles for downstream robotic manipulation improves performance over baselines that either lack explicit 3D information or rely on memory-intensive dense 3D inputs without object-centric structure. Code and videos are available at https://eubooks3003.github.io/3d-dlp.

Existing approaches to 3D scene understanding in Vision-Language Models (VLMs) either rely on complex, model-specific geometry encoders or large training budgets in pursuit of spatial reasoning. Instead, OneCanvas aggregates patch features from all views onto a single equirectangular panoramic canvas. Namely, each patch is unprojected to a 3D world coordinate using its depth and camera pose, then placed on the canvas at the continuous longitude and latitude of that point as seen from the canvas origin, with no rasterization or aggregation across overlapping views. A 3D position embedding of the patch's metric coordinates is added to its feature, restoring the depth lost when collapsing the world position to an angular canvas coordinate. Patches from all frames thus share one spatial coordinate system with no fusion or major architectural modifications of the backbone. The pretrained VLM consumes this representation as if it were an ordinary image. Because the canvas can be centered on any pose of interest, the same representation directly supports situated reasoning from a specific viewpoint, a common requirement in robotics and embodied AI. Thanks to this representation, we can also introduce a spatial pretraining curriculum: by procedurally placing patch features of objects, drawn from real images, at chosen 3D world positions on an otherwise empty canvas, we generate on-the-fly supervision spanning a broad range of spatial reasoning tasks, with answer distributions controlled to reduce spatial reasoning shortcuts. OneCanvas achieves state-of-the-art accuracy on SQA3D and VSI-Bench, and generalizes to out-of-distribution data on SPBench, using an order of magnitude less training compute than the strongest competing methods.

Motion forecasting is central to visual intelligence: agents must anticipate how objects will move in order to plan actions, reason about physical interactions, and synthesize realistic futures. We argue that 3D points in world coordinates provide a general representation that is class-agnostic, view-stable, compact, and directly useful for downstream tasks. We formalize the task of goal-conditioned 3D point motion forecasting: given a short visual history, a set of 3D query points on an object of interest, and a language description of the intended goal, the model predicts the future 3D trajectory of each point. We introduce a full stack to study this task at scale: (1) MolmoMotion-1M is a large corpus of action-described, object-grounded 3D point trajectories annotated from 1.16M unconstrained videos; (2) PointMotionBench is a human-verified benchmark spanning 111 object categories and 61 motion types; and (3) MolmoMotion is a general motion forecasting model that supports both autoregressive coordinate prediction and flow-matching-based trajectory generation. MolmoMotion accurately predicts diverse motion patterns with different language instructions, and significantly outperforms existing motion prediction baselines on PointMotionBench. Finally, we show that the learned 3D motion prior transfers well to downstream applications: it improves training efficiency and generalization for robot manipulation, and its predicted trajectories provide effective motion guidance for generative models to synthesize videos with more realistic object motion.

Semantic 3D occupancy provides a voxelized world state for autonomous driving and robot decision making, but object and rare-class errors can affect free-space interpretation, collision checking, and temporal state propagation. We show that a common VLM strategy, aligning 3D voxel or object features with crop-caption embeddings, improves text-space similarity without reliably improving closed-set occupancy mIoU. Motivated by this mismatch, we propose VISA, a training-time semantic auditing approach for existing occupancy world models. VISA queries an offline VLM on a representative crop of each physical object instance, obtains a structured audit with class hypotheses, plausible confusions, reliability, attributes, and evidence, and propagates it along the object track. The audit is grounded to matched 3D object voxels and distilled into semantic logits through reliability-weighted taxonomy, attribute-factor, and scene-level audit graph losses, while inference remains unchanged and requires no VLM. On nuScenes, averaged across three runs, VISA improves OccWorld from 19.06 to 20.05 mIoU and GaussianWorld from 21.36 to 21.91 mIoU; on GaussianWorld, object mIoU improves from 18.18 to 19.16 and rare-class mIoU from 15.60 to 16.79. These results suggest that VLMs are better suited to closed-set occupancy as reliability-aware semantic auditors than as generic caption-embedding targets.

This paper introduces EPS3D, a new end-to-end feed-forward framework for open-vocabulary 3D panoptic segmentation. Unlike existing methods relying on additional preprocessing, we design an end-to-end architecture, with a distillation-based training strategy on diverse 3D scenes to predict 3D-aware semantic and instance features from multi-view images, improving 3D consistency and avoiding error accumulation. We further propose a mutual enhancement module to enforce inherent semantic-instance consistency. By aligning semantics within instances (Ins2Sem) and refining instance features with semantic guidance (Sem2Ins), we achieve more coherent 3D scene understanding. Ultimately, EPS3D outperforms SOTA baselines on two benchmarks (e.g., +13% mIoU for semantics on Replica) with high efficiency (e.g., 1s per scene), supporting tasks like robotic manipulation and 3D scene editing.

Recent advances in 3D multimodal large language models (3D-MLLMs) have enabled unified solutions for 3D scene understanding tasks, including visual question answering, captioning, and referring segmentation. However, existing 3D-MLLMs remain largely object-centric, limiting their ability to model fine-grained part structures that are essential for embodied interaction with 3D environments. In this work, we present PAR3D, a unified part-aware 3D-MLLM framework that enables models to understand, reason about, and ground both objects and their parts in 3D scenes. To enable training and evaluation of part-aware 3D scene understanding, we introduce ScenePart, a synthetic 3D scene dataset with part-level annotations and language instructions. We further develop Part-Aware 3D Representation Learning to enrich 3D visual representations with fine-grained part-level semantics, and propose Hierarchical Segmentation Query Generation to ground part targets via hierarchical object-part queries. Extensive experiments show that our method substantially improves part-level question answering and referring segmentation, while also achieving strong performance across object-level vision-language tasks.

Feed-forward 3D Gaussian Splatting methods reconstruct a scene from posed or pose-free images in a single forward pass, yet current approaches predict one Gaussian per input pixel, tying the representation budget to camera resolution rather than scene complexity. A flat wall and a richly textured object thus produce equally many Gaussians despite very different geometric needs. We propose ZipSplat, a token-based feed-forward model that decouples Gaussian placement from the pixel grid. A multi-view backbone extracts dense visual tokens, and k-means clustering compresses them into a compact set of scene tokens. Cross- and self-attention refine these tokens, and a lightweight MLP decodes each into a group of Gaussians with unconstrained 3D positions. Because clustering is applied at inference, a single trained model spans the quality-efficiency curve without retraining. ZipSplat operates without ground-truth poses or intrinsics, yet sets a new state of the art on DL3DV and RealEstate10K with ${\sim}6{\times}$ fewer Gaussians than pixel-aligned methods, surpassing the best pose-free baseline by 2.1dB and 1.2dB PSNR, respectively. It further generalizes zero-shot to Mip-NeRF360 and ScanNet++, outperforming all comparable baselines. Our project page is at https://veichta.com/zipsplat.

High-fidelity semantic 3D scene representations are crucial for numerous applications, including robotics, autonomous driving, and simulation. Beyond this, the ability to edit such representations enables developers to adapt these applications more easily to specific target scenarios. Current approaches provide limited support for controllable editing. We introduce TASE, a method that projects pretrained 2D semantic features into a truncation-aware embedding space to enable flexible 3D scene editing. Our method explicitly optimizes a feature space in which progressively reducing feature channels yields increasingly abstract semantic representations, while retaining more channels preserves fine-grained detail. Additionally, we improve multi-view consistency of the features using a scale- and translation-equivariance loss. The resulting truncation-aware embedding space enables text-driven edits to 3D scenes, providing explicit control over how strongly edits adhere to the original scene content and allowing more substantial modifications than prior methods. Moreover, we propose a finetuning stage for the editing diffusion model to mitigate artifacts caused by geometric changes. Experimental results demonstrate competitive performance in 3D scene editing, substantially outperforming prior methods on edits involving large geometric modifications.

Recently, zero-shot 3D scene understanding via 2D Vision-Language Models (VLMs) has gained increasing research interest due to their promising spatial reasoning capabilities. Typically, multiple 2D views are sampled from a 3D point cloud and fed into pre-trained VLMs to answer a given question. This paradigm highlights the critical role of input context quality and raises the challenge of retaining as many task-relevant 3D details as possible under a limited input budget. We propose \texttt{KeyVT}, a hierarchical approach for input context collection at both the view and token levels. Specifically, we combine pixel features with camera parameters and assess view importance based on both semantic content and geometric position, resulting in spatially consistent and task-relevant views. Furthermore, we address redundancy among patches across selected views by identifying representative tokens under the optimal transport (OT) framework, where view tokens and key tokens are formulated as two discrete distributions in the embedding space. These key tokens are expected to cover all view features by minimizing the OT distance. We evaluate our framework on three widely used benchmarks, demonstrating significant improvements over existing tuning-free methods and performance comparable to training-based approaches.

Open-set 3D recognition requires models that generalize to rare or unseen categories. Recent approaches address this by distilling language-vision knowledge into 3D encoders, typically relying on heavy 2D ViTs and aligning each point cloud with a single image or caption, thus anchoring representations to partial views. We propose aligning each point cloud with multiple images and textual descriptions to capture a more holistic understanding of 3D objects. To realize this idea, it is essential to design a loss function capable of jointly aligning a 3D instance with multiple matched signals, multi-view images and multiple texts, while separating positive aggregation from negative competition. We introduce such a function, termed the decoupled multi-positive contrastive loss. Our formulation enhances the loss's hardness-aware focus on challenging negatives, avoiding the "spotlight crowding" that occurs when many positives share the same softmax with all the negatives. Complementing this, we present a lightweight text adapter applied only to web captions, reducing the domain gap to curated annotations and enabling effective use of large-scale unsupervised text. Our model demonstrates state-of-the-art open-vocabulary performance on long-tail benchmarks, yielding substantial zero-shot improvements while sustaining high frame rates.

Vision Language Models (VLMs) enable a unified model to solve various vision tasks through prompting. They have shown promising performance in semantic understanding. However, 3D understanding still largely relies on expert vision models with complex task-specific designs. The key argument this work wants to make is that VLMs are native 3D learners. Our in-depth large scale study shows that 1) focal length unification, 2) text-based pixel reference and 3) data mixture and scaling, are all you need for effective 3D learning. Model architecture changes, large models, heavy data augmentations, and complex losses including the regression formulation, many of which form the foundation of expert vision models, are actually not necessary conditions. As a result, we propose VLM3, a scalable method with the simplest design that enables standard VLMs to master diverse 3D tasks. VLM3 not only advances the VLM depth estimation accuracy by a large margin (0.84 -> 0.9), but also enables diverse 3D tasks such as pixel correspondence, camera pose estimation and object-level 3D understanding, matching expert vision model accuracy while maintaining standard architectures and text-based training. We believe VLM3 opens up a new paradigm for simple and scalable 3D learning.

Point cloud segmentation is a fundamental task in 3D scene understanding. Its progress is constrained by the high cost and time required for dense 3D annotations, making labeled samples difficult to obtain. Beyond annotation scarcity, different sensing modalities face inherent limitations. 2D images provide rich texture and appearance cues, yet they lack explicit depth and geometric structure. In contrast, 3D point clouds capture accurate spatial geometry but are sparse and contain no texture information. As a result, relying on a single modality restricts the richness of learned representations and weakens generalization. Although recent multi-modal methods that combine 3D point clouds with 2D images have demonstrated strong performance in tasks such as classification and retrieval, they typically depend on large-scale labeled datasets and have not been fully exploited for data-efficient dense prediction. To address these limitations, we propose a novel cross-modal knowledge distillation framework, xModel-KD, for 3D point cloud segmentation. Our method exploits the complementary strengths of 2D texture and 3D geometry by learning unified per-point representations through cross-modal alignment. Specifically, we design a cross-modal fusion encoder trained with a contrastive objective that enforces feature consistency between corresponding 2D and 3D representations across multiple views. By integrating powerful pre-trained backbones with a targeted fusion strategy, the proposed framework effectively transfers appearance cues from images to geometry-aware point features. Experimental results show that cross-modal fusion achieves a 2% absolute improvement in mIoU over a LiDAR-only baseline, demonstrating the benefit of leveraging complementary multi-modal information for scalable and annotation-efficient 3D scene understanding.