Jun 2026· arXiv.org· Vol abs/2606.30638· 0 citations· 45 references
Computer Science
TL;DR
GaussDet is presented, a method that circumvents the need for dense CLIP features by leveraging discrete, open-vocabulary 2D object detectors with referring expression capabilities and achieves a substantial 16.7% mIoU improvement in referential grounding within a strict zero-shot setting.
Abstract
3D Gaussian Splatting (3DGS) has emerged at the forefront of 3D scene reconstruction. Extending 3DGS with language-driven, open-vocabulary understanding has gained significant attention for real-world applications such as embodied AI. Recent methods achieve this by learning an instance feature attribute and assigning semantics by distilling high-dimensional Contrastive Language-Image Pretraining (CLIP) features directly into the scene representation. However, the instance grouping mechanisms of these methods either require a predefined number of instances or suffer from noise in their bottom-up grouping strategies. Furthermore, the reliance on CLIP restricts semantic understanding to simple noun phrases, preventing complex spatial reasoning and referential expression grounding. We present GaussDet, a method that circumvents the need for dense CLIP features by leveraging discrete, open-vocabulary 2D object detectors with referring expression capabilities. We learn instance features for individual Gaussians to decompose the scene into 3D instance groups. By rendering these groups and aggregating semantic votes from multi-view 2D detections, we generate a robust View-Aggregated Semantic Label Distribution (VASD) for each 3D instance. This view-aggregation strategy acts as a strong regularizer, attenuating spurious labels caused by low-quality instance grouping. Our approach enables a straightforward, zero-shot extension from simple language queries to complex referential grounding. Extensive evaluations across two key tasks -- open-vocabulary segmentation (LeRF-OVS, ScanNet) and referring expression grounding (Ref-LeRF) -- demonstrate that GaussDet achieves consistent improvements over existing methods. Most notably, we achieve a substantial 16.7% mIoU improvement in referential grounding within a strict zero-shot setting.
Recent advancements in 3D Gaussian Splatting (3DGS) have enabled language-guided scene understanding. However, existing Referring 3D Gaussian Splatting (R3DGS) methods are fundamentally restricted to single-target queries. To reflect the ambiguity of real-world instructions, we introduce the Generalized Referring 3D Gaussian Splatting Segmentation (GR3DGS) task, which requires dynamically segmenting an arbitrary number of targets (0, 1, or $N$). To facilitate comprehensive evaluation of this new task, we construct two new benchmarks: GR-LERF and GR-ScanNet. Crucially, existing R3DGS paradigms exhibit fundamental technical bottlenecks that severely limit their performance on the GR3DGS task: they lack intrinsic 3D point-level understanding by operating merely on 2D rendered pixels, and they incur prohibitive computational overhead by requiring per-scene optimization to embed heavy semantic features. To dismantle these bottlenecks, we propose ZeroSplat, a novel training-free and zero-feature framework. ZeroSplat lifts 2D Vision-Language Model (VLM) priors into 3D space through robust multi-view geometric constraints. This strategy enables intrinsic point-level understanding without incurring any additional feature storage. Extensive experiments demonstrate that ZeroSplat significantly outperforms state-of-the-art methods across generalized and single-target scenarios while maintaining exceptional efficiency. Project Page: https://inkmind-ai.github.io/ZeroSplat
Jiayu Ding, Meilu Song, Xiaoyi Zhang et al.· 0 citations
3D Gaussian splatting (3DGS) has recently emerged as an efficient and scalable method for high-fidelity 3D scene reconstruction, representation, and real-time rendering. In addition to geometric reconstruction, increasing research attention focuses on enriching 3D Gaussian primitives with semantic information, which can be related to an arbitrary application or domain, as well as common knowledge. However, the existing approaches to semantic 3DGS significantly differ in how semantics are represented, learned, and accessed, which makes systematic analysis difficult. This paper provides a review on semantic extensions to 3DGS. We introduce a unified multi-axis taxonomy that enables us to classify the available methods in terms of five complementary categories: semantic vocabulary space, representation form, functional role, knowledge source, and query mechanism. The analysis reveals key design trade-offs related to the flexibility, efficiency, and semantic expressiveness of the methods. Furthermore, we review datasets, benchmarks, and evaluation metrics used in the field, indicating the diversity of approaches and the lack of common evaluation frameworks. Based on this analysis, we also identify open challenges and possible future research directions. The presented survey is relevant to advances in games and immersive technologies, where semantically enriched real-time 3D representations are essential for interactive environments, AR/VR, and intelligent scene understanding. The systematic analysis presented in this survey aims to facilitate a deeper understanding of semantic 3DGS and support the development of more general, efficient, and task-aware 3D scene understanding systems.
Camera-reconstructed 3D scenes often require offline visual cleanup before inspection, presentation, or reuse as renderable virtual-scene assets. Representative applications include removing temporary furniture, parked vehicles, equipment, signage, and other distracting or obsolete objects from reconstructed indoor and outdoor environments. Such editing requires not only accurate target localization across viewpoints but also plausible recovery of the previously occluded background. Existing methods often depend on manually specified masks or category-restricted detectors, while projection-based pipelines independently inpaint multiple views and subsequently refine the 3D representation, potentially introducing cross-view appearance and geometry inconsistencies. We present OVR-GS (Open-Vocabulary Removal in Gaussian Splatting), an instruction-driven object-removal framework for pre-trained 3D Gaussian Splatting (3DGS) scenes. Given a free-form instruction, a language parser generates target-oriented queries and a textual background-completion condition. Grounding DINO and the Segment Anything Model (SAM) produce multi-view candidate masks, which are filtered using Contrastive Language–Image Pre-training (CLIP). The proposed Semantic-Aware Gaussian Selector (SAGS) aggregates rendering-contribution-weighted mask evidence, groups spatially coherent candidates, and identifies the target Gaussian subset through rendered-cluster semantic verification. After removal, new Gaussians are initialized from boundary-adjacent primitives and interior samples and optimized locally using Score Distillation Sampling (SDS), while the original background remains fixed. On IMFine, SPIn-NeRF, and Inpaint360GS, OVR-GS achieves peak signal-to-noise ratio (PSNR) values of 19.78, 17.82, and 24.62 dB and Fréchet inception distance (FID) values of 142.30, 148.60, and 34.80, respectively. The results demonstrate the effectiveness of localized Gaussian optimization for instruction-driven cleanup of reconstructed environments before visual inspection, presentation, or reuse as renderable virtual-scene assets.
Yongpeng Ding, Feng Ouyang, Jiawei Fan et al.· Italian National Conference...· 0 citations
Simultaneously reconstructing and understanding 3D environments is essential for embodied agents. Toward this goal, feed-forward semantic 3D Gaussian Splatting (3DGS) efficiently constructs semantic scene representations from sparse multi-view observations. However, existing methods lack explicit instance discrimination and mainly support category- or phrase-based semantic queries. To this end, we propose GroupForward, an instance-grouped feed-forward Gaussian splatting model that reconstructs geometry, appearance, instance structure, and semantics from sparse, unposed, and uncalibrated multi-view images. Unlike existing methods that attach high-dimensional semantic features to each Gaussian, GroupForward learns compact instance embeddings that group Gaussians into cross-view consistent 3D instances, reformulating feed-forward semantic 3DGS from per-Gaussian semantic feature rendering to instance-level semantic aggregation and propagation. Building on these instance groups, we further propose a Referential Scene Reasoning Framework (RSRF) for complex 3D referring segmentation. RSRF constructs an instance-grouped 3D scene graph and retrieves candidate instances for a given referring expression. A vision-language model then reasons over structured instance evidence and multi-view observations to identify the referred instance among the candidates. RSRF thereby extends language interaction from simple semantic querying to complex referential scene reasoning. Experiments on semantic reconstruction and referential reasoning demonstrate the effectiveness of our instance-grouped reconstruction and reasoning framework.
Qijian Tian, Zimeng Wu, Xuhong Wang et al.· 0 citations
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.
Xiaosheng Yan, Hefeng Wu, Yanghui Xu et al.· 0 citations
Unsupervised 3D instance segmentation of outdoor LiDAR scans has traditionally relied on handcrafted geometric priors such as density-based clustering, motion cues, or projected 2D detections. In this work, we investigate whether a frozen, self-supervised point transformer already contains the structural information required to isolate object instances without any handcrafted geometric prior. Using this transformer purely as a feature extractor, we probe its internal representations across the SemanticKITTI, nuScenes, and Waymo Perception datasets. Our analysis yields four core insights: (1) the instance signal concentrates in the attention queries and keys rather than in the values or final output features; (2) output features semantically collapse, merging adjacent same-class objects that the queries and keys keep distinct; (3) this instance signal is bimodal in depth, strongest at the shallowest and deepest encoder stages; and (4) this signal is driven predominantly by the rotary position encoding (RoPE), whose removal collapses its advantage. We put these findings into our method TokenGraph3D, a training-free segmenter that groups points via connected components on a key-similarity graph, using neither density-based clustering nor proximity priors. Under identical prior-free conditions, we substantially outperform output-feature baselines, making the emergent 3D instance structure visible.
Ted Lentsch, Santiago Montiel-Mar'in, Holger Caesar et al.· 0 citations