Pith. sign in

REVIEW 3 major objections 4 minor 76 references

Zero-shot 3D grounding can find objects no Object Lookup Table lists, so long as a chain of contextual cues leads to them.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · deepseek-v4-flash

2026-08-03 13:42 UTC pith:7Z4BZZSH

load-bearing objection Useful step for zero-shot 3DVG, but the OpenTarget benchmark is built the way the method reasons, so the headline open-world gap should be read with caution. the 3 major comments →

arxiv 2512.23020 v3 pith:7Z4BZZSH submitted 2025-12-28 cs.CV cs.AI

OpenGround: Planning-based Online Perception for Open-World 3D Visual Grounding

Wenyuan Huang , Zhenyu Zhang , Zhao Wang , Zhou Wei , Ting Huang , Fang Zhao , Jian Yang This is my paper
classification cs.CV cs.AI
keywords zero-shot 3D visual groundingopen-world groundingobject lookup tablevision-language modelcognitive task chainonline perceptionScanReferOpenTarget dataset
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

OpenGround argues that zero-shot 3D visual grounding fails not because vision-language models are weak, but because the pipeline is locked to a pre-defined Object Lookup Table (OLT) that decides in advance what can be found. To break this, it plans a cognitive task chain from the query—first grounding context objects, then the target—and, whenever an object is missing from the OLT, actively perceives it online by segmenting 2D views around previously grounded objects and lifting the masks into 3D to extend the table. On a new open-world benchmark, OpenTarget, built from fine-grained part-level objects, the method reaches 46.2% Acc@0.25 with no ground-truth list, far above zero-shot baselines that were given the ground-truth OLT. On the standard ScanRefer benchmark it reports 61.8% Acc@0.25, the best zero-shot number in its comparison. If the numbers hold, it is evidence that progressive, context-driven grounding plus online dictionary expansion can carry open-world 3D grounding without exhaustive pre-definition.

Core claim

The paper's central claim is that the hard part of open-world 3D grounding is not the final VLM call but the cognitive preparation before it. The Active Cognition-based Reasoning (ACR) module decomposes a query into an ordered task chain of context-to-target sub-goals; then, for each sub-goal whose object is absent from the OLT, the Active Cognition Enhancement (ACE) module selects viewpoints around objects already grounded, segments the novel object with an open-vocabulary 2D segmenter, lifts the masks to 3D, merges spatially overlapping masks, and appends new entries to a continuously updated OLT. Single-Step Grounding then reasons over annotated views that mark both previously grounded ob

What carries the argument

The load-bearing mechanism is the object lookup table (OLT) turned dynamic. In prior zero-shot work the OLT is a fixed dictionary of IDs, labels, and boxes that the VLM can reason over; OpenGround's ACE module rewrites the table mid-query. A Cognitive Task Chain Construction module prompts the VLM to parse the query into target plus relevant objects, retrieves candidate IDs by text similarity, and orders them into a sequence (validated against human-constructed chains by weighted edit distance). ACE then greedily selects up to V views that maximize coverage of previously grounded objects, runs an open-vocabulary 2D segmenter (GroundedSAM) on those views, lifts masks into 3D, merges them by I

Load-bearing premise

ACE only looks for new objects around objects already grounded, so if the next object named in the query is far from the current chain, it will never be observed and can never enter the object table.

What would settle it

Take the OpenTarget queries and split them by the distance between the first grounded context object and the target's decisive reference. If accuracy on the far-far subset (e.g., reference farther than 3 m from any earlier object) does not collapse toward random while near queries stay high, the spatial-proximity assumption is not the active constraint. A direct experiment would annotate each query with the minimal distance from the target to the nearest previously grounded object and plot accuracy against it.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • Open-world 3D grounding is feasible without an exhaustive object dictionary: the ACR+ACE pipeline grounds objects never listed, as long as they are spatially reachable from already grounded context.
  • The approach is a plug-in: replacing the single-step grounding backbone with prior methods (SeeGround, SeqVLM) still confers open-world ability, so the cognitive-chain idea transfers as a module.
  • Reasoning structure, not VLM scale, drives performance: a 32B VLM in the framework still beats zero-shot baselines given the ground-truth OLT, per the ablations.
  • Long, multi-step queries benefit the most: accuracy stays high for chain lengths 8–10 while flat baselines drop, suggesting decomposition pays off precisely where single-shot reasoning is weakest.
  • Without any initial OLT the method still outperforms baselines that are given ground-truth boxes, indicating the online perception loop is doing the grounding work rather than the initial detector.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The spatial-proximity premise means the method is a 'local' reasoner: if a query's decisive reference is far from the current chain, the ACE cameras never point there; a natural extension is a global BEV-guided region selection before local ACE, which the authors themselves outline as future work.
  • Task-chain ordering could be bootstrapped from repeated grounding attempts rather than a learned model: the weighted edit-distance agreement with humans suggests an active-learning loop that adapts chain construction to the VLM's failure modes.
  • The IoU-based mask merging and the dependence on an open-vocabulary segmenter make the ceiling set by 2D open-world segmentation; swapping in stronger segmenters (or open-vocabulary 3D segmentation, which the authors test) is a direct lever for accuracy.
  • Compatibility results imply the ACE module can be retrofitted to any OLT-based zero-shot 3DVG pipeline, which would let the field upgrade existing systems without changing their reasoning core.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

3 major / 4 minor

Summary. The paper proposes OpenGround, a zero-shot framework for 3D visual grounding that does not require the target object to appear in a predefined object lookup table (OLT). It introduces an Active Cognition-based Reasoning (ACR) module that (i) decomposes a query into a cognitive task chain of context-to-target sub-goals, (ii) actively extends the OLT by perceiving novel objects around previously grounded objects, and (iii) performs single-step VLM grounding with contextual annotations. The paper also introduces OpenTarget, a benchmark derived from ScanNet++ and Articulate3D with 7,724 VLM-generated and partially manually filtered query-object pairs, designed to test grounding of fine-grained/open-world objects. Experiments report 46.2%/34.2% Acc@0.25/0.50 on OpenTarget (a 17.6% improvement over baselines given ground-truth OLT), 61.8%/53.1% on ScanRefer (best among zero-shot methods in the comparison), and 61.7% on Nr3D.

Significance. If the claims hold, the contribution is meaningful: OpenGround offers a principled way to escape the fixed-OLT bottleneck of existing VLM-based zero-shot 3D grounding, and the OpenTarget benchmark addresses a real evaluation gap. The paper has several strengths: the task-chain decomposition is well motivated and compared against human-constructed orderings; the ACE module is evaluated in isolation and in combination with other single-step grounding methods (Fig. 8b), showing a degree of generality; the ablations cover OLT initialization, task-chain strategies, annotation design, VLM choice, perspective selection, and segmentation backbones. The central idea of progressive, context-driven perception is useful and likely to stimulate follow-up work. However, the main open-world evidence rests on a dataset whose construction closely mirrors the method's own assumptions, and the quantitative comparisons are single-run point estimates without error bars or significance tests. These issues affect the strength of the central claim but are addressable with additional evaluation and analysis.

major comments (3)
  1. [Sec. 3, Fig. 2, Table 3, Appendix Tab. 9] The OpenTarget benchmark appears to encode the same hierarchical, spatially proximal structure that OpenGround's ACR/ACE modules are explicitly designed to exploit. The annotation prompt in Tab. 9 supplies the label hierarchy (e.g., cabinet→drawer→handle) and the parent object's annotation as context reference, so the resulting descriptions naturally name the parent object and often locate it near the target. ACR constructs its task chain precisely to ground such parent/context objects first, and ACE restricts perception to their spatial neighborhood. The strong OpenTarget margin (46.2/34.2 vs. 28.6/20.4 for VLM-Grounder+GT) may therefore reflect compatibility with the benchmark-generation pipeline rather than a general open-world capability. The paper's Limitation admits the spatial-proximity assumption, but the dataset construction likely guarantees it for most queries. I request an ad
  2. [Sec. 5.2, Tables 1-4, Fig. 8] All quantitative results are single-run point estimates. No error bars, confidence intervals, or significance tests are reported. This is particularly important for the ablation claims, where several differences are small: Full vs. Relevance task chains differ by 1.6 points at Acc@0.50 (Tab. 4, rows 1 and 4), V=3 vs. V=4 differ by 0.52 points (Fig. 8a), and the claimed 17.6% improvement on OpenTarget could also be affected by sampling or VLM nondeterminism. At minimum, the authors should provide repeated runs (or bootstrap over queries) for the main OpenTarget result and for the key ablations, and state the variance.
  3. [Table 3 and Table 6] VLM-Grounder, the strongest OpenTarget baseline, is evaluated only on a randomly selected 300-sample subset, with no confidence interval or representativeness check. The main table reports VLM-Grounder+GT at 28.6/20.4, yet this number comes from 300 samples rather than the full 7,724, making the comparison to OpenGround's full-benchmark numbers difficult to interpret. Please evaluate VLM-Grounder on the full benchmark (with the same efficiency optimizations used elsewhere) or, if that is infeasible, provide stratified bootstrap intervals and show that the 300-sample subset has the same query-type distribution as the full set.
minor comments (4)
  1. [Sec. 6] In the Conclusion, 'SceneRefer' should be 'ScanRefer'.
  2. [Algorithm 2, line 26] Line 26 updates the candidate's observed region using P_Oi (previously grounded object) instead of P_cj (the candidate). This makes the pseudocode inconsistent with the text in Sec. 8.4.
  3. [Sec. 3 and Sec. 7.1] The text says 'manual filtering' ensures quality, while Sec. 7.1 describes a VLM-voting stage followed by human review. The number of human annotators, the inter-annotator agreement, and the fraction of queries rejected at each stage are not reported; this information is important for a new benchmark.
  4. [General] The paper does not state whether code, the OpenTarget benchmark, or the evaluation scripts will be released. For a benchmark paper and for reproducibility of the reported results, this should be clarified.

Circularity Check

0 steps flagged

No significant circularity: the central zero-shot grounding pipeline is a VLM-driven process evaluated on external benchmarks, and the OpenTarget alignment concern is benchmark design rather than an equivalence-by-construction.

full rationale

OpenGround's derivation is a zero-shot inference pipeline built from independent modules: task-chain construction is a VLM call (Eq. 4), ACE extends the OLT via 2D segmentation, lifting, and merging (Eqs. 5–8), and single-step grounding follows an existing paradigm with contextual annotations. None of these equations define the target in terms of the output, and no parameter is fitted to the evaluation set before prediction. The ScanRefer and Nr3D results are external benchmarks, so the state-of-the-art claims there rest on independent evidence. The OpenTarget benchmark is author-constructed and its annotation prompt supplies the label hierarchy and parent-object context reference (Table 9), which does align with the hierarchical task-chain structure ACR consumes; that is a legitimate benchmark-design concern about favorability, but it is not a circular derivation because the method's outputs are not restatements of the dataset construction and the baselines receive GT OLT without the task chain. The paper explicitly acknowledges its spatial-proximity assumption as a Limitation, and the supplementary discusses extensions, which is a scope caveat rather than a circular step. No self-citation is load-bearing: the only author-overlapping reference, 3D-R1 [15], appears in related work as a supervised baseline, not as justification for the open-world claim. Overall, the claimed predictions do not reduce to their inputs by construction.

Axiom & Free-Parameter Ledger

6 free parameters · 5 axioms · 0 invented entities

The method contributes no new physical entities. Its open-world capability is carried by free parameters (thresholds, view count) and by strong assumptions about VLM ability, segmentation coverage, spatial locality of references, and the validity of VLM-generated benchmark queries. The Mask3D-OLT-exclusion assumption is unquantified.

free parameters (6)
  • tau_cand (candidate retrieval threshold) = 0.9
    Cosine-similarity threshold in Eq. 3 that decides which OLT entries are candidates; fixed following prior work, but directly controls the candidate pool size and downstream VLM choices.
  • tau_iou (mask merging threshold) = 0.5
    IoU threshold used in ACE to merge lifted 2D masks; taken from Open3DIS, but determines whether novel objects are over-segmented or under-segmented before entering the OLT.
  • V (max ACE views) = 3
    Max number of observation perspectives selected in Algorithm 1; chosen after ablation on OpenTarget where V=4 gives +0.52% Acc@0.50; tuning on the target benchmark means the headline result uses a validation-fitted hyperparameter.
  • alpha (soft margin in single-step view selection) = not reported
    Lexicographic trade-off in Algorithm 2 between candidate coverage and context coverage; no numeric value or sensitivity analysis is given.
  • GroundedSAM text/box thresholds = 0.4
    GroundingDINO thresholds for open-vocabulary detection in ACE; affects which novel objects are perceived and added to the OLT.
  • tau (fallback coverage threshold) = not reported
    Threshold in Algorithm 1's fallback strategy deciding whether a view adds enough new 3D points; no value given.
axioms (5)
  • domain assumption Spatial proximity of relevant objects
    ACE only selects views around previously grounded objects (Sec. 4.2, Algorithm 1); the Limitation section admits the method 'assumes spatially proximal relevant objects.' If a decisive reference object is far away, it is never observed and cannot be grounded.
  • domain assumption VLM reliability for parsing, ordering, and reasoning
    Objects Parsing (Eq. 2), Task Chain Construction (Eq. 4), condition extraction, and final object selection all depend on VLM outputs with no fallback; if the VLM misparses the query or orders the chain wrongly, all later steps inherit the error.
  • domain assumption Open-vocabulary segmentation adequacy
    ACE extends the OLT only with masks from SEG(I_k, L_{T_t}); Table 8 shows Acc@0.50 drops from 34.2 (SAM-H) to 26.7 (SAM-B), confirming that the central open-world capability is bounded by segmentation quality.
  • ad hoc to paper OpenTarget queries are valid open-world queries
    Queries are generated by Qwen3-VL from labeled images/hierarchies and filtered by multiple VLMs plus manual review; they are not collected from human speakers, so OpenTarget measures grounding of VLM-style descriptions, which may align with the method's VLM-driven task chains.
  • domain assumption Mask3D OLT excludes the OpenTarget targets
    The open-world claim assumes part-level objects (drawer handles, hinges) are absent from the initial Mask3D OLT; the paper does not quantify how many of the 7,724 target segments are accidentally segmented by Mask3D.

pith-pipeline@v1.3.0-alltime-deepseek · 24775 in / 13924 out tokens · 123446 ms · 2026-08-03T13:42:35.974648+00:00 · methodology

0 comments
read the original abstract

3D visual grounding aims to locate objects based on natural language descriptions in 3D scenes. Existing supervised methods are limited by generalization and recent zero-shot methods typically rely on a predefined Object Lookup Table (OLT) to query Visual Language Models (VLMs) for reasoning about object locations via a single step grounding, which limits the applications in scenarios with undefined targets and complex queries. To address these problems, we present OpenGround, a novel zero-shot framework for open-world 3D visual grounding that remains compatible with recent zero-shot methods. OpenGround integrates Task-Chain Planning to decompose a query into a plan of context-to-target sub-goals for progressive grounding, and Context-Guided Perception to perceive novel objects online under context guidance from the task chain. We also propose a new dataset named OpenTarget, which contains over 7000 object-description pairs to mimic open-world evaluation. Extensive experiments demonstrate that OpenGround achieves competitive performance on Nr3D, state-of-the-art on ScanRefer, and delivers a substantial 17.6\% improvement on OpenTarget. Project Page at https://why-102.github.io/openground.io/.

Figures

Figures reproduced from arXiv: 2512.23020 by Fang Zhao, Jian Yang, Ting Huang, Wenyuan Huang, Zhao Wang, Zhenyu Zhang, Zhou Wei.

Figure 1
Figure 1. Figure 1: Comparative overview of previous zero-shot meth￾ods’ paradigm and ours. Previous paradigm is single-step grounding based on predefined OLT , unable to ground undefined objects. In contrast, our method introduces Active Cognition￾based Reasoning to the paradigm, which enhances cognition be￾fore each grounding step and grounds target object progressively. to targets of different granularities, from coarse ob… view at source ↗
Figure 2
Figure 2. Figure 2: Data Collection Pipeline. The pipeline generates discriminative object descriptions via three annotation stages and two￾stage verification. It leverages hierarchical labels (e.g., cabinet→drawer→handle), selects target–distractor views, and employs VLMs for context-aware descriptions using parent annotations. Quality is ensured via VLM voting and manual refinement. Qwen-VL, and InternVL series, show strong… view at source ↗
Figure 3
Figure 3. Figure 3: Overview of the OpenGround framework. The core of our framework is the Active Cognition-based Reasoning (ACR) module. First, the ACR invokes Cognitive Task Chain Construction module to obtain a sequential task chain to guide step-by-step grounding. Next, the ACR module progresses along the task chain to ground objects progressively. For objects not present in the OLT , it activates the Active Cognition Enh… view at source ↗
Figure 4
Figure 4. Figure 4: (a) illustrates three sequential steps (Objects Parsing, [PITH_FULL_IMAGE:figures/full_fig_p004_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Illustration of perspective selection strategies in ACE. Task Chain Construction. Human cognitive planning for object grounding balances semantic relevance and task dif￾ficulty. To emulate this, we prompt the VLM with parsed object labels and their candidate set sizes (seen as difficulty in Nr3D [1]) to generate the ordered indices of L as the Cognitive Task Chain T : T = VLM(Q,L, {∥Cj∥ | j ∈ [1, n + 1]}),… view at source ↗
Figure 6
Figure 6. Figure 6: Qualitative Comparisons on OpenTarget. (a) compares our method with previous open-source SoTA method, SeqVLM [25] with predefined OLT . They fail to ground objects out-of-OLT . (b) compares our method with previous methods equipped with ground-truth OLT . these baselines either overlook key objects (e.g., toy, door) or are distracted by numerous relevant objects, leading to failures [PITH_FULL_IMAGE:figur… view at source ↗
Figure 8
Figure 8. Figure 8: Ablation Study: (a) evaluates the impact of differ [PITH_FULL_IMAGE:figures/full_fig_p008_8.png] view at source ↗
Figure 7
Figure 7. Figure 7: Task Chain Construction Strategy Results Compar￾ison. The orange mask area is the objects skipped by the jump strategy. Jump strategy incorrectly identifies the object (orange) because it skips the key object, rack. Full strategy correctly iden￾tifies the object (teal), considering the rack (purple). 5.3. Ablation Study To comprehensively validate each module in OpenGround, we conduct extensive ablation st… view at source ↗
Figure 9
Figure 9. Figure 9: Dataset Statistics. (a) shows the word count distribu￾tion, indicating balanced conciseness. (b) shows the frequency of sentence categories including attributes (fundamental object properties), parts (object components), intra-class spatial rela￾tions (same-category object positions), inter-class spatial relations (different-category object positions), ordinal numbers (sequential order), superlatives (extr… view at source ↗
Figure 10
Figure 10. Figure 10: Edit Distances with Human Inconsistency. [PITH_FULL_IMAGE:figures/full_fig_p011_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Annotation Visualization. The query is “Locate the drawer directly beneath the machine-like object marked with ‘Ulti￾maker’. The drawer is a part of the cabinet under the machine-like object.” (a) is the origin image without any annotations. (b) anno￾tates all objects mentioned (“drawer”, “cabinet”, “object”) in the query. (c) only annotates all candidates for “drawer”. (d) anno￾tates previously grounded … view at source ↗
Figure 12
Figure 12. Figure 12: Comparison across Task Chain Length. “VLMGrounder∗ +GT” is evaluated only on a subset of 300 sam￾ples (due to its extremely high computational cost) and lacks val￾ues at L = 1 and L = 10 because such chain lengths do not appear in the sampled subset. biguity and consistent performance degradation. For fair￾ness, we also evaluate our method under the same setting by replacing our initial OLT with the full … view at source ↗
Figure 13
Figure 13. Figure 13: Visualization of OpenGround on the OpenTarget benchmark. Correct predictions are shown in teal and incorrect predictions in orange. Key linguistic cues used for grounding are underlined [PITH_FULL_IMAGE:figures/full_fig_p016_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Visualization of OpenGround on Application in ScanNet++ [63]. The target object (Ground Truth) is even out of OpenTarget, actually open-world object [PITH_FULL_IMAGE:figures/full_fig_p017_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: Visualization of OpenGround on Application in Torondo3D [48]. The target object (Ground Truth) is also out of predefined categories in Toronto3D, actually open-world object [PITH_FULL_IMAGE:figures/full_fig_p017_15.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

76 extracted references · 19 linked inside Pith

  1. [1]

    Referit3d: Neural listeners for fine-grained 3d object identification in real-world scenes

    Panos Achlioptas, Ahmed Abdelreheem, Fei Xia, Mohamed Elhoseiny, and Leonidas Guibas. Referit3d: Neural listeners for fine-grained 3d object identification in real-world scenes. In European conference on computer vision, pages 422–440. Springer, 2020. 1, 2, 3, 5, 6, 7

    2020

  2. [2]

    Qwen-vl: A versatile vision-language model for un- derstanding, localization, text reading, and beyond

    Jinze Bai, Shuai Bai, Shusheng Yang, Shijie Wang, Sinan Tan, Peng Wang, Junyang Lin, Chang Zhou, and Jingren Zhou. Qwen-vl: A versatile vision-language model for un- derstanding, localization, text reading, and beyond. arXiv preprint arXiv:2308.12966, 2023. 2

    Pith/arXiv arXiv 2023

  3. [3]

    Qwen2.5-vl technical report

    Shuai Bai, Keqin Chen, Xuejing Liu, Jialin Wang, Wenbin Ge, Sibo Song, Kai Dang, Peng Wang, Shijie Wang, Jun Tang, Humen Zhong, Yuanzhi Zhu, Mingkun Yang, Zhao- hai Li, Jianqiang Wan, Pengfei Wang, Wei Ding, Zheren Fu, Yiheng Xu, Jiabo Ye, Xi Zhang, Tianbao Xie, Zesen Cheng, Hang Zhang, Zhibo Yang, Haiyang Xu, and Jun- yang Lin. Qwen2.5-vl technical repor...

    Pith/arXiv arXiv 2025

  4. [4]

    Arkitscenes: A diverse real-world dataset for 3d indoor scene understanding using mobile rgb-d data

    Gilad Baruch, Zhuoyuan Chen, Afshin Dehghan, Tal Dimry, Yuri Feigin, Peter Fu, Thomas Gebauer, Brandon Joffe, Daniel Kurz, Arik Schwartz, et al. Arkitscenes: A diverse real-world dataset for 3d indoor scene understanding using mobile rgb-d data. arXiv preprint arXiv:2111.08897, 2021. 1

    Pith/arXiv arXiv 2021

  5. [5]

    V olcengine, 2025

    ByteDance. V olcengine, 2025. Accessed: 2025-02-18. 7

    2025

  6. [6]

    Scanrefer: 3d object localization in rgb-d scans using natural language

    Dave Zhenyu Chen, Angel X Chang, and Matthias Nießner. Scanrefer: 3d object localization in rgb-d scans using natural language. In European conference on computer vision, pages 202–221. Springer, 2020. 1, 2, 3, 6, 7

    2020

  7. [7]

    Clip2scene: Towards label-efficient 3d scene understanding by clip

    Runnan Chen, Youquan Liu, Lingdong Kong, Xinge Zhu, Yuexin Ma, Yikang Li, Yuenan Hou, Yu Qiao, and Wen- ping Wang. Clip2scene: Towards label-efficient 3d scene understanding by clip. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 7020–7030, 2023. 1

    2023

  8. [8]

    Internvl: Scaling up vision founda- tion models and aligning for generic visual-linguistic tasks

    Zhe Chen, Jiannan Wu, Wenhai Wang, Weijie Su, Guo Chen, Sen Xing, Muyan Zhong, Qinglong Zhang, Xizhou Zhu, Lewei Lu, et al. Internvl: Scaling up vision founda- tion models and aligning for generic visual-linguistic tasks. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 24185–24198, 2024. 2

    2024

  9. [9]

    Scannet: Richly-annotated 3d reconstructions of indoor scenes

    Angela Dai, Angel X Chang, Manolis Savva, Maciej Hal- ber, Thomas Funkhouser, and Matthias Nießner. Scannet: Richly-annotated 3d reconstructions of indoor scenes. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 5828–5839, 2017. 1, 6

    2017

  10. [10]

    Text-guided sparse voxel prun- ing for efficient 3d visual grounding

    Wenxuan Guo, Xiuwei Xu, Ziwei Wang, Jianjiang Feng, Jie Zhou, and Jiwen Lu. Text-guided sparse voxel prun- ing for efficient 3d visual grounding. In Proceedings of the Computer Vision and Pattern Recognition Conference, pages 3666–3675, 2025. 1, 2, 7

    2025

  11. [11]

    Holistic understanding of 3d scenes as universal scene description

    Anna-Maria Halacheva, Yang Miao, Jan-Nico Zaech, Xi Wang, Luc Van Gool, and Danda Pani Paudel. Holistic understanding of 3d scenes as universal scene description. arXiv preprint arXiv:2412.01398, 2024. 2, 3

    Pith/arXiv arXiv 2024

  12. [12]

    Visual language maps for robot navigation

    Chenguang Huang, Oier Mees, Andy Zeng, and Wolfram Burgard. Visual language maps for robot navigation. arXiv preprint arXiv:2210.05714, 2022. 1

    Pith/arXiv arXiv 2022

  13. [13]

    Viewsrd: 3d vi- sual grounding via structured multi-view decomposition

    Ronggang Huang, Haoxin Yang, Yan Cai, Xuemiao Xu, Huaidong Zhang, and Shengfeng He. Viewsrd: 3d vi- sual grounding via structured multi-view decomposition. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 9726–9736, 2025. 1, 2, 6, 7

    2025

  14. [14]

    Real-time object detection meets dinov3

    Shihua Huang, Yongjie Hou, Longfei Liu, Xuanlong Yu, and Xi Shen. Real-time object detection meets dinov3. arXiv,

  15. [15]

    3d-r1: Enhancing reasoning in 3d vlms for unified scene understanding

    Ting Huang, Zeyu Zhang, and Hao Tang. 3d-r1: Enhancing reasoning in 3d vlms for unified scene understanding. arXiv preprint arXiv:2507.23478, 2025. 1, 2, 7

    Pith/arXiv arXiv 2025

  16. [16]

    Assister: As- sistive navigation via conditional instruction generation

    Zanming Huang, Zhongkai Shangguan, Jimuyang Zhang, Gilad Bar, Matthew Boyd, and Eshed Ohn-Bar. Assister: As- sistive navigation via conditional instruction generation. In European Conference on Computer Vision, pages 271–289. Springer, 2022. 1

    2022

  17. [17]

    Bottom up top down detection transform- ers for language grounding in images and point clouds

    Ayush Jain, Nikolaos Gkanatsios, Ishita Mediratta, and Kate- rina Fragkiadaki. Bottom up top down detection transform- ers for language grounding in images and point clouds. In European Conference on Computer Vision, pages 417–433. Springer, 2022. 2

    2022

  18. [18]

    Pointgroup: Dual-set point grouping for 3d instance segmentation

    Li Jiang, Hengshuang Zhao, Shaoshuai Shi, Shu Liu, Chi- Wing Fu, and Jiaya Jia. Pointgroup: Dual-set point grouping for 3d instance segmentation. In Proceedings of the IEEE/CVF conference on computer vision and Pattern recognition, pages 4867–4876, 2020. 2, 3

    2020

  19. [19]

    Spazer: Spatial- semantic progressive reasoning agent for zero-shot 3d visual grounding, 2025

    Zhao Jin, Rong-Cheng Tu, Jingyi Liao, Wenhao Sun, Xiao Luo, Shunyu Liu, and Dacheng Tao. Spazer: Spatial- semantic progressive reasoning agent for zero-shot 3d visual grounding, 2025. 2, 6, 7, 5

    2025

  20. [20]

    Berg, Wan-Yen Lo, Piotr Doll ´ar, and Ross Girshick

    Alexander Kirillov, Eric Mintun, Nikhila Ravi, Hanzi Mao, Chloe Rolland, Laura Gustafson, Tete Xiao, Spencer White- head, Alexander C. Berg, Wan-Yen Lo, Piotr Doll ´ar, and Ross Girshick. Segment anything. arXiv:2304.02643, 2023. 3

    Pith/arXiv arXiv 2023

  21. [21]

    Robo3d: Towards robust and reliable 3d perception against corruptions

    Lingdong Kong, Youquan Liu, Xin Li, Runnan Chen, Wen- wei Zhang, Jiawei Ren, Liang Pan, Kai Chen, and Ziwei Liu. Robo3d: Towards robust and reliable 3d perception against corruptions. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 19994–20006, 2023. 1

    2023

  22. [22]

    Binary coors capable or ‘correcting dele- tions, insertions, and reversals

    VI Lcvenshtcin. Binary coors capable or ‘correcting dele- tions, insertions, and reversals. In Soviet physics-doklady,

  23. [23]

    Cityanchor: City-scale 3d visual grounding with multi-modality llms

    Jinpeng Li, Haiping Wang, Yuan Liu, Zhiyang Dou, Yuexin Ma, Sibei Yang, Yuan Li, Wenping Wang, Zhen Dong, Bisheng Yang, et al. Cityanchor: City-scale 3d visual grounding with multi-modality llms. In The Thirteenth International Conference on Learning Representations. 1, 2

  24. [24]

    Seeground: See and ground for zero-shot open-vocabulary 3d visual grounding

    Rong Li, Shijie Li, Lingdong Kong, Xulei Yang, and Jun- wei Liang. Seeground: See and ground for zero-shot open-vocabulary 3d visual grounding. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2025. 2, 6, 7, 9, 4, 5

    2025

  25. [25]

    Seqvlm: Proposal- guided multi-view sequences reasoning via vlm for zero- shot 3d visual grounding

    Jiawen Lin, Shiran Bian, Yihang Zhu, Wenbin Tan, Yachao Zhang, Yuan Xie, and Yanyun Qu. Seqvlm: Proposal- guided multi-view sequences reasoning via vlm for zero- shot 3d visual grounding. In Proceedings of the 33rd ACM International Conference on Multimedia, page 3094–3103, New York, NY , USA, 2025. Association for Computing Ma- chinery. 2, 6, 7, 9, 4, 5

    2025

  26. [26]

    Groundflow: A plug-in module for temporal reasoning on 3d point cloud sequential grounding

    Zijun Lin, Shuting He, Cheston Tan, and Bihan Wen. Groundflow: A plug-in module for temporal reasoning on 3d point cloud sequential grounding. arXiv preprint arXiv:2506.21188, 2025. 1, 2

    arXiv 2025

  27. [27]

    Improved baselines with visual instruction tuning, 2023

    Haotian Liu, Chunyuan Li, Yuheng Li, and Yong Jae Lee. Improved baselines with visual instruction tuning, 2023. 2

    2023

  28. [28]

    Visual instruction tuning, 2023

    Haotian Liu, Chunyuan Li, Qingyang Wu, and Yong Jae Lee. Visual instruction tuning, 2023

    2023

  29. [29]

    Llava-next: Im- proved reasoning, ocr, and world knowledge, 2024

    Haotian Liu, Chunyuan Li, Yuheng Li, Bo Li, Yuanhan Zhang, Sheng Shen, and Yong Jae Lee. Llava-next: Im- proved reasoning, ocr, and world knowledge, 2024. 2

    2024

  30. [30]

    Grounding dino: Marrying dino with grounded pre-training for open-set object detection

    Shilong Liu, Zhaoyang Zeng, Tianhe Ren, Feng Li, Hao Zhang, Jie Yang, Chunyuan Li, Jianwei Yang, Hang Su, Jun Zhu, et al. Grounding dino: Marrying dino with grounded pre-training for open-set object detection. arXiv preprint arXiv:2303.05499, 2023. 3, 4

    Pith/arXiv arXiv 2023

  31. [31]

    Aligning cyber space with physical world: A comprehensive survey on embodied ai

    Yang Liu, Weixing Chen, Yongjie Bai, Xiaodan Liang, Guanbin Li, Wen Gao, and Liang Lin. Aligning cyber space with physical world: A comprehensive survey on embodied ai. IEEE/ASME Transactions on Mechatronics, 2025. 1

    2025

  32. [32]

    Group-free 3d object detection via transformers

    Ze Liu, Zheng Zhang, Yue Cao, Han Hu, and Xin Tong. Group-free 3d object detection via transformers. In Proceedings of the IEEE/CVF international conference on computer vision, pages 2949–2958, 2021. 2

    2021

  33. [33]

    Rea- songrounder: Lvlm-guided hierarchical feature splatting for open-vocabulary 3d visual grounding and reasoning

    Zhenyang Liu, Yikai Wang, Sixiao Zheng, Tongying Pan, Longfei Liang, Yanwei Fu, and Xiangyang Xue. Rea- songrounder: Lvlm-guided hierarchical feature splatting for open-vocabulary 3d visual grounding and reasoning. In CVPR, pages 3718–3727. Computer Vision Foundation / IEEE, 2025. 2

    2025

  34. [34]

    An examination of the compositionality of large generative vision-language mod- els

    Teli Ma, Rong Li, and Junwei Liang. An examination of the compositionality of large generative vision-language mod- els. arXiv preprint arXiv:2308.10509, 2023. 1

    Pith/arXiv arXiv 2023

  35. [35]

    A guided tour to approximate string matching

    Gonzalo Navarro. A guided tour to approximate string matching. ACM Comput. Surv., 33(1):31–88, 2001. 5

    2001

  36. [36]

    Open3dis: Open-vocabulary 3d instance segmentation with 2d mask guidance

    Phuc Nguyen, Tuan Duc Ngo, Evangelos Kalogerakis, Chuang Gan, Anh Tran, Cuong Pham, and Khoi Nguyen. Open3dis: Open-vocabulary 3d instance segmentation with 2d mask guidance. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 4018–4028, 2024. 5, 6

    2024

  37. [37]

    Gpt-4 technical report, 2024

    OpenAI. Gpt-4 technical report, 2024. 2, 6, 7

    2024

  38. [38]

    Gpt4scene: Understand 3d scenes from videos with vision-language models

    Zhangyang Qi, Zhixiong Zhang, Ye Fang, Jiaqi Wang, and Hengshuang Zhao. Gpt4scene: Understand 3d scenes from videos with vision-language models. arXiv preprint arXiv:2501.01428, 2025. 1, 2, 6, 7, 4

    Pith/arXiv arXiv 2025

  39. [39]

    Multi-branch collabora- tive learning network for 3d visual grounding

    Zhipeng Qian, Yiwei Ma, Zhekai Lin, Jiayi Ji, Xiawu Zheng, Xiaoshuai Sun, and Rongrong Ji. Multi-branch collabora- tive learning network for 3d visual grounding. In European Conference on Computer Vision, pages 381–398. Springer,

  40. [40]

    Learning transferable visual models from natural language supervision, 2021

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, Gretchen Krueger, and Ilya Sutskever. Learning transferable visual models from natural language supervision, 2021. 6

    2021

  41. [41]

    Sam 2: Segment anything in images and videos

    Nikhila Ravi, Valentin Gabeur, Yuan-Ting Hu, Ronghang Hu, Chaitanya Ryali, Tengyu Ma, Haitham Khedr, Roman R¨adle, Chloe Rolland, Laura Gustafson, Eric Mintun, Junt- ing Pan, Kalyan Vasudev Alwala, Nicolas Carion, Chao- Yuan Wu, Ross Girshick, Piotr Doll´ar, and Christoph Feicht- enhofer. Sam 2: Segment anything in images and videos. arXiv preprint arXiv:...

    Pith/arXiv arXiv 2024

  42. [42]

    Grounding dino 1.5: Ad- vance the ”edge” of open-set object detection, 2024

    Tianhe Ren, Qing Jiang, Shilong Liu, Zhaoyang Zeng, Wen- long Liu, Han Gao, Hongjie Huang, Zhengyu Ma, Xiaoke Jiang, Yihao Chen, Yuda Xiong, Hao Zhang, Feng Li, Peijun Tang, Kent Yu, and Lei Zhang. Grounding dino 1.5: Ad- vance the ”edge” of open-set object detection, 2024. 4

    2024

  43. [43]

    Grounded sam: Assembling open-world models for diverse visual tasks,

    Tianhe Ren, Shilong Liu, Ailing Zeng, Jing Lin, Kun- chang Li, He Cao, Jiayu Chen, Xinyu Huang, Yukang Chen, Feng Yan, Zhaoyang Zeng, Hao Zhang, Feng Li, Jie Yang, Hongyang Li, Qing Jiang, and Lei Zhang. Grounded sam: Assembling open-world models for diverse visual tasks,

  44. [44]

    Mask3d: Mask trans- former for 3d semantic instance segmentation.arXiv preprint arXiv:2210.03105, 2022

    Jonas Schult, Francis Engelmann, Alexander Hermans, Or Litany, Siyu Tang, and Bastian Leibe. Mask3d: Mask trans- former for 3d semantic instance segmentation.arXiv preprint arXiv:2210.03105, 2022. 2, 3, 6, 7, 8, 4

    Pith/arXiv arXiv 2022

  45. [45]

    Zhihong Shao, Peiyi Wang, Qihao Zhu, Runxin Xu, Junx- iao Song, Mingchuan Zhang, Y .K. Li, Y . Wu, and Daya Guo. Deepseekmath: Pushing the limits of mathematical reason- ing in open language models, 2024. 2

    2024

  46. [46]

    Chain of semantics programming in 3d gaussian splatting representation for 3d vision grounding

    Jiaxin Shi, Mingyue Xiang, Hao Sun, Yixuan Huang, and Zhi Weng. Chain of semantics programming in 3d gaussian splatting representation for 3d vision grounding. In CVPR, pages 24560–24569. Computer Vision Foundation / IEEE,

  47. [47]

    Scanverse: An extended reality authoring platform for in- dustrial digital twins

    Serkan Solmaz, Lode Jorissen, and Vasilios Zogopoulos. Scanverse: An extended reality authoring platform for in- dustrial digital twins. Procedia CIRP, 136:624–629, 2025. 1

    2025

  48. [48]

    Toronto-3D: A large-scale mobile lidar dataset for semantic segmenta- tion of urban roadways

    Weikai Tan, Nannan Qin, Lingfei Ma, Ying Li, Jing Du, Guorong Cai, Ke Yang, and Jonathan Li. Toronto-3D: A large-scale mobile lidar dataset for semantic segmenta- tion of urban roadways. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops, pages 202–203, 2020. 8

    2020

  49. [49]

    Qwen3 technical report, 2025

    Qwen Team. Qwen3 technical report, 2025. 2, 3, 8, 1

    2025

  50. [50]

    Step-3 is large yet affordable: Model-system co-design for cost-effective decoding, 2025

    StepFun Team. Step-3 is large yet affordable: Model-system co-design for cost-effective decoding, 2025. 8, 1

    2025

  51. [51]

    Glm-4.5v and glm-4.1v-thinking: Towards versatile multi- modal reasoning with scalable reinforcement learning, 2025

    V Team, Wenyi Hong, Wenmeng Yu, Xiaotao Gu, Guo Wang, Guobing Gan, Haomiao Tang, Jiale Cheng, Ji Qi, Junhui Ji, Lihang Pan, Shuaiqi Duan, Weihan Wang, Yan Wang, Yean Cheng, Zehai He, Zhe Su, Zhen Yang, Ziyang Pan, Aohan Zeng, Baoxu Wang, Bin Chen, Boyan Shi, Changyu Pang, Chenhui Zhang, Da Yin, Fan Yang, Guoqing Chen, Jiazheng Xu, Jiale Zhu, Jiali Chen, J...

    2025

  52. [52]

    Four ways to improve verbo-visual fusion for dense 3d visual grounding

    Ozan Unal, Christos Sakaridis, Suman Saha, and Luc Van Gool. Four ways to improve verbo-visual fusion for dense 3d visual grounding. In European Conference on Computer Vision, pages 196–213. Springer, 2024. 2

    2024

  53. [53]

    Four ways to improve verbo-visual fusion for dense 3d visual grounding

    Ozan Unal, Christos Sakaridis, Suman Saha, and Luc Van Gool. Four ways to improve verbo-visual fusion for dense 3d visual grounding. In European Conference on Computer Vision (ECCV), 2024. 1

    2024

  54. [54]

    Softgroup for 3d instance segmentation on point clouds

    Thang Vu, Kookhoi Kim, Tung M Luu, Thanh Nguyen, and Chang D Yoo. Softgroup for 3d instance segmentation on point clouds. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 2708–2717,

  55. [55]

    Qwen2-vl: Enhancing vision-language model’s perception of the world at any resolution

    Peng Wang, Shuai Bai, Sinan Tan, Shijie Wang, Zhihao Fan, Jinze Bai, Keqin Chen, Xuejing Liu, Jialin Wang, Wenbin Ge, Yang Fan, Kai Dang, Mengfei Du, Xuancheng Ren, Rui Men, Dayiheng Liu, Chang Zhou, Jingren Zhou, and Jun- yang Lin. Qwen2-vl: Enhancing vision-language model’s perception of the world at any resolution. arXiv preprint arXiv:2409.12191, 2024. 2, 7

    Pith/arXiv arXiv 2024

  56. [56]

    Internvl3.5: Advancing open-source multimodal models in versatility, reasoning, and efficiency

    Weiyun Wang, Zhangwei Gao, Lixin Gu, Hengjun Pu, Long Cui, Xingguang Wei, Zhaoyang Liu, Linglin Jing, Sheng- long Ye, Jie Shao, et al. Internvl3.5: Advancing open-source multimodal models in versatility, reasoning, and efficiency. arXiv preprint arXiv:2508.18265, 2025. 2, 3

    Pith/arXiv arXiv 2025

  57. [57]

    Gˆ 3-lq: Marry- ing hyperbolic alignment with explicit semantic-geometric modeling for 3d visual grounding

    Yuan Wang, Yali Li, and Shengjin Wang. Gˆ 3-lq: Marry- ing hyperbolic alignment with explicit semantic-geometric modeling for 3d visual grounding. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 13917–13926, 2024. 2

    2024

  58. [58]

    Deepseek-vl2: Mixture-of-experts vision-language models for advanced multimodal understanding, 2024

    Zhiyu Wu, Xiaokang Chen, Zizheng Pan, Xingchao Liu, Wen Liu, Damai Dai, Huazuo Gao, Yiyang Ma, Chengyue Wu, Bingxuan Wang, Zhenda Xie, Yu Wu, Kai Hu, Jiawei Wang, Yaofeng Sun, Yukun Li, Yishi Piao, Kang Guan, Aixin Liu, Xin Xie, Yuxiang You, Kai Dong, Xingkai Yu, Haowei Zhang, Liang Zhao, Yisong Wang, and Chong Ruan. Deepseek-vl2: Mixture-of-experts visio...

    2024

  59. [59]

    Vlm-grounder: A vlm agent for zero-shot 3d visual grounding

    Runsen Xu, Zhiwei Huang, Tai Wang, Yilun Chen, Jiang- miao Pang, and Dahua Lin. Vlm-grounder: A vlm agent for zero-shot 3d visual grounding. In CoRL, 2024. 2, 6, 7, 8, 4, 5

    2024

  60. [60]

    Pointllm: Empowering large language models to understand point clouds

    Runsen Xu, Xiaolong Wang, Tai Wang, Yilun Chen, Jiang- miao Pang, and Dahua Lin. Pointllm: Empowering large language models to understand point clouds. In European Conference on Computer Vision, pages 131–147. Springer,

  61. [61]

    Llm- grounder: Open-vocabulary 3d visual grounding with large language model as an agent

    Jianing Yang, Xuweiyi Chen, Shengyi Qian, Nikhil Madaan, Madhavan Iyengar, David F Fouhey, and Joyce Chai. Llm- grounder: Open-vocabulary 3d visual grounding with large language model as an agent. In 2024 IEEE International Conference on Robotics and Automation (ICRA), pages 7694–7701. IEEE, 2024. 2

    2024

  62. [62]

    Gupta, Rilyn Han, Li Fei-Fei, and Saining Xie

    Jihan Yang, Shusheng Yang, Anjali W. Gupta, Rilyn Han, Li Fei-Fei, and Saining Xie. Thinking in Space: How Multi- modal Large Language Models See, Remember and Recall Spaces. arXiv preprint arXiv:2412.14171, 2024. 2

    Pith/arXiv arXiv 2024

  63. [63]

    Scannet++: A high-fidelity dataset of 3d in- door scenes

    Chandan Yeshwanth, Yueh-Cheng Liu, Matthias Nießner, and Angela Dai. Scannet++: A high-fidelity dataset of 3d in- door scenes. In Proceedings of the International Conference on Computer Vision (ICCV), 2023. 2, 3, 8

    2023

  64. [64]

    Visual programming for zero-shot open-vocabulary 3d visual grounding

    Zhihao Yuan, Jinke Ren, Chun-Mei Feng, Hengshuang Zhao, Shuguang Cui, and Zhen Li. Visual programming for zero-shot open-vocabulary 3d visual grounding. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 20623–20633, 2024. 2, 6, 7

    2024

  65. [65]

    Sort3d: Spatial object-centric reasoning toolbox for zero-shot 3d grounding using large language models

    Nader Zantout, Haochen Zhang, Pujith Kachana, Jinkai Qiu, Ji Zhang, and Wenshan Wang. Sort3d: Spatial object-centric reasoning toolbox for zero-shot 3d grounding using large language models. CoRR, abs/2504.18684, 2025. 2

    Pith/arXiv arXiv 2025

  66. [66]

    Freeq-graph: Free-form querying with semantic con- sistent scene graph for 3d scene understanding, 2025

    Chenlu Zhan, Yufei Zhang, Gaoang Wang, and Hongwei Wang. Freeq-graph: Free-form querying with semantic con- sistent scene graph for 3d scene understanding, 2025. 2

    2025

  67. [67]

    Multi3drefer: Grounding text description to multiple 3d objects

    Yiming Zhang, ZeMing Gong, and Angel X Chang. Multi3drefer: Grounding text description to multiple 3d objects. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 15225–15236, 2023. 1, 2

    2023

  68. [68]

    Prototype correlation matching and class- relation reasoning for few-shot medical image seg- mentation

    Yumin Zhang, Hongliu Li, Yajun Gao, Haoran Duan, Yawen Huang, and Yefeng Zheng. Prototype correlation matching and class- relation reasoning for few-shot medical image seg- mentation. IEEE Transactions on Medical Imaging, 43(11): 4041–4054, 2024. 1

    2024

  69. [69]

    Vision-and-language navigation today and tomorrow: A survey in the era of foundation models

    Yue Zhang, Ziqiao Ma, Jialu Li, Yanyuan Qiao, Zun Wang, Joyce Chai, Qi Wu, Mohit Bansal, and Parisa Kordjamshidi. Vision-and-language navigation today and tomorrow: A survey in the era of foundation models. arXiv preprint arXiv:2407.07035, 2024. 1

    Pith/arXiv arXiv 2024

  70. [70]

    3dvg- transformer: Relation modeling for visual grounding on point clouds

    Lichen Zhao, Daigang Cai, Lu Sheng, and Dong Xu. 3dvg- transformer: Relation modeling for visual grounding on point clouds. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 2928–2937, 2021. 2

    2021

  71. [71]

    Densegrounding: Improving dense language-vision seman- tics for ego-centric 3d visual grounding

    Henry Zheng, Hao Shi, Qihang Peng, Yong Xien Chng, Rui Huang, Yepeng Weng, Zhongchao Shi, and Gao Huang. Densegrounding: Improving dense language-vision seman- tics for ego-centric 3d visual grounding. arXiv preprint arXiv:2505.04965, 2025. 1, 2

    Pith/arXiv arXiv 2025

  72. [72]

    Point-SAM: Promptable 3d segmentation model for point clouds

    Yuchen Zhou, Jiayuan Gu, Tung Yen Chiang, Fanbo Xiang, and Hao Su. Point-SAM: Promptable 3d segmentation model for point clouds. In The Thirteenth International Conference on Learning Representations, 2025. 3

    2025

  73. [73]

    Llava-3d: A simple yet effective pathway to empowering lmms with 3d-awareness

    Chenming Zhu, Tai Wang, Wenwei Zhang, Jiangmiao Pang, and Xihui Liu. Llava-3d: A simple yet effective pathway to empowering lmms with 3d-awareness. arXiv preprint arXiv:2409.18125, 2024. 2

    Pith/arXiv arXiv 2024

  74. [74]

    Struct2d: A perception-guided framework for spatial reasoning in large multimodal models

    Fangrui Zhu, Hanhui Wang, Yiming Xie, Jing Gu, Tianye Ding, Jianwei Yang, and Huaizu Jiang. Struct2d: A perception-guided framework for spatial reasoning in large multimodal models. CoRR, abs/2506.04220, 2025. 2

    arXiv 2025

  75. [75]

    3d-vista: Pre-trained transformer for 3d vision and text alignment

    Ziyu Zhu, Xiaojian Ma, Yixin Chen, Zhidong Deng, Siyuan Huang, and Qing Li. 3d-vista: Pre-trained transformer for 3d vision and text alignment. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 2911– 2921, 2023. 2

    2023

  76. [76]

    Move to understand a 3d scene: Bridg- ing visual grounding and exploration for efficient and versa- tile embodied navigation

    Ziyu Zhu, Xilin Wang, Yixuan Li, Zhuofan Zhang, Xiaojian Ma, Yixin Chen, Baoxiong Jia, Wei Liang, Qian Yu, Zhi- dong Deng, et al. Move to understand a 3d scene: Bridg- ing visual grounding and exploration for efficient and versa- tile embodied navigation. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 8120– 8132, 2025. 1

    2025