Pith. sign in

REVIEW 3 major objections 5 minor 61 references

OmniMap: A General Mapping Framework Integrating Optics, Geometry, and Semantics

T0 review · 3 major / 5 minor · reviewed 2026-08-04 · deepseek-v4-flash

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

Pith's one-line read One online RGB-D framework can simultaneously deliver photorealistic rendering, dense geometry, and open-vocabulary semantics in real time.

desk verdict A well-engineered integration of 3DGS and voxel mapping with strong benchmark results, but the rendering claim is weakened by an unstated evaluation protocol for the per-keyframe camera parameters. read the letter →

arxiv 2509.07500 v1 pith:WPRDKFX5 submitted 2025-09-09 cs.RO

Yinan Deng , Yufeng Yue , Jianyu Dou , Jingyu Zhao , Jiahui Wang , Yujie Tang , Yi Yang , Mengyin Fu This is my paper
classification cs.RO
keywords onlinemapping3DGaussianSplattingTSDFvoxelfusionopen-vocabularysemanticszero-shotsegmentationRGB-Dreconstructioninstanceroboticperception
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

OmniMap is an online RGB-D mapping framework designed to show that one system can simultaneously provide photorealistic rendering, dense geometric reconstruction, and open-vocabulary instance understanding while running in real time. The paper argues that prior methods force a choice among these: voxel maps are stable but coarse, Gaussian Splatting renders well but does not attach semantics online, and open-vocabulary maps lack density. Its central claim is that a tightly coupled hybrid—probabilistic voxels that seed and anchor Gaussian primitives—resolves the trade-off. On the Replica and ScanNet benchmarks the framework reports higher rendering fidelity, mesh accuracy, and zero-shot semantic segmentation than the compared methods, at 5.55 FPS with a 14.2 MB model. These results are obtained with externally supplied ground-truth poses in static scenes, so the online claim rests on the quality of the external position stream.

What carries the argument

The load-bearing object is the hybrid 3DGS–voxel map. Voxels from TSDF-Fusion are the stable scaffold: they hold probabilistic instance-ID tuples and an embedding codebook, provide the keyframe-selection signal through the ratio of unregistered voxels, and decide where to place new Gaussians. The Gaussian field is the fine-grained surface: it renders RGB, depth, and normals and carries photometric detail beyond voxel resolution. The two are coupled by initializing Gaussians only at newly allocated voxels, and by a four-parameter differentiable camera model that corrects per-frame blur and exposure during optimization.

What would settle it

Perturb ground-truth ScanNet poses with drift typical of a real SLAM trajectory and stream the sequence through OmniMap; if PSNR or mesh F-score collapses below the compared baseline under that noise, the practical online-mapping claim fails. Alternatively, move a person through a static indoor scene and test whether the probabilistic instance tuples fragment or merge the moving object.

Watch

Extended reading notes

Core claim

One online framework can hold the three scene attributes prior work separates—photorealistic appearance, dense geometry, open-vocabulary semantics—if the representations are coupled the right way. Voxels store probabilistic instance-ID tuples with Dirichlet pseudo-counts, giving semantic fusion an explicit uncertainty model; new Gaussians are initialized only from newly allocated voxels, so growth follows observed structure. A four-parameter differentiable camera model approximates blur and exposure during optimization, and RGB, depth, and normal losses supervise the field. The paper positions this as the first general online mapping framework to combine all three. Reported results with grou

Load-bearing premise

The framework does not track itself: it consumes external camera poses, and the benchmark scenes are static, so all reported map quality is conditioned on a noise-free pose stream and a non-dynamic world.

Editorial extensions

If this is right

  • A robot with an RGB-D camera and any external pose source can maintain a map that is simultaneously renderable, meshable, and queryable by natural-language object descriptions.
  • Because Gaussian growth is driven by newly allocated voxels, model size is tied to observed structure rather than frame count; the reported 6.2 MB model at coarser voxel resolution makes the accuracy-efficiency trade-off explicit.
  • Downstream tasks such as scene question answering, object editing, pick-and-place, and navigation can read from the same unified map instead of separate representations.
  • If the central claim holds, the perceived trade-off between online 3DGS mapping and open-vocabulary mapping is not fundamental; both can be maintained simultaneously.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The four optimized camera parameters could serve as a per-frame quality signal: large translation or brightness weights would flag motion-blurred or badly exposed frames for re-capture before they contaminate the map.
  • Because the framework accepts poses from any external source, the next natural stress test is a trajectory with realistic SLAM drift; the paper's results bracket only the perfect-pose regime.
  • The Dirichlet pseudo-counts on each voxel provide an uncertainty estimate the paper does not exploit; this could drive next-best-view selection or tell a robot when to re-observe an instance.
  • The modular front end implies that swapping the detection and segmentation models should change semantic accuracy and frame rate nearly independently, making the semantic pipeline tunable without touching the reconstruction core.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. OmniMap proposes an online RGB-D mapping framework that couples a 3D Gaussian Splatting (3DGS) field with a probabilistic voxel map to jointly represent appearance, geometry, and open-vocabulary semantics. The method introduces: (i) a hybrid representation in which newly allocated TSDF voxels initialize Gaussian primitives; (ii) a per-keyframe four-parameter camera model to compensate for motion blur and exposure; (iii) normal supervision from rendered depth; and (iv) an instance-level probabilistic fusion formulated as MLE/MAP over Dirichlet-Categorical updates with an embedding codebook. Experiments on Replica and ScanNet compare against 3DGS-based SLAM and open-vocabulary mapping baselines, reporting state-of-the-art PSNR/SSIM/LPIPS, mesh F-score, zero-shot semantic mIoU, 5.55 FPS, and 14.2 MB model size. The paper also demonstrates downstream applications: scene Q&A, interactive editing, manipulation, and navigation.

Significance. If the empirical claims hold, the paper makes a substantial contribution: it demonstrates a single online pipeline that simultaneously maintains photorealistic rendering, accurate geometry, and open-vocabulary instance understanding, with real-time operation and a compact model. The benchmark coverage is unusually broad for a mapping paper: eight 3DGS baselines on rendering/reconstruction, seven semantic baselines, ablations, parameter-sensitivity plots, and real-robot demos. The probabilistic instance-fusion formulation is a principled alternative to heuristic IoU association, and the voxel-guided Gaussian initialization is a clean mechanism for incremental densification. However, the central claims rest on the optical evaluation protocol: the paper does not explicitly state that evaluated frames are held out from the per-keyframe camera-parameter optimization, and no error bars or multi-seed statistics are reported. The framework also relies on externally provided poses and is limited to static scenes, as acknowledged in the Conclusion. These issues affect the strength of the 'first general online mapping framework' and 'SOTA' claims, but they appear fixable with additional clari

major comments (3)
  1. [Sec. VI-D and Sec. VII-B, Eq. (29)-(33), Tables II-III] The optical evaluation protocol is underspecified in a load-bearing way. The four camera parameters [ω_raw, ω_trans, x_trans, y_trans] are optimized per keyframe with the same RGB and SSIM losses (Eqs. 30-31) that are later used to report rendering quality. The text says 'rendered novel views using camera trajectories P' but does not state whether the evaluated frames are disjoint from the keyframes used during optimization, nor what camera parameters are used at evaluation time. If the evaluated frames are the same keyframes whose parameters were fitted, the camera model can absorb per-frame brightness and translation residuals, inflating PSNR/SSIM/LPIPS without improving true novel-view synthesis. The Fig. 6 'validation' is also only a correlation on the optimized frames. Please specify the train/test split, report held-out-frame metrics with camera parameters frozen or handled by a co
  2. [Sec. VII-A, Tables II-VII] All benchmark numbers are single-run point estimates. The pipeline uses stochastic 3DGS optimization and random selection of 19 keyframes per input frame (Sec. VII-A), so the reported differences—especially the smaller margins over HI-SLAM2 and GS-ICP-SLAM—may not be statistically robust. The paper repeatedly claims OmniMap 'consistently outperforms' and is 'superior' across metrics; supporting this requires mean±std over at least 3 seeds (or equivalent significance testing) for the headline tables. Without this, the SOTA claim is not yet quantitatively grounded.
  3. [Sec. VII-A and Sec. IX] The framework does not estimate camera poses; all experiments use ground-truth poses, and the Conclusion explicitly states that mapping quality depends on external positioning systems and that dynamic scenes are not handled. This is an honest limitation, but it directly bounds the 'general online mapping framework' claim in the Abstract and Sec. I. The claimed online performance is only demonstrated for externally localized, static environments. To support the broader claim, please either add experiments with realistic pose noise (or an integrated tracker) and dynamic-object sequences, or qualify the abstract and contribution statements accordingly.
minor comments (5)
  1. [Eq. (21)] The Gaussian exponent should use the inverse covariance Σ_i^{-1}; the current Σ_i^T is incorrect as written.
  2. [Eq. (27) and Fig. 6] The text states ω_raw and ω_trans are in [0,1], but Fig. 6 reports optimized values such as 1.408 and -8.820. Please clarify the actual parameter bounds or state that they are unconstrained post-initialization.
  3. [Sec. VII-D] The sentence 'Tab. V and Tab. VI summarize the qualitative results' should say 'quantitative results'; the figures provide the qualitative results.
  4. [Sec. VII-A] The keyframe discrimination thresholds—the unregistered-region ratio τ_t and the maximum number N of consecutive non-keyframes—are never given numerical values. Please report them in the implementation details.
  5. [Sec. V-C] The Dirichlet closed-form expectation in Eq. (16) cites reference [14]; a standard Bayesian statistics reference would be more appropriate for the conjugate prior update.

Circularity Check

1 steps flagged · score 3.0 of 10

Camera-model 'validation' in Fig. 6 is a fitting artifact, but the main SOTA claims rest on external benchmarks and ablations.

  1. fitted input called prediction [Sec. VI-C Eq. (29), Sec. VI-D Eqs. (30)-(31), Fig. 6]
    "The numbers under images correspond to the optimized parameter set [ωraw, ωtrans, xtrans, ytrans]. Quantitative analysis reveals: (1) high-brightness images show elevated ωraw+ωtrans (blue bar) relative to low-brightness conditions, and (2) blurred images exhibit consistently higher |xtrans|+|ytrans| (green bar) compared to sharp counterparts. These correlations validate OmniMap’s effectiveness in camera modeling."

    The four camera parameters are optimized per keyframe against the very same observed frames via Eq. (29) supervised by Lrgb (Eq. 30). Therefore ωraw+ωtrans and |xtrans|+|ytrans| are free variables chosen to minimize the photometric residual for each frame. A bright frame forces the optimizer to raise ωraw+ωtrans, and a blurred frame with motion forces a nonzero translation mixture; the reported correlation is a property of the fitted parameters, not an independent validation. No held-out blur/exposure prediction is made, so Fig. 6's 'validate' is circular. The central rendering claim remains partly supported by the external ablation and benchmarks.

full rationale

The paper's core derivations are not circular: rendering, mesh, and semantic metrics are external benchmarks (Replica, ScanNet), baseline methods are reproduced, and ablations (Table VIII) show the proposed components matter. The only genuine circular step is Fig. 6, where per-frame camera parameters optimized to minimize photometric loss are then used to 'validate' the camera model by showing correlations with brightness and blur; this correlation is a direct consequence of the fitting procedure, not a predictive test. Self-citations, including the OpenVox-derived probabilistic voxel formulation, are background and not load-bearing because the Dirichlet update and association equations are derived in the paper. The evaluation protocol does not explicitly state that rendered 'novel views' are held out from camera-parameter optimization, but that is an ambiguity/risk rather than a demonstrated reduction, so it is not counted as circularity. Overall score 3 reflects this partial, non-central circularity while the main SOTA claims retain independent support.

Assumptions & free parameters 6 free parameters · 6 assumptions · 0 invented entities

The central claims rest on established but unreproduced components (3DGS, TSDF-Fusion, YOLO-World, TAP, SBERT) plus several hand-set thresholds and per-frame fitted parameters. The most fragile inputs are the external-pose assumption and the adequacy of the two-image blur model.

free parameters (6)
  • Per-keyframe camera model parameters (omega_raw, omega_trans, x_trans, y_trans) = Initialized [0.5, 0.5, 0, 0]; fitted per keyframe, examples in Fig. 6
    Fitted against the RGB rendering loss (Eqs. 29-31) to absorb motion blur and exposure; used to claim optical robustness.
  • Instance fusion threshold xi = Not reported
    Hand-chosen threshold controlling when a new instance is created (Alg. 1); Fig. 11 shows it affects segmentation granularity.
  • Voxel resolution = 0.03 m default; 0.06 m in ablation (6.2 MB model)
    Chosen to balance precision and memory; determines Gaussian primitive density and mesh detail.
  • Keyframe insertion thresholds (unregistered ratio tau_t and N consecutive frames) = Not reported
    Determines when frames enter the keyframe buffer; no numeric values given.
  • Fusion weights between geometric and embedding similarity = Not reported
    Eq. (9) and (10) are combined by 'weighted fusion' but the weights are never specified.
  • Loss balancing weights for L_rgb, L_ssim, L_depth, L_normal = Not reported
    The combined loss is not written explicitly; the relative weights are unstated and likely tuned.
assumptions (6)
  • domain assumption 3DGS differentiable rendering (Eqs. 21-23) correctly models the scene appearance and depth.
    Adopted from Kerbl et al.; not rederived, and online incremental optimization may not satisfy the same convergence properties.
  • domain assumption TSDF-Fusion provides a stable voxel grid with reliable color and surface information for Gaussian initialization.
    Used as the structural backbone (Sec. III, V); errors in TSDF integration propagate to Gaussian placement.
  • domain assumption YOLO-World and TAP produce masks and captions accurate enough that probabilistic fusion can recover from their noise.
    Front-end segmentation quality is not quantified; claims of robustness rest on the fusion absorbing this noise.
  • domain assumption Scene is static and camera poses are externally accurate.
    Acknowledged in Conclusion: no tracking module, cannot handle dynamic scenes; benchmarks supply ground-truth poses.
  • ad hoc to paper Motion blur and exposure can be represented as a weighted combination of a sharp render and a globally shifted sharp render (Eq. 29).
    This simplification ignores rotation, depth-dependent parallax, and spatially varying blur; it is introduced for efficiency.
  • domain assumption Dirichlet-Categorical conjugate updating (Eqs. 12-16) is a valid sensor model for instance segmentation observations.
    Borrowed from counting sensor models; its validity for instance masks is asserted without independent evidence.

how reviews work

0 comments
Cite this review

Pith. "Pith review of OmniMap: A General Mapping Framework Integrating Optics, Geometry, and Semantics." pith.science (2026). https://pith.science/paper/WPRDKFX5

@misc{pith2026250907500,
  author       = {Pith},
  title        = {Pith review of: OmniMap: A General Mapping Framework Integrating Optics, Geometry, and Semantics},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/WPRDKFX5}},
  note         = {Machine review of arXiv:2509.07500}
}
read the original abstract

Robotic systems demand accurate and comprehensive 3D environment perception, requiring simultaneous capture of photo-realistic appearance (optical), precise layout shape (geometric), and open-vocabulary scene understanding (semantic). Existing methods typically achieve only partial fulfillment of these requirements while exhibiting optical blurring, geometric irregularities, and semantic ambiguities. To address these challenges, we propose OmniMap. Overall, OmniMap represents the first online mapping framework that simultaneously captures optical, geometric, and semantic scene attributes while maintaining real-time performance and model compactness. At the architectural level, OmniMap employs a tightly coupled 3DGS-Voxel hybrid representation that combines fine-grained modeling with structural stability. At the implementation level, OmniMap identifies key challenges across different modalities and introduces several innovations: adaptive camera modeling for motion blur and exposure compensation, hybrid incremental representation with normal constraints, and probabilistic fusion for robust instance-level understanding. Extensive experiments show OmniMap's superior performance in rendering fidelity, geometric accuracy, and zero-shot semantic segmentation compared to state-of-the-art methods across diverse scenes. The framework's versatility is further evidenced through a variety of downstream applications, including multi-domain scene Q&A, interactive editing, perception-guided manipulation, and map-assisted navigation.

Figures

Figures reproduced from arXiv: 2509.07500 by the authors.

Figure 1
Figure 1. We introduce OmniMap, a general online mapping framework integrating optics, geometry, and semantics. OmniMap incrementally maintains an open-vocabulary instance-level voxel representation and a 3DGS (3D Gaussian Splatting) representation, from which color and geometric meshes are derived. OmniMap supports multi-modal rendering (RGB / depth / normal / instance), and achieves state-of-the-art performance in rendering… view at source ↗
Figure 2
Figure 2. OmniMap facilitates a holistic representation that integrates optics, geometry, and semantics, and provides five key advantages. incremental framework (replaceable with other voxel-based mapping methods), strategically initializing Gaussians only within newly assigned voxels to maintain model compactness and geometric stability. Additionally, OmniMap incorporates normal-based supervision directly derived from the re… view at source ↗
Figure 3
Figure 3. The system framework of OmniMap consists of three main modules. The 2D Language Embeddings Extractor sequentially combines multiple open-set models for detection, segmentation, captioning, and embedding extraction of objects. The Probabilistic Voxel Reconstructor incrementally integrates per-frame instance masks and embeddings into 3D space while maintaining voxel-aligned probabilistic instance tuples and a global e… view at source ↗
Figures from the paper (11 more)
Figure 4
Figure 4. Figure 4: The map is represented by probabilistic voxels V, with each voxel v j storing a probabilistic instance ID tuple θ j . In practice, we apply the TSDF-Fusion algorithm [57] integrated with the Open3D framework. To optimize memory usage, the voxel map is composed of activ…
Figure 5
Figure 5. Figure 5: OmniMap initializes the Gaussian by distinguishing the newly added voxels, providing an efficient yet effective coverage of newly observed regions. After initialization, only minimal Gaussian additions are required for subsequent updates. Gaussian primitives are depth-…
Figure 6
Figure 6. Figure 6: Schematic of camera modeling. The numbers under images correspond to the optimized parameter set [ωraw, ωtrans, xtrans, ytrans]. Quantitative analysis reveals: (1) high-brightness images show elevated ωraw + ωtrans (blue bar) relative to low-brightness conditions, and …
Figure 7
Figure 7. Figure 7: OmniMap’s rendering produces smooth and geometrically consistent normal images despite being trained with noisy supervi￾sion, demonstrating significantly better reconstruction quality than approaches without Lnormal. RGB Loss: For keyframe with time t ′ , we employ Eq.…
Figure 8
Figure 8. Figure 8: The rendering comparisons across diverse scenes demonstrate OmniMap’s superior visual quality, achieving both higher overall fidelity and more accurate texture details than all baseline methods. Qualitative results for representative scenes are shown in [PITH_FULL_IMA…
Figure 9
Figure 9. Figure 9: Mesh reconstruction quality comparison. OmniMap generates notably smoother surfaces and more geometrically precise reconstruc￾tions, particularly excelling in both continuous surface representation and fine structural detail recovery. highlights OmniMap’s advantage in …
Figure 10
Figure 10. Figure 10: Zero-shot semantic segmentation results. OmniMap’s probabilistic voxel representation demonstrates robustness to sensor noise while maintaining accurate object segmentation and scene understanding capabilities. iMap’s ability to maintain high mapping performance witho…
Figure 11
Figure 11. Figure 11: The impact of two key parameters on mapping performance: voxel resolution and instance fusion threshold. The bottom row shows the final instance-level maps for different fusion thresholds. The mural is actually a black-and￾white framed photograph. It depicts a boat re…
Figure 12
Figure 12. Figure 12: The process of 2D Scene Q&A. OmniMap enables scene understanding and question answering through object retrieval and image rendering capabilities, augmented by VQA techniques. fusion results at varying thresholds, highlighting differences in challenging cases such as …
Figure 14
Figure 14. Figure 14: Scene editing demonstration. Through object querying and direct manipulation of Gaussian primitives, our framework enables flexible scene modification with multi-view rendering capabilities. quality, offering an efficient approach to address critical data augmentation…
Figure 15
Figure 15. Figure 15: OmniMap supports robotic arm operations by reconstructing workspace optics, geometry, and semantics through end-effector￾mounted camera scanning, enabling task completion. fidelity, with its real-time rendered output closely matching the actual RGB feed. This achievem…

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

61 extracted references · 56 canonical work pages

  1. [1]

    Octomap: An efficient probabilistic 3d mapping framework based on octrees,

    A. Hornung, K. M. Wurm, M. Bennewitz, et al., “Octomap: An efficient probabilistic 3d mapping framework based on octrees,”Autonomous robots, vol. 34, no. 3, pp. 189–206, 2013

    work page 2013

  2. [2]

    V oxblox: Incremental 3d euclidean signed distance fields for on-board mav planning,

    H. Oleynikova, Z. Taylor, M. Fehr, et al., “V oxblox: Incremental 3d euclidean signed distance fields for on-board mav planning,” in2017 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pp. 1366–1373. IEEE, 2017

    work page 2017

  3. [3]

    Kinectfusion: Real-time dense surface mapping and tracking,

    R. A. Newcombe, S. Izadi, O. Hilliges, et al., “Kinectfusion: Real-time dense surface mapping and tracking,” in2011 10th IEEE international symposium on mixed and augmented reality, pp. 127–136. Ieee, 2011

    work page 2011

  4. [4]

    Bundlefusion: Real-time glob- ally consistent 3d reconstruction using on-the-fly surface reintegration,

    A. Dai, M. Nießner, M. Zollh ¨ofer, et al., “Bundlefusion: Real-time glob- ally consistent 3d reconstruction using on-the-fly surface reintegration,” ACM Transactions on Graphics (ToG), vol. 36, no. 4, p. 1, 2017

    work page 2017

  5. [5]

    Nerf: Representing scenes as neural radiance fields for view synthesis,

    B. Mildenhall, P. P. Srinivasan, M. Tancik, et al., “Nerf: Representing scenes as neural radiance fields for view synthesis,”Communications of the ACM, vol. 65, no. 1, pp. 99–106, 2021. 19

    work page 2021

  6. [6]

    3d gaussian splatting for real-time radiance field rendering,

    B. Kerbl, G. Kopanas, T. Leimk ¨uhler, et al., “3d gaussian splatting for real-time radiance field rendering,”ACM Transactions on Graphics, vol. 42, no. 4, pp. 1–14, 2023

    work page 2023

  7. [7]

    Conceptgraphs: Open- vocabulary 3d scene graphs for perception and planning,

    Q. Gu, A. Kuwajerwala, S. Morin, et al., “Conceptgraphs: Open- vocabulary 3d scene graphs for perception and planning,” in2016 IEEE international conference on robotics and automation (ICRA). IEEE, 2024

    work page 2024

  8. [8]

    Hierarchical Open-V ocabulary 3D Scene Graphs for Language-Grounded Robot Navigation,

    A. Werby, C. Huang, M. B ¨uchner, et al., “Hierarchical Open-V ocabulary 3D Scene Graphs for Language-Grounded Robot Navigation,” inPro- ceedings of Robotics: Science and Systems, Delft, Netherlands, Jul. 2024

    work page 2024

Show all 61 references
  1. [9]

    Lerf: Language embedded radi- ance fields,

    J. Kerr, C. M. Kim, K. Goldberg, et al., “Lerf: Language embedded radi- ance fields,” inProceedings of the IEEE/CVF International Conference on Computer Vision, pp. 19 729–19 739, 2023

    2023

  2. [10]

    Langsplat: 3d language gaussian splatting,

    M. Qin, W. Li, J. Zhou, et al., “Langsplat: 3d language gaussian splatting,” inIEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2024

    2024

  3. [11]

    Occupancy grid mapping without ray-casting for high-resolution lidar sensors,

    Y . Cai, F. Kong, Y . Ren, et al., “Occupancy grid mapping without ray-casting for high-resolution lidar sensors,”IEEE Transactions on Robotics, vol. 40, pp. 172–192, 2023

    2023

  4. [12]

    Continuous occupancy mapping in dynamic environments using particles,

    G. Chen, W. Dong, P. Peng, et al., “Continuous occupancy mapping in dynamic environments using particles,”IEEE Transactions on Robotics, vol. 40, pp. 64–84, 2023

    2023

  5. [13]

    Uni-fusion: Universal continuous mapping,

    Y . Yuan and A. N ¨uchter, “Uni-fusion: Universal continuous mapping,” IEEE Transactions on Robotics, vol. 40, pp. 1373–1392, 2024

    2024

  6. [14]

    See-csom: Sharp-edged and efficient continuous semantic occupancy mapping for mobile robots,

    Y . Deng, M. Wang, Y . Yang, et al., “See-csom: Sharp-edged and efficient continuous semantic occupancy mapping for mobile robots,” IEEE Transactions on Industrial Electronics, 2023

    2023

  7. [15]

    Semantic octree mapping and shannon mutual information computation for robot exploration,

    A. Asgharivaskasi and N. Atanasov, “Semantic octree mapping and shannon mutual information computation for robot exploration,”IEEE Transactions on Robotics, vol. 39, no. 3, pp. 1910–1928, 2023

    1910

  8. [16]

    Hd-ccsom: Hierarchical and dense collaborative continuous semantic occupancy mapping through label diffusion,

    Y . Deng, M. Wang, Y . Yang, et al., “Hd-ccsom: Hierarchical and dense collaborative continuous semantic occupancy mapping through label diffusion,” in2022 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pp. 2417–2422. IEEE, 2022

    2022

  9. [17]

    Gaussian-slam: Photo-realistic dense slam with gaussian splatting,

    V . Yugay, Y . Li, T. Gevers, et al., “Gaussian-slam: Photo-realistic dense slam with gaussian splatting,”arXiv preprint arXiv:2312.10070, 2023

    2023 arXiv

  10. [18]

    Splatam: Splat track & map 3d gaussians for dense rgb-d slam,

    N. Keetha, J. Karhade, K. M. Jatavallabhula, et al., “Splatam: Splat track & map 3d gaussians for dense rgb-d slam,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 21 357–21 366, 2024

    2024

  11. [19]

    Gaussian splatting slam,

    H. Matsuki, R. Murai, P. H. Kelly, et al., “Gaussian splatting slam,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 18 039–18 048, 2024

    2024

  12. [20]

    Conceptfusion: Open- set multimodal 3d mapping,

    K. Jatavallabhula, A. Kuwajerwala, Q. Gu, et al., “Conceptfusion: Open- set multimodal 3d mapping,”Robotics: Science and Systems (RSS), 2023

    2023

  13. [21]

    Opengraph: Open-vocabulary hier- archical 3d graph representation in large-scale outdoor environments,

    Y . Deng, J. Wang, J. Zhao, et al., “Opengraph: Open-vocabulary hier- archical 3d graph representation in large-scale outdoor environments,” IEEE Robotics and Automation Letters, vol. 9, no. 10, pp. 8402–8409, 2024

    2024

  14. [22]

    Openin: Open-vocabulary instance- oriented navigation in dynamic domestic environments,

    Y . Tang, M. Wang, Y . Deng, et al., “Openin: Open-vocabulary instance- oriented navigation in dynamic domestic environments,”IEEE Robotics and Automation Letters, no. 99, pp. 1–8, 2025

    2025

  15. [23]

    Gsfusion: Online rgb-d mapping where gaussian splatting meets tsdf fusion,

    J. Wei and S. Leutenegger, “Gsfusion: Online rgb-d mapping where gaussian splatting meets tsdf fusion,”IEEE Robotics and Automation Letters, 2024

    2024

  16. [24]

    Nice-slam: Neural implicit scalable encoding for slam,

    Z. Zhu, S. Peng, V . Larsson, et al., “Nice-slam: Neural implicit scalable encoding for slam,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp. 12 786–12 796, 2022

    2022

  17. [25]

    Activegs: Active scene reconstruction using gaussian splatting,

    L. Jin, X. Zhong, Y . Pan, et al., “Activegs: Active scene reconstruction using gaussian splatting,”IEEE Robotics and Automation Letters, 2025

    2025

  18. [26]

    Deblur-nerf: Neural radiance fields from blurry images,

    L. Ma, X. Li, J. Liao, et al., “Deblur-nerf: Neural radiance fields from blurry images,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp. 12 861–12 870, 2022

    2022

  19. [27]

    Deblurring 3d gaussian splatting,

    B. Lee, H. Lee, X. Sun, et al., “Deblurring 3d gaussian splatting,” in European Conference on Computer Vision, pp. 127–143. Springer, 2024

    2024

  20. [28]

    Bad-gaussians: Bundle adjusted deblur gaussian splatting,

    L. Zhao, P. Wang, and P. Liu, “Bad-gaussians: Bundle adjusted deblur gaussian splatting,” inEuropean Conference on Computer Vision, pp. 233–250. Springer, 2024

    2024

  21. [29]

    Open-fusion: Real-time open- vocabulary 3d mapping and queryable scene representation,

    K. Yamazaki, T. Hanyu, K. V o, et al., “Open-fusion: Real-time open- vocabulary 3d mapping and queryable scene representation,” in2024 IEEE International Conference on Robotics and Automation (ICRA), pp. 9411–9417. IEEE, 2024

    2024

  22. [30]

    Occupancy grid models for robot mapping in changing environments,

    D. Meyer-Delius, M. Beinhofer, and W. Burgard, “Occupancy grid models for robot mapping in changing environments,” inTwenty-Sixth AAAI Conference on Artificial Intelligence, 2012

    2012

  23. [31]

    Ssgm: Spatial semantic graph matching for loop closure detection in indoor environments,

    Y . Tang, M. Wang, Y . Deng, et al., “Ssgm: Spatial semantic graph matching for loop closure detection in indoor environments,” in2023 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pp. 9163–9168. IEEE, 2023

    2023

  24. [32]

    Multi-view robust collaborative localization in high outlier ratio scenes based on semantic features,

    Y . Tang, M. Wang, Y . Deng, et al., “Multi-view robust collaborative localization in high outlier ratio scenes based on semantic features,” in2023 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pp. 11 042–11 047. IEEE, 2023

    2023

  25. [33]

    Robust large-scale collabora- tive localization based on semantic submaps with extreme outliers,

    Y . Tang, M. Wang, Y . Yang, et al., “Robust large-scale collabora- tive localization based on semantic submaps with extreme outliers,” IEEE/ASME Transactions on Mechatronics, vol. 29, no. 4, pp. 2649– 2660, 2023

    2023

  26. [34]

    Semanticfusion: Dense 3d semantic mapping with convolutional neural networks,

    J. McCormac, A. Handa, A. Davison, et al., “Semanticfusion: Dense 3d semantic mapping with convolutional neural networks,” in2017 IEEE International Conference on Robotics and automation (ICRA), pp. 4628–

  27. [35]

    Gpmap: A unified framework for robotic mapping based on sparse gaussian processes,

    S. Kim and J. Kim, “Gpmap: A unified framework for robotic mapping based on sparse gaussian processes,” inField and Service Robotics: Results of the 9th International Conference, pp. 319–332. Springer, 2015

    2015

  28. [36]

    Hilbert maps: Scalable continuous occupancy mapping with stochastic gradient descent,

    F. Ramos and L. Ott, “Hilbert maps: Scalable continuous occupancy mapping with stochastic gradient descent,”The International Journal of Robotics Research, vol. 35, no. 14, pp. 1717–1730, 2016

    2016

  29. [37]

    Bayesian generalized kernel inference for terrain traversability mapping,

    T. Shan, J. Wang, B. Englot, et al., “Bayesian generalized kernel inference for terrain traversability mapping,” inConference on Robot Learning, pp. 829–838. PMLR, 2018

    2018

  30. [38]

    S-mki: Incremental dense semantic occupancy reconstruction through multi-entropy kernel inference,

    Y . Deng, M. Wang, D. Wang, et al., “S-mki: Incremental dense semantic occupancy reconstruction through multi-entropy kernel inference,” in 2022 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pp. 3824–3829. IEEE, 2022

    2022

  31. [39]

    Lgsdf: Continual global learning of signed distance fields aided by local updating,

    Y . Yue, Y . Deng, Y . Tang, et al., “Lgsdf: Continual global learning of signed distance fields aided by local updating,”IEEE Robotics and Automation Letters, 2025

    2025

  32. [40]

    Macim: Multi-agent collaborative implicit mapping,

    Y . Deng, Y . Tang, Y . Yang, et al., “Macim: Multi-agent collaborative implicit mapping,”IEEE Robotics and Automation Letters, 2024

    2024

  33. [41]

    imap: Implicit mapping and positioning in real-time,

    E. Sucar, S. Liu, J. Ortiz, et al., “imap: Implicit mapping and positioning in real-time,” inProceedings of the IEEE/CVF international conference on computer vision, pp. 6229–6238, 2021

    2021

  34. [42]

    Gs-slam: Dense visual slam with 3d gaussian splatting,

    C. Yan, D. Qu, D. Xu, et al., “Gs-slam: Dense visual slam with 3d gaussian splatting,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 19 595–19 604, 2024

    2024

  35. [43]

    Rtg-slam: Real-time 3d reconstruction at scale using gaussian splatting,

    Z. Peng, T. Shao, Y . Liu, et al., “Rtg-slam: Real-time 3d reconstruction at scale using gaussian splatting,” inACM SIGGRAPH 2024 Conference Papers, pp. 1–11, 2024

    2024

  36. [44]

    Splat-slam: Globally optimized rgb-only slam with 3d gaussians,

    E. Sandstr ¨om, G. Zhang, K. Tateno, et al., “Splat-slam: Globally optimized rgb-only slam with 3d gaussians,” inProceedings of the Computer Vision and Pattern Recognition Conference, pp. 1680–1691, 2025

    2025

  37. [45]

    Rgbd gs-icp slam,

    S. Ha, J. Yeon, and H. Yu, “Rgbd gs-icp slam,” inEuropean Conference on Computer Vision, pp. 180–197. Springer, 2024

    2024

  38. [46]

    Cartgs: Computational alignment for real-time gaussian splatting slam,

    D. Feng, Z. Chen, Y . Yin, et al., “Cartgs: Computational alignment for real-time gaussian splatting slam,”IEEE Robotics and Automation Letters, 2025

    2025

  39. [47]

    Hi-slam2: Geometry-aware gaussian slam for fast monocular scene reconstruction,

    W. Zhang, Q. Cheng, D. Skuddis, et al., “Hi-slam2: Geometry-aware gaussian slam for fast monocular scene reconstruction,”arXiv preprint arXiv:2411.17982, 2024

    2024

  40. [48]

    Decomposing nerf for editing via feature field distillation,

    S. Kobayashi, E. Matsumoto, and V . Sitzmann, “Decomposing nerf for editing via feature field distillation,”Advances in Neural Information Processing Systems, vol. 35, pp. 23 311–23 330, 2022

    2022

  41. [49]

    Openobj: Open-vocabulary object- level neural radiance fields with fine-grained understanding,

    Y . Deng, J. Wang, J. Zhao, et al., “Openobj: Open-vocabulary object- level neural radiance fields with fine-grained understanding,”IEEE Robotics and Automation Letters, 2024

    2024

  42. [50]

    Openmulti: Open-vocabulary instance- level multi-agent distributed implicit mapping,

    J. Dou, Y . Deng, J. Wang, et al., “Openmulti: Open-vocabulary instance- level multi-agent distributed implicit mapping,”IEEE Robotics and Automation Letters, 2025

    2025

  43. [51]

    Gaussian grouping: Segment and edit anything in 3d scenes,

    M. Ye, M. Danelljan, F. Yu, et al., “Gaussian grouping: Segment and edit anything in 3d scenes,” inEuropean Conference on Computer Vision, pp. 162–179. Springer, 2024

    2024

  44. [52]

    Opengaussian: Towards point-level 3d gaussian-based open vocabulary understanding,

    Y . Wu, J. Meng, H. Li, et al., “Opengaussian: Towards point-level 3d gaussian-based open vocabulary understanding,”Advances in Neural Information Processing Systems, vol. 37, pp. 19 114–19 138, 2024

    2024

  45. [53]

    Openvox: Real-time instance-level open-vocabulary probabilistic voxel representation,

    Y . Deng, B. Yao, Y . Tang, et al., “Openvox: Real-time instance-level open-vocabulary probabilistic voxel representation,” in2025 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). IEEE, 2025. 20

    2025

  46. [54]

    Yolo-world: Real-time open- vocabulary object detection,

    T. Cheng, L. Song, Y . Ge, et al., “Yolo-world: Real-time open- vocabulary object detection,” inProceedings of the IEEE/CVF Confer- ence on Computer Vision and Pattern Recognition, pp. 16 901–16 911, 2024

    2024

  47. [55]

    Tokenize anything via prompting,

    T. Pan, L. Tang, X. Wang, et al., “Tokenize anything via prompting,” in Proceedings of the European conference on computer vision (ECCV), 2024

    2024

  48. [56]

    Sentence-bert: Sentence embeddings using siamese bert-networks,

    N. Reimers and I. Gurevych, “Sentence-bert: Sentence embeddings using siamese bert-networks,” inProceedings of the 2019 Conference on Empirical Methods in Natural Language Processing, 11 2019

    2019

  49. [57]

    A volumetric method for building complex models from range images,

    B. Curless and M. Levoy, “A volumetric method for building complex models from range images,” inProceedings of the 23rd annual confer- ence on Computer graphics and interactive techniques, pp. 303–312, 1996

    1996

  50. [58]

    Bayesian spatial kernel smoothing for scalable dense semantic mapping,

    L. Gan, R. Zhang, J. W. Grizzle, et al., “Bayesian spatial kernel smoothing for scalable dense semantic mapping,”IEEE Robotics and Automation Letters, vol. 5, no. 2, pp. 790–797, 2020

    2020

  51. [59]

    Graspsplats: Efficient manipulation with 3d feature splatting,

    M. Ji, R.-Z. Qiu, X. Zou, et al., “Graspsplats: Efficient manipulation with 3d feature splatting,” inConference on Robot Learning, 2024

    2024

  52. [60]

    The replica dataset: A digital replica of indoor spaces,

    J. Straub, T. Whelan, L. Ma, et al., “The replica dataset: A digital replica of indoor spaces,”arXiv preprint arXiv:1906.05797, 2019

    1906 arXiv

  53. [61]

    Scannet: Richly-annotated 3d reconstructions of indoor scenes,

    A. Dai, A. X. Chang, M. Savva, et al., “Scannet: Richly-annotated 3d reconstructions of indoor scenes,” inProceedings of the IEEE conference on computer vision and pattern recognition, pp. 5828–5839, 2017. Yinan Deng(Student Member, IEEE) received the B.S. degree from Beijing ...

    2017

Pith tools

Reviewed August 4, 2026 · model on record in the stance chip above.