Pith. sign in

REVIEW 3 major objections 4 minor 2 cited by

SIREN: Semantic, Initialization-Free Registration of Multi-Robot Gaussian Splatting Maps

T0 review · 3 major / 4 minor · reviewed 2026-08-08 · deepseek-v4-flash

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

Pith's one-line read SIREN claims that semantic features alone can register and fuse multi-robot Gaussian Splatting maps, with no source images, camera poses, or initial inter-map transform.

desk verdict The first semantics-based GSplat map registration method that truly needs no poses or initialization, and worth refereeing, but the coarse-objective derivation in Section IV-B has a real algebraic gap that needs fixing. read the letter →

arxiv 2502.06519 v1 pith:CI6DND3H submitted 2025-02-10 cs.RO cs.CV

Ola Shorinwa , Jiankai Sun , Mac Schwager , Anirudha Majumdar This is my paper · ORCID
classification cs.ROcs.CV
keywords multi-robotmappingGaussianSplattingmapregistrationsemanticfeaturesinitialization-freefusionradiancefieldsopen-vocabularysemantics
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

Registering two maps of the same scene usually requires knowing roughly where they meet or having access to the original camera poses and images. SIREN claims to remove both requirements for Gaussian Splatting maps: it aligns and fuses maps using only the trained maps themselves, by exploiting the semantic features already embedded in the 3D scene representation. The method selects feature-rich regions through open-vocabulary language queries, matches Gaussians across maps by semantic similarity, computes a coarse non-rigid alignment in closed form, and then refines it photometrically using images rendered from the maps. On real robot data and standard benchmarks, the authors report large accuracy gains over baselines, including roughly 90x smaller rotation errors, 300x smaller translation errors, and 44x smaller scale errors in the most challenging scenes. If correct, this would let robot teams fuse maps without sharing raw sensor data or guessing an initial transform.

What carries the argument

The central mechanism is the semantically weighted Gaussian-to-Gaussian registration objective. Each Gaussian is assigned a semantic embedding by querying a neural field $\psi:\mathbb{R}^3\to\mathbb{R}^d$ trained jointly with the splat, and open-vocabulary text queries select feature-rich Gaussians for matching. Correspondences are proposed by cosine similarity between embeddings, then the coarse transform is obtained in closed form from an SVD of $\bar{Q}W\bar{P}^\top + \sum_{(i,j)\in E} w_{ij}\bar{H}_{q_j}\bar{H}_{p_i}^\top$, where the weights $w_{ij}$ are semantic similarities and the $\bar{H}$ terms encode each Gaussian's orientation and scale; RANSAC prunes false matches. The photometric stage renders novel views from both maps at corresponding poses, filters image pairs with vision-language similarity, and refines the transform through image matching and bundle adjustment. This two-stage design converts semantics into a reliable correspondence set and then into a high-accuracy photorealistic alignment.

What would settle it

Take a pair of independently trained Gaussian Splatting maps of the same scene whose only overlap is a uniform, untextured wall and floor with no objects a vision-language model can label; if SIREN cannot recover the known ground-truth transform, or produces a correspondence set too small for RANSAC, then the claim of general initialization-free registration is shown to depend on semantic content being present in the overlap.

Watch

Extended reading notes

Core claim

SIREN's central claim is that semantic grounding makes Gaussian Splatting map registration possible without camera poses, source images, or any initial inter-map transform. The pipeline trains a semantic field alongside each splat, uses open-vocabulary queries to isolate feature-rich Gaussians, matches those Gaussians by semantic similarity, and solves a weighted Gaussian-to-Gaussian alignment in closed form to obtain a coarse non-rigid transform. It then renders images from both maps at corresponding poses, filters the image pairs by semantic similarity, and runs image matching with bundle adjustment to refine the transform, producing a fused map at high visual fidelity. In experiments on real robot data and standard benchmarks, the most challenging scenes show roughly 90x smaller rotation errors, 300x smaller translation errors, and 44x smaller scale errors than competing methods.

Load-bearing premise

The maps must contain consistent, discriminative semantic features in the region they overlap; if semantic matching returns too few correct correspondences, the coarse alignment has nothing to start from and the whole pipeline cannot initialize.

Editorial extensions

If this is right

  • A team of robots can produce a single fused Gaussian Splatting map of an area no single robot can cover, as the paper demonstrates with a quadruped, a drone, and fixed-base manipulators.
  • Registration now requires only the trained maps as input, removing the need to store or share the original camera images and poses.
  • Semantic selection of feature-rich regions replaces the manual keypoint annotation and initial-guess requirements of earlier radiance-field registration methods.
  • The coarse-to-fine design yields fused maps that are photorealistic after refinement, not merely geometrically aligned, and the fused map can be finetuned with rendered data to remove floaters.
  • In the scenes evaluated, rotation, translation, and scale errors drop by roughly 90x, 300x, and 44x relative to baselines in the hardest cases.

Reading between the lines

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

  • If semantic features remain stable when maps are trained independently from different vantage points, the same coarse-to-fine idea could transfer to other implicit scene representations that carry semantic fields, not just Gaussian splatting maps.
  • Because only the trained maps are exchanged, SIREN points toward privacy- and bandwidth-friendly collaborative mapping: robots never need to share raw imagery or poses.
  • A direct stress test would vary the overlap's object content and vocabulary size; accuracy should degrade gracefully as semantic overlap shrinks, and fail when it vanishes.
  • The finetuning step suggests a self-supervised loop: rendered views of the fused map can be used to clean artifacts without any real-world data, which may make long-term map maintenance cheaper.
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 / 4 minor

Summary. The paper proposes SIREN, a three-stage pipeline for registering independently trained Gaussian Splatting maps from multiple robots: (i) semantic feature extraction and matching via CLIP-based semantic fields distilled into each GSplat, (ii) coarse Gaussian-to-Gaussian registration via a claimed closed-form solution to a weighted least-squares problem over means and covariances, and (iii) fine photometric registration using rendered images, CLIP filtering, feature matching, and bundle adjustment. The method claims to require no camera poses, no source images, and no inter-map initialization. Experiments compare SIREN variants with existing GSplat and point-cloud registration methods on Mip-NeRF360 scenes and on robot-collected data from a quadruped, drone, and manipulator, reporting geometric and photometric metrics. The paper also includes ablations and a finetuning procedure for the fused map.

Significance. If the empirical results hold, SIREN would be a meaningful advance: it removes the initialization and image/pose requirements of prior GSplat registration work and demonstrates cross-embodiment map fusion. The paper is also honest about its main limitation in Section VII: the method requires maps with embedded semantic codes and may fail in scenes without semantically relevant features. However, the central theoretical claim is compromised by an incorrect algebraic equivalence between Eq. (2) and Eq. (3), and the empirical evaluation omits the ground-truth transform protocol and run-to-run variance. These issues must be addressed before the contribution can be fully assessed.

major comments (3)
  1. [IV-B, Eqs. (2)-(3)] The derivation of the closed-form coarse registration is invalid as stated. The paper claims that the covariance term in Eq. (2), ||s_c^2 R \Sigma_{p_i} R^T - \Sigma_{q_j}||_F^2, can be expressed in the form of the term in Eq. (3), ||s_c R H_{p_i} \Lambda_{p_i} - H_{q_j} \Lambda_{q_j}||_F^2, using \Sigma_{p_i} = H_{p_i} \Lambda_{p_i} \Lambda_{p_i}^T H_{p_i}^T. This equivalence is false: the former equals ||(s_c R H_{p_i} \Lambda_{p_i})(s_c R H_{p_i} \Lambda_{p_i})^T - (H_{q_j} \Lambda_{q_j})(H_{q_j} \Lambda_{q_j})^T||_F^2, and for general matrices A and B, ||AA^T - BB^T||_F does not equal ||A - B||_F (e.g., A = I, B = -I gives 0 versus ||2I||_F). Consequently, the SVD solution in Eqs. (4)-(6) and Appendix A solves Eq. (3), not the stated objective Eq. (2). The text does not describe Eq. (3) as a surrogate or relaxation. Because the coarse alignment is the foundation for the fine photometric stage, this is a load-bearing correctness gap in the central claim of a rigorous registration pipeline. Please either present Eq. (3) explicitly as a surrogate objective with a justification, or solve the actual covariance objective.
  2. [V, Experimental Setup] The paper never states how ground-truth transforms were computed for the Mip-NeRF360 scene splits and for the robot-collected submaps. The reported rotation, translation, and scale errors are only meaningful relative to a known ground-truth alignment; without the procedure used to obtain those transforms (e.g., manually aligned poses, COLMAP-based poses from the full dataset, or marker-based calibration), the numerical error values cannot be reproduced or interpreted. In addition, the pipeline contains stochastic steps (random M-sampling in feature matching and RANSAC in coarse registration), yet all error metrics are reported as point values with no variance across repeated runs. Please document the ground-truth derivation and report means and standard deviations over multiple runs.
  3. [Abstract and V-B] The quantitative claims are inconsistent between the abstract and the results. The abstract states 'about 90x smaller rotation errors, 300x smaller translation errors, and 44x smaller scale errors in the most challenging scenes,' while Section V-B reports factors such as '160x, 465x, and 488x' in the Kitchen scene and '415x and 1287x' in the Workshop scene. Please reconcile these numbers or clarify which scene and comparison they refer to.
minor comments (4)
  1. [IV-B, Eq. (5)] The symbol \Sigma in the numerator of Eq. (5) is undefined; it should be \Sigma_c, the diagonal singular-value matrix from the SVD in Eq. (6), rather than the covariance matrices used elsewhere in the paper.
  2. [IV-C, Eq. (7)] The rotation matrices are denoted R_{a_i} and R_{b_j} in the surrounding text but R_{c_i} and R_{d_j} in Eq. (7); please unify the notation.
  3. [Appendix B, Table IV] The PhotoReg row in the Kitchen and Workshop columns appears to have the PSNR and SSIM values swapped (e.g., a PSNR of 0.75 and an SSIM of 11.6 cannot be correct); please verify the table formatting.
  4. [Algorithm 1 and IV-A] The correspondence set is denoted C in Algorithm 1 but E in Section IV-A; please use a consistent symbol.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: SIREN's registration chain is grounded in externally trained semantic features, standard closed-form similarity alignment, and image-based bundle adjustment; no fitted parameter is renamed as a prediction.

full rationale

SIREN's derivation chain is self-contained against external evidence. The semantic features come from pretrained vision-language models and are distilled into independently trained GSplat maps; correspondences are selected by cosine similarity on those features; the coarse stage uses a standard weighted Kabsch-Umeyama-style closed-form solution with an added covariance term; and the fine stage recovers camera poses from rendered images using SuperPoint/SuperGlue matching and bundle adjustment. The relative geometry between reconstructed source and target cameras is image-derived, so the final transform is not forced to equal the coarse transform; the coarse transform is used only to propose overlapping views. No inter-map ground-truth transform is used as an input or as a fit target, and no claimed prediction is defined in terms of the quantity being predicted. The paper's self-citations (FastSplat, Splat-Mover, Splat-Nav) are context, application examples, or a speed suggestion in an ablation discussion; they are not load-bearing for the derivation. The stated limitation that input maps must have embedded semantic codes is an explicit requirement, not a hidden circular assumption. The reviewer concern that Eq. (3) is not algebraically equivalent to Eq. (2) is a correctness or surrogate-objective issue, not a circularity: solving (3) does not encode the target result for (2). Appendix A's deferral of the optimality proof to [46] is an external citation, not a self-citation. Therefore, no circular step can be exhibited.

Assumptions & free parameters 6 free parameters · 4 assumptions · 0 invented entities

The method introduces no new physical entities, but relies on several unquantified hyperparameters and domain assumptions about semantic consistency and GSplat depth accuracy. The absence of reported values for these parameters makes the method hard to replicate exactly.

free parameters (6)
  • Semantic loss weights gamma and beta = not specified
    Relative weights balancing photometric and semantic losses in Eq. (1); values are not reported.
  • M, number of sampled candidate target Gaussians per source Gaussian = not specified
    Controls recall and precision of the correspondence set in feature matching; value not reported.
  • Distance threshold for pruning candidate matches = not specified
    Filters matches by proximity in the coarse matching step; value not reported.
  • CLIP text prompts and null prompt = not specified
    Semantic relevance scores depend on user queries; the queries used in experiments are not listed.
  • RANSAC iterations and inlier threshold = not specified
    Robustness to outliers in coarse registration depends on these; values not reported.
  • Weight beta_ij in fine registration objective = not specified
    Balances translation and rotation residuals in Eq. (7); value not reported.
assumptions (4)
  • domain assumption Gaussian Splatting provides accurate depth estimates for back-projecting pixels without depth supervision
    Invoked in Section IV-A to train the semantic field by querying 3D points along rays; if depth is inaccurate, semantic features are misaligned with the geometry.
  • domain assumption CLIP semantic features are consistent and discriminative across independently trained maps of the same scene
    Central to feature extraction and matching in Section IV-A; the paper acknowledges in Section VII that scenes lacking semantic features are a failure case.
  • domain assumption Scenes contain semantically relevant objects in the overlap region
    Stated in Section VII as a requirement; if the overlap has no recognizable objects, matching produces few correct correspondences.
  • ad hoc to paper The limit point of the fine registration problem as beta approaches zero is a valid approximation for the actual beta used
    Equation (8) uses the beta-to-zero limit to compute the optimal transform, but no derivation or bound is given for nonzero beta.

how reviews work

0 comments
Cite this review

Pith. "Pith review of SIREN: Semantic, Initialization-Free Registration of Multi-Robot Gaussian Splatting Maps." pith.science (2026). https://pith.science/paper/CI6DND3H

@misc{pith2026250206519,
  author       = {Pith},
  title        = {Pith review of: SIREN: Semantic, Initialization-Free Registration of Multi-Robot Gaussian Splatting Maps},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/CI6DND3H}},
  note         = {Machine review of arXiv:2502.06519}
}
read the original abstract

We present SIREN for registration of multi-robot Gaussian Splatting (GSplat) maps, with zero access to camera poses, images, and inter-map transforms for initialization or fusion of local submaps. To realize these capabilities, SIREN harnesses the versatility and robustness of semantics in three critical ways to derive a rigorous registration pipeline for multi-robot GSplat maps. First, SIREN utilizes semantics to identify feature-rich regions of the local maps where the registration problem is better posed, eliminating the need for any initialization which is generally required in prior work. Second, SIREN identifies candidate correspondences between Gaussians in the local maps using robust semantic features, constituting the foundation for robust geometric optimization, coarsely aligning 3D Gaussian primitives extracted from the local maps. Third, this key step enables subsequent photometric refinement of the transformation between the submaps, where SIREN leverages novel-view synthesis in GSplat maps along with a semantics-based image filter to compute a high-accuracy non-rigid transformation for the generation of a high-fidelity fused map. We demonstrate the superior performance of SIREN compared to competing baselines across a range of real-world datasets, and in particular, across the most widely-used robot hardware platforms, including a manipulator, drone, and quadruped. In our experiments, SIREN achieves about 90x smaller rotation errors, 300x smaller translation errors, and 44x smaller scale errors in the most challenging scenes, where competing methods struggle. We will release the code and provide a link to the project page after the review process.

Figures

Figures reproduced from arXiv: 2502.06519 by the authors.

Figure 1
Figure 1. SIREN enables robust registration (i.e., fusion) of multi-robot Gaussian Splatting maps, with no access to camera poses, images, and inter-map relative poses, via semantics-grounded optimization centered on feature-rich regions of each map. Abstract—We present SIREN for registration of multi-robot Gaussian Splatting (GSplat) maps, with zero access to camera poses, images, and inter-map transforms for initialization … view at source ↗
Figure 2
Figure 2. SIREN consists of three steps: (a) semantic feature extraction and matching of Gaussians across the local maps, (b) coarse Gaussian-to-Gaussian registration for coarsely aligning the local maps, (c) fine photometric registration for high-accuracy fusion of the local maps, through image-to-image registration and bundle adjustment. enables us to avoid training proposal networks (as required in NeRFs) that generate sam… view at source ↗
Figure 3
Figure 3. Although RANSAC-GR achieves the highest mean [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figures from the paper (6 more)
Figure 4
Figure 4. Figure 4: Rendered images from the fused GSplat maps of the [PITH_FULL_IMAGE:figures/full_fig_p009_4.png]
Figure 5
Figure 5. Figure 5: Stillshots of a quadruped mapping different areas of a kitchen and workshop and a drone mapping an apartment-like [PITH_FULL_IMAGE:figures/full_fig_p010_5.png]
Figure 6
Figure 6. Figure 6: Rendered images from the fused GSplat maps of the [PITH_FULL_IMAGE:figures/full_fig_p011_6.png]
Figure 7
Figure 7. Figure 7: Tabletop robotics tasks, e.g., manipulation, generally require robots to map the scene prior to completing the task. [PITH_FULL_IMAGE:figures/full_fig_p011_7.png]
Figure 8
Figure 8. Figure 8: Rendered images of the local maps of a tabletop scene trained by two manipulators. The maps provide high-fidelity [PITH_FULL_IMAGE:figures/full_fig_p012_8.png]
Figure 9
Figure 9. Figure 9: Rendered images from the fused GSplat maps generated by [PITH_FULL_IMAGE:figures/full_fig_p019_9.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 2 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. VISTA: Open-Vocabulary, Task-Relevant Robot Exploration with Online Semantic Gaussian Splatting

    cs.RO 2025-07 conditional novelty 6.0 of 10

    VISTA couples a view-diversity information metric with CLIP semantics in a receding-horizon planner to improve open-vocabulary object search during online Gaussian Splatting mapping on robots.

  2. WoMAP: World Models For Embodied Open-Vocabulary Object Localization

    cs.RO 2025-06 conditional novelty 6.0 of 10

    WoMAP generates training data from Gaussian Splatting scenes, distills detector confidence into a latent world model, and uses that model to refine vision-language action proposals for open-vocabulary object localization.

Reference graph

Works this paper leans on

56 extracted references · 36 canonical work pages · cited by 2 Pith papers

  1. [1]

    Netvlad: Cnn architecture for weakly supervised place recognition

    Relja Arandjelovic, Petr Gronat, Akihiko Torii, Tomas Pajdla, and Josef Sivic. Netvlad: Cnn architecture for weakly supervised place recognition. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 5297–5307, 2016

    work page 2016

  2. [2]

    Mip-nerf 360: Unbounded anti-aliased neural radiance fields

    Jonathan T Barron, Ben Mildenhall, Dor Verbin, Pratul P Srinivasan, and Peter Hedman. Mip-nerf 360: Unbounded anti-aliased neural radiance fields. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 5470–5479, 2022

    2022

  3. [3]

    Method for registration of 3-d shapes

    Paul J Besl and Neil D McKay. Method for registration of 3-d shapes. In Sensor fusion IV: control paradigms and data structures , volume 1611, pages 586–606. Spie, 1992

    work page 1992

  4. [4]

    Sparse iterative closest point

    Sofien Bouaziz, Andrea Tagliasacchi, and Mark Pauly. Sparse iterative closest point. In Computer graphics forum, volume 32, pages 113–123. Wiley Online Library, 2013

    work page 2013

  5. [5]

    Emerging properties in self-supervised vision transformers

    Mathilde Caron, Hugo Touvron, Ishan Misra, Herv´e J´egou, Julien Mairal, Piotr Bojanowski, and Armand Joulin. Emerging properties in self-supervised vision transformers. In Proceedings of the IEEE/CVF international conference on computer vision , pages 9650–9660, 2021

    work page 2021

  6. [6]

    Gaussreg: Fast 3d registration with gaussian splatting

    Jiahao Chang, Yinglin Xu, Yihao Li, Yuantao Chen, Wensen Feng, and Xiaoguang Han. Gaussreg: Fast 3d registration with gaussian splatting. In European Conference on Computer Vision , pages 407–423. Springer, 2025

    work page 2025

  7. [7]

    Splat-nav: Safe real-time robot navigation in gaussian splatting maps

    Timothy Chen, Ola Shorinwa, Joseph Bruno, Javier Yu, Weijia Zeng, Keiko Nagami, Philip Dames, and Mac Schwager. Splat-nav: Safe real-time robot navigation in gaussian splatting maps. arXiv preprint arXiv:2403.02751, 2024

    arXiv 2024

  8. [8]

    Object modelling by registration of multiple range images

    Yang Chen and G ´erard Medioni. Object modelling by registration of multiple range images. Image and vision computing, 10(3):145–155, 1992

    work page 1992

Show all 56 references
  1. [9]

    Dreg-nerf: Deep registra- tion for neural radiance fields

    Yu Chen and Gim Hee Lee. Dreg-nerf: Deep registra- tion for neural radiance fields. In Proceedings of the IEEE/CVF International Conference on Computer Vision , pages 22703–22713, 2023

    2023

  2. [10]

    Random sample consensus: a paradigm for model fitting with applications to image analysis and automated cartography

    Martin A Fischler and Robert C Bolles. Random sample consensus: a paradigm for model fitting with applications to image analysis and automated cartography. Communications of the ACM , 24(6):381–395, 1981

    1981

  3. [11]

    Robust point cloud registration framework based on deep graph matching

    Kexue Fu, Shaolei Liu, Xiaoyuan Luo, and Manning Wang. Robust point cloud registration framework based on deep graph matching. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages 8893–8902, 2021

    2021

  4. [12]

    nerf2nerf: Pairwise registration of neural radiance fields

    Lily Goli, Daniel Rebain, Sara Sabour, Animesh Garg, and Andrea Tagliasacchi. nerf2nerf: Pairwise registration of neural radiance fields. In 2023 IEEE International Conference on Robotics and Automation (ICRA) , pages 9354–9361. IEEE, 2023

    2023

  5. [13]

    Open-vocabulary object detection via vision and language knowledge distillation

    Xiuye Gu, Tsung-Yi Lin, Weicheng Kuo, and Yin Cui. Open-vocabulary object detection via vision and language knowledge distillation. arXiv preprint arXiv:2104.13921 , 2021

    2021 arXiv

  6. [14]

    Sugar: Surface- aligned gaussian splatting for efficient 3d mesh recon- struction and high-quality mesh rendering

    Antoine Gu ´edon and Vincent Lepetit. Sugar: Surface- aligned gaussian splatting for efficient 3d mesh recon- struction and high-quality mesh rendering. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 5354–5363, 2024

    2024

  7. [15]

    Registration with the point cloud library: A modular framework for aligning in 3-d

    Dirk Holz, Alexandru E Ichim, Federico Tombari, Radu B Rusu, and Sven Behnke. Registration with the point cloud library: A modular framework for aligning in 3-d. IEEE Robotics & Automation Magazine , 22(4):110–124, 2015

    2015

  8. [16]

    2d gaussian splatting for geometrically accurate radiance fields

    Binbin Huang, Zehao Yu, Anpei Chen, Andreas Geiger, and Shenghua Gao. 2d gaussian splatting for geometrically accurate radiance fields. In ACM SIGGRAPH 2024 Conference Papers, pages 1–11, 2024

    2024

  9. [17]

    Graspsplats: Efficient manipulation with 3d feature splatting

    Mazeyu Ji, Ri-Zhao Qiu, Xueyan Zou, and Xiaolong Wang. Graspsplats: Efficient manipulation with 3d feature splatting. arXiv preprint arXiv:2409.02084 , 2024

    2024 arXiv

  10. [18]

    Registering neural radiance fields as 3d density images

    Han Jiang, Ruoxuan Li, Haosen Sun, Yu-Wing Tai, and Chi-Keung Tang. Registering neural radiance fields as 3d density images. arXiv preprint arXiv:2305.12843 , 2023

    2023 arXiv

  11. [19]

    Image matching using sift, surf, brief and orb: perfor- mance comparison for distorted images

    Ebrahim Karami, Siva Prasad, and Mohamed Shehata. Image matching using sift, surf, brief and orb: perfor- mance comparison for distorted images. arXiv preprint arXiv:1710.02726, 2017

    2017 arXiv

  12. [20]

    3d gaussian splatting for real- time radiance field rendering

    Bernhard Kerbl, Georgios Kopanas, Thomas Leimk ¨uhler, and George Drettakis. 3d gaussian splatting for real- time radiance field rendering. ACM Trans. Graph. , 42(4): 139–1, 2023

    2023

  13. [21]

    Lerf: Language embed- ded radiance fields

    Justin Kerr, Chung Min Kim, Ken Goldberg, Angjoo Kanazawa, and Matthew Tancik. Lerf: Language embed- ded radiance fields. In Proceedings of the IEEE/CVF International Conference on Computer Vision , pages 19729–19739, 2023

    2023

  14. [22]

    Decomposing nerf for editing via feature field distillation

    Sosuke Kobayashi, Eiichi Matsumoto, and Vincent Sitz- mann. Decomposing nerf for editing via feature field distillation. Advances in Neural Information Processing Systems, 35:23311–23330, 2022

    2022

  15. [23]

    Deblurring 3d gaussian splatting

    Byeonghyeon Lee, Howoong Lee, Xiangyu Sun, Usman Ali, and Eunbyung Park. Deblurring 3d gaussian splatting. In European Conference on Computer Vision , pages 127–

  16. [24]

    Manigaussian: Dynamic gaussian splatting for multi-task robotic manipulation

    Guanxing Lu, Shiyi Zhang, Ziwei Wang, Changliu Liu, Jiwen Lu, and Yansong Tang. Manigaussian: Dynamic gaussian splatting for multi-task robotic manipulation. In European Conference on Computer Vision, pages 349–366. Springer, 2025

    2025

  17. [25]

    Image segmentation using text and image prompts

    Timo L¨uddecke and Alexander Ecker. Image segmentation using text and image prompts. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 7086–7096, 2022

    2022

  18. [26]

    Clip4clip: An empirical study of clip for end to end video clip retrieval and captioning

    Huaishao Luo, Lei Ji, Ming Zhong, Yang Chen, Wen Lei, Nan Duan, and Tianrui Li. Clip4clip: An empirical study of clip for end to end video clip retrieval and captioning. Neurocomputing, 508:293–304, 2022

    2022

  19. [27]

    Nerf: Representing scenes as neural radiance fields for view synthesis

    Ben Mildenhall, Pratul P Srinivasan, Matthew Tancik, Jonathan T Barron, Ravi Ramamoorthi, and Ren Ng. Nerf: Representing scenes as neural radiance fields for view synthesis. Communications of the ACM , 65(1):99–106, 2021

    2021

  20. [28]

    Simple open-vocabulary object detection

    Matthias Minderer, Alexey Gritsenko, Austin Stone, Maxim Neumann, Dirk Weissenborn, Alexey Dosovitskiy, Aravindh Mahendran, Anurag Arnab, Mostafa Dehghani, Zhuoran Shen, et al. Simple open-vocabulary object detection. In European Conference on Computer Vision , pages 728–755. ...

    2022

  21. [29]

    Clip- cap: Clip prefix for image captioning

    Ron Mokady, Amir Hertz, and Amit H Bermano. Clip- cap: Clip prefix for image captioning. arXiv preprint arXiv:2111.09734, 2021

    2021 arXiv

  22. [30]

    Dinov2: Learning robust visual features without supervision

    Maxime Oquab, Timoth´ee Darcet, Th ´eo Moutakanni, Huy V o, Marc Szafraniec, Vasil Khalidov, Pierre Fernandez, Daniel Haziza, Francisco Massa, Alaaeldin El-Nouby, et al. Dinov2: Learning robust visual features without supervision. arXiv preprint arXiv:2304.07193 , 2023

    2023 arXiv

  23. [31]

    Colored point cloud registration revisited

    Jaesik Park, Qian-Yi Zhou, and Vladlen Koltun. Colored point cloud registration revisited. In Proceedings of the IEEE international conference on computer vision , pages 143–152, 2017

    2017

  24. [32]

    Langsplat: 3d language gaussian splatting

    Minghan Qin, Wanhua Li, Jiawei Zhou, Haoqian Wang, and Hanspeter Pfister. Langsplat: 3d language gaussian splatting. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 20051– 20060, 2024

    2024

  25. [33]

    Geotransformer: Fast and robust point cloud registration with geometric transformer

    Zheng Qin, Hao Yu, Changjian Wang, Yulan Guo, Yuxing Peng, Slobodan Ilic, Dewen Hu, and Kai Xu. Geotransformer: Fast and robust point cloud registration with geometric transformer. IEEE Transactions on Pattern Analysis and Machine Intelligence , 45(8):9806–9821, 2023

    2023

  26. [34]

    Learning generalizable feature fields for mobile manipulation

    Ri-Zhao Qiu, Yafei Hu, Ge Yang, Yuchen Song, Yang Fu, Jianglong Ye, Jiteng Mu, Ruihan Yang, Nikolay Atanasov, Sebastian Scherer, et al. Learning generalizable feature fields for mobile manipulation. arXiv preprint arXiv:2403.07563, 2024

    2024 arXiv

  27. [35]

    Learning transferable visual models from natural language supervision

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. Learning transferable visual models from natural language supervision. In International conference on machine learning, pa...

    2021

  28. [36]

    Language embedded radiance fields for zero-shot task-oriented grasping

    Adam Rashid, Satvik Sharma, Chung Min Kim, Justin Kerr, Lawrence Yunliang Chen, Angjoo Kanazawa, and Ken Goldberg. Language embedded radiance fields for zero-shot task-oriented grasping. In 7th Annual Conference on Robot Learning , 2023

    2023

  29. [37]

    Efficient variants of the icp algorithm

    Szymon Rusinkiewicz and Marc Levoy. Efficient variants of the icp algorithm. In Proceedings third international conference on 3-D digital imaging and modeling , pages 145–152. IEEE, 2001

    2001

  30. [38]

    Fast point feature histograms (fpfh) for 3d registration

    Radu Bogdan Rusu, Nico Blodow, and Michael Beetz. Fast point feature histograms (fpfh) for 3d registration. In 2009 IEEE international conference on robotics and automation, pages 3212–3217. IEEE, 2009

    2009

  31. [39]

    Superglue: Learning feature matching with graph neural networks

    Paul-Edouard Sarlin, Daniel DeTone, Tomasz Malisiewicz, and Andrew Rabinovich. Superglue: Learning feature matching with graph neural networks. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 4938–4947, 2020

    2020

  32. [40]

    Structure-from-motion revisited

    Johannes L Schonberger and Jan-Michael Frahm. Structure-from-motion revisited. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 4104–4113, 2016

    2016

  33. [41]

    Distilled feature fields enable few-shot language-guided manipulation

    William Shen, Ge Yang, Alan Yu, Jansen Wong, Leslie Pack Kaelbling, and Phillip Isola. Distilled feature fields enable few-shot language-guided manipulation. arXiv preprint arXiv:2308.07931 , 2023

    2023 arXiv

  34. [42]

    Fast- splat: Fast, ambiguity-free semantics transfer in gaussian splatting

    Ola Shorinwa, Jiankai Sun, and Mac Schwager. Fast- splat: Fast, ambiguity-free semantics transfer in gaussian splatting. arXiv preprint arXiv:2411.13753 , 2024

    2024 arXiv

  35. [43]

    Splat-mover: Multi- stage, open-vocabulary robotic manipulation via editable gaussian splatting

    Ola Shorinwa, Johnathan Tucker, Aliyah Smith, Aiden Swann, Timothy Chen, Roya Firoozi, Monroe David Kennedy, and Mac Schwager. Splat-mover: Multi- stage, open-vocabulary robotic manipulation via editable gaussian splatting. In 8th Annual Conference on Robot Learning, 2024

    2024

  36. [44]

    Registration of 3d point clouds and meshes: A survey from rigid to nonrigid

    Gary KL Tam, Zhi-Quan Cheng, Yu-Kun Lai, Frank C Langbein, Yonghuai Liu, David Marshall, Ralph R Martin, Xian-Fang Sun, and Paul L Rosin. Registration of 3d point clouds and meshes: A survey from rigid to nonrigid. IEEE transactions on visualization and computer graphics , 19 ...

    2012

  37. [45]

    Nerfstudio: A modular framework for neural radiance field development

    Matthew Tancik, Ethan Weber, Evonne Ng, Ruilong Li, Brent Yi, Terrance Wang, Alexander Kristoffersen, Jake Austin, Kamyar Salahi, Abhik Ahuja, et al. Nerfstudio: A modular framework for neural radiance field development. In ACM SIGGRAPH 2023 Conference Proceedings , pages 1–12, 2023

    2023

  38. [46]

    Least-squares estimation of transfor- mation parameters between two point patterns

    Shinji Umeyama. Least-squares estimation of transfor- mation parameters between two point patterns. IEEE Transactions on Pattern Analysis & Machine Intelligence , 13(04):376–380, 1991

    1991

  39. [47]

    Clip-nerf: Text-and-image driven manipulation of neural radiance fields

    Can Wang, Menglei Chai, Mingming He, Dongdong Chen, and Jing Liao. Clip-nerf: Text-and-image driven manipulation of neural radiance fields. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 3835–3844, 2022

    2022

  40. [48]

    Deep closest point: Learning representations for point cloud registration

    Yue Wang and Justin M Solomon. Deep closest point: Learning representations for point cloud registration. In Proceedings of the IEEE/CVF international conference on computer vision , pages 3523–3532, 2019

    2019

  41. [49]

    Cris: Clip-driven referring image segmentation

    Zhaoqing Wang, Yu Lu, Qiang Li, Xunqiang Tao, Yan- dong Guo, Mingming Gong, and Tongliang Liu. Cris: Clip-driven referring image segmentation. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 11686–11695, 2022

    2022

  42. [50]

    pixelnerf: Neural radiance fields from one or few images

    Alex Yu, Vickie Ye, Matthew Tancik, and Angjoo Kanazawa. pixelnerf: Neural radiance fields from one or few images. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages 4578– 4587, 2021

    2021

  43. [51]

    Mip-splatting: Alias-free 3d gaussian splatting

    Zehao Yu, Anpei Chen, Binbin Huang, Torsten Sattler, and Andreas Geiger. Mip-splatting: Alias-free 3d gaussian splatting. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 19447– 19456, 2024

    2024

  44. [52]

    Photoreg: Photometrically regis- tering 3d gaussian splatting models

    Ziwen Yuan, Tianyi Zhang, Matthew Johnson-Roberson, and Weiming Zhi. Photoreg: Photometrically regis- tering 3d gaussian splatting models. arXiv preprint arXiv:2410.05044, 2024

    2024 arXiv

  45. [53]

    Nerf++: Analyzing and improving neural radiance fields

    Kai Zhang, Gernot Riegler, Noah Snavely, and Vladlen Koltun. Nerf++: Analyzing and improving neural radiance fields. arXiv preprint arXiv:2010.07492 , 2020

    2010 arXiv

  46. [54]

    Fast global registration

    Qian-Yi Zhou, Jaesik Park, and Vladlen Koltun. Fast global registration. In Computer Vision–ECCV 2016: 14th European Conference, Amsterdam, The Netherlands, October 11-14, 2016, Proceedings, Part II 14 , pages 766–

    2016

  47. [55]

    Feature 3dgs: Supercharging 3d gaussian splatting to enable distilled feature fields

    Shijie Zhou, Haoran Chang, Sicheng Jiang, Zhiwen Fan, Zehao Zhu, Dejia Xu, Pradyumna Chari, Suya You, Zhangyang Wang, and Achuta Kadambi. Feature 3dgs: Supercharging 3d gaussian splatting to enable distilled feature fields. In Proceedings of the IEEE/CVF Conference on Computer...

    2024

  48. [56]

    Loopsplat: Loop closure by registering 3d gaussian splats

    Liyuan Zhu, Yue Li, Erik Sandstr ¨om, Shengyu Huang, Konrad Schindler, and Iro Armeni. Loopsplat: Loop closure by registering 3d gaussian splats. arXiv preprint arXiv:2408.10154, 2024. APPENDIX A COARSE GAUSSIAN -TO-G AUSSIAN REGISTRATION We discuss the derivation of the close...

    2024 arXiv

Pith tools

Reviewed August 8, 2026 · model on record in the stance chip above.