Pith. sign in

REVIEW 4 major objections 4 minor 35 references

The paper argues that real-time multi-camera 3D reconstruction can be decoupled into target-free calibration and stateless fusion, with total cost linear in cameras and pixels.

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

2026-08-02 12:23 UTC pith:MD7DMBKP

load-bearing objection The FUSE half is a plausible engineering contribution, but GMAC's reported rotation-error gains are mathematically impossible under its own scale-only update, and the missing equations make the rest unverifiable. the 4 major comments →

arxiv 2606.04376 v2 pith:MD7DMBKP submitted 2026-06-03 eess.IV cs.MM

FUSE-Flow: A Decoupled Framework for Calibration and Stateless Real-Time Multi-View Point Cloud Fusion

Chentian Sun This is my paper
classification eess.IV cs.MM
keywords multi-camera 3D reconstructionextrinsic calibrationmetric scale recoverypoint cloud fusionconfidence weightingadaptive spatial hashingreal-time performancestateless fusion
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.

FUSE-Flow is a two-module framework for synchronized RGB-D camera arrays. It tries to show that the usual coupling between extrinsic calibration, multi-view fusion, and global optimization is not necessary: a calibration module, GMAC, lifts the relative-scale pose predictions of a pretrained multi-view backbone into metric camera poses using only geometric consistency, and a fusion module, FUSE, merges each frame's depth maps with confidence weighting and adaptive spatial hashing while carrying no state across frames. If the framework is right, large camera setups could be calibrated without checkerboards or bundle adjustment and reconstruct dynamic scenes in real time with memory and compute that grow only linearly with camera count. The key and most fragile premise is that the backbone's relative translations differ from true metric translations by a single global scale factor.

Core claim

The central claim is that the accuracy bottleneck in real-time multi-camera reconstruction is not the quality of any single component but their entanglement. GMAC shows that one scalar, estimated in closed form by minimizing depth residuals at depth-guided correspondences and pruned by cycle-consistency checks, can convert a backbone's globally coupled but scale-ambiguous pose output into metric extrinsics. FUSE shows that per-frame point clouds can be fused without a global map: every point receives a reliability weight from depth-gradient statistics and cross-view depth residuals, and adaptive spatial hashing keeps representative points per cell. The paper reports that this combination out

What carries the argument

The carrying mechanism is a scale-only correction combined with a stateless fusion rule. GMAC's closed-form scale estimator is a weighted least-squares solution for a single global scale applied uniformly to the backbone's translation vectors; cycle-consistency validation rejects graph edges that break triplet coherence before re-estimating the scale. FUSE's two-level adaptive spatial hashing subdivides dense regions into finer voxels and then selects representative points by a score that rewards both high measurement confidence and cross-view consistency; the harmonic-mean fusion weight means a point must pass both gates to survive. These two pieces are what let calibration run without bund

Load-bearing premise

The load-bearing premise is that a pretrained backbone's predicted camera translations are globally related to true metric translations by a single scalar, so that one closed-form factor can lift the whole pose graph into metric units; if different camera pairs carry different scale errors, or if the backbone's rotations are wrong, the scale-only estimator cannot fix them.

What would settle it

Measure each camera pair's true baseline (e.g., with a tape or laser) and compare it with the pairwise translation magnitude predicted by the backbone on the same synchronized frame. If the per-pair ratios are not approximately equal across all pairs, the single-scale model cannot hold, and GMAC's closed-form estimator would be fitting a nonexistent parameter. A second, targeted check: since the paper reports rotation-error improvements, an experiment with a backbone whose rotation predictions are artificially corrupted but whose translations are left unchanged should show no rotation recovery

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

Share X Bluesky LinkedIn Reddit HN

If this is right

  • A camera rig can be calibrated automatically from ordinary video if a pretrained pose backbone is available, with no checkerboards, calibration targets, or global bundle adjustment.
  • Fusion cost and memory stay linear in camera count and pixels, so adding more cameras to an array should degrade frame rate gracefully rather than quadratically.
  • Dynamic scenes are handled better than stateful TSDF or SLAM pipelines because no previous frame is needed to fuse the current one.
  • Residual calibration error is absorbed by the confidence weighting, so reconstruction quality should degrade gradually under pose noise instead of collapsing.
  • The pipeline is positioned for immersive media and interactive use, not sub-millimeter metrology, since global optimization is deliberately omitted.

Where Pith is reading between the lines

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

  • Because the scale estimator assumes one global scalar, a natural generalization is per-cluster scale factors with cluster-level cycle consistency; this would be a testable fix for the paper's own concern about ultra-large camera arrays.
  • The closed-form solution fixes correspondences while the scale varies; an iterative variant that re-projects and re-selects correspondences after each scale update would reveal whether the fixed-correspondence approximation biases the estimate.
  • The graceful-degradation claim could be stress-tested directly: perturb input extrinsics by controlled amounts and check whether fusion metrics degrade linearly; if they degrade sharply, the stateless fusion is less robust than stated.
  • A moving-camera version would stress the neighbor-graph assumption, since the bounded neighbor set is precomputed; dynamic topology might require rebuilding the graph, which could reintroduce super-linear cost.

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

4 major / 4 minor

Summary. The paper proposes FUSE-Flow, a decoupled framework for real-time multi-camera 3D reconstruction. It consists of two modules: GMAC, which converts scale-ambiguous relative poses from pretrained multi-view reconstruction backbones into metric extrinsics using a single global scale factor estimated from depth-guided reprojection and cycle-consistency validation; and FUSE, a stateless, confidence-guided point cloud fusion module built on two-level adaptive spatial hashing and representative point selection. The authors claim linear O(NHW) per-frame complexity, no inter-frame state, and synergy between calibration and fusion. Experiments report improved rotation/translation errors against backbone baselines and better reconstruction metrics against R3D3.

Significance. If the claims were sound, the decoupled design and the stateless linear-complexity fusion would be a practical contribution for large multi-camera arrays. The explicit complexity analysis and the design goal of avoiding global optimization are commendable. However, the central calibration result is internally inconsistent: the described scale-only update cannot produce the reported rotation-error improvements, and the closed-form estimator is not actually specified in the manuscript. The FUSE component may be viable as an engineering contribution, but the paper's primary empirical evidence for GMAC—and for the combined pipeline—rests on a mechanism that the equations do not support. No code or data are provided, which further limits verification of the empirical claims.

major comments (4)
  1. [§III.B.3, Eqs. (4), (14), and Tables I–III] The described GMAC mechanism cannot produce the reported rotation-error reductions. Equation (4) defines a single global scale s applied to translations, and Eq. (14) outputs corrected extrinsics as [R | s·t]. Rotations are never updated. The cycle-consistency validation in §III.B.3 prunes graph edges and re-estimates s, but no mechanism changes R. Yet Table I reports large rotation-error improvements for every backbone (e.g., VGGT: 1.04°→0.63°; MapAnything: 1.19°→0.74°; VGGSfM: 1.31°→0.89°), and Table III repeats this pattern in ablations. Since the rotation errors of 'Ours' should equal 'Original' under the stated equations, the primary evidence for GMAC's standalone contribution is internally inconsistent. The limitation statement in §V ('the single-scale design') confirms that only a scalar scale is used.
  2. [§III.B.3, Eqs. (9)–(12)] The closed-form scale estimator is not actually specified. The text states 'Let [Eqs. (9)–(11)]' and then presents Eq. (12), but the definitions of the quantities a, b, c (or their algebraic forms) are missing from the manuscript. The quadratic objective and the claimed minimizer in Eq. (12) therefore cannot be audited. Equation (13) likewise introduces the cycle transformation T_{ijk} without a definition. This is a load-bearing derivational gap: the scale estimator and cycle validation are the core of GMAC, and the paper provides no way to check the claimed closed-form result.
  3. [§IV.B, Tables I–III and VIII] The translation-error comparison for the 'Original' baseline is ill-posed. The paper repeatedly states that backbone translations are scale-ambiguous (e.g., §III.B.2 and §III.B.3). Table I reports 'Original' translation error in millimeters against metric ground truth, but no procedure is given for placing raw relative-scale predictions into metric units before computing the error. Without such a rescaling step, the 'Original' numbers and the percentage improvements in §IV.B.1 have no defined meaning. The same issue affects Table VIII, where 'Original' extrinsics are used to run FUSE; if these are truly scale-ambiguous, the resulting fusion metrics are also undefined.
  4. [§III.C.4] The linear-complexity proof is incomplete. The text asserts O(NHW) per-frame complexity because the neighbor graph has bounded degree k_c and each fine cell retains at most k_f representative points. However, no method is given for constructing a bounded-degree overlap graph for arbitrary N, and no argument shows that k_c remains constant as the camera count grows in a dense array. The statement that 'k_f and k_c are bounded constants' is an assumption, not a derivation. Since linear scalability is a central contribution, this gap needs either an explicit algorithm with proven bounds or a clear limitation statement for dense configurations.
minor comments (4)
  1. [Abstract and §III.D] The abstract states that 'confidence-aware fusion corrects calibration biases,' but §III.D and §IV.D describe FUSE as absorbing or tolerating residual extrinsic errors, not actively correcting them. Please align the wording with the actual mechanism.
  2. [§II.C and §III.C.4] The complexity notation is inconsistent: §II.C uses O(N²) and O(V) without defining V, while §III.C.4 uses O(NHW) and O(M). Define all quantities and keep a single notational convention.
  3. [§IV.A.1] The real-world multi-camera dataset is described only as 'synchronized indoor video sequences' with a fixed array. Camera count, baseline distances, synchronization method, and scene diversity are missing, which hampers reproducibility.
  4. [General] Several references are dated 2025 or 2026 (e.g., [5], [12], [30], [35]); please verify these citations and ensure they are accessible and correctly formatted.

Circularity Check

0 steps flagged

No load-bearing circularity: GMAC's scale fit is the stated method, not a disguised evaluation target; FUSE is evaluated under controlled extrinsics.

full rationale

The derivation chain is self-contained. GMAC takes backbone relative poses and depth-guided correspondences and solves a single global scale s via closed-form least squares (Eqs. 4-12), then applies it in Eq. (14) as [R | s t]. This is a calibration fit to depth reprojection consistency, not a fit to the reported reconstruction metrics (PFR, PNCC, OR, CD) or to the ground-truth translation error used for evaluation; translation-error improvement is the expected consequence of anchoring scale, not a re-labeling of the fitted objective. FUSE is evaluated in isolation with ground-truth/pseudo-ground-truth extrinsics (Section IV.C), so its fusion quality is not determined by GMAC's fitted scale. The main empirical oddity is that Tables I-III report rotation-error reductions even though Eq. (14) leaves rotations untouched; the Conclusion's own limitation statement ('the single-scale design') reinforces this. That is an internal consistency/correctness problem, not circularity, because no rotation parameter is fitted from the evaluation target. The only mild coupling is that the same depth frames are used both to recover scale and to measure reconstruction quality, but the paper's controlled-extrinsic experiments break that coupling. No self-citation chain is load-bearing; references are external. Therefore no circular step can be exhibited.

Axiom & Free-Parameter Ledger

6 free parameters · 6 axioms · 0 invented entities

The framework introduces no new physical entities. Its free parameters are mostly unspecified hyperparameters for confidence scoring and spatial hashing. The domain assumptions are load-bearing: if the backbone poses are not globally similarity-consistent, or depth is not metric, the whole pipeline loses its claimed calibration and fusion accuracy.

free parameters (6)
  • global scale s = least-squares estimate, not reported per scene
    Estimated from depth-guided correspondences in Eq. (12) to lift relative poses to metric scale; central output of GMAC.
  • confidence weights α, β = not reported
    Weights in Eq. (18) for depth-gradient and local-variance terms in measurement confidence modeling.
  • minimum depth threshold δ = not reported
    Threshold in Eq. (18) to suppress invalid or too-close depth measurements.
  • relative depth error threshold ε = not reported; said to match sensor accuracy
    Used in Eq. (21) to gate cross-view consistency.
  • coarse grid size l_c = not reported
    Level-1 spatial hashing cell size in Eq. (23).
  • fine voxel parameters l_f, bounds [l_min,l_max], base count k_0, ratio r, max K_max = not reported
    Adaptive fine-grid subdivision and representative point selection parameters in Eqs. (24)-(26).
axioms (6)
  • domain assumption The chosen multi-view reconstruction backbone (VGGT/MapAnything/VGGSfM) produces globally coupled relative poses with correct inter-camera distance ratios.
    GMAC relies on this to assume a single scale factor s can convert predictions to metric extrinsics (§III.B.2, Eq. 4).
  • domain assumption A single scalar scale factor is sufficient to correct all translation magnitudes.
    If per-camera scale drift exists, one s cannot fix poses; authors acknowledge 'best-fit', but the central claim depends on it (§III.B.3).
  • domain assumption Depth maps are metric and correspond to the same scene as the RGB images; camera intrinsics are known.
    Both GMAC scale estimation and FUSE fusion treat depth as ground truth (§III.A).
  • domain assumption Correspondences can be treated as fixed while s varies.
    The closed-form estimator relies on the fixed-correspondence approximation (§III.B.3, Eqs. 9-12); if correspondences shift with scale, the quadratic form is approximate.
  • domain assumption Sensor depth noise variance grows quadratically with depth.
    MCM normalizes gradient and variance by d^2 to make confidence depth-invariant (§III.C.1, Eqs. 16-17).
  • domain assumption The neighboring-camera graph has bounded degree and is precomputed from field-of-view overlap.
    The linear-complexity proof rests on |N(i)| ≤ k_n constant (§III.C.2, §III.C.4).

pith-pipeline@v1.3.0-alltime-deepseek · 17787 in / 14012 out tokens · 129923 ms · 2026-08-02T12:23:19.115505+00:00 · methodology

0 comments
Cite this review

Pith. "Pith review of FUSE-Flow: A Decoupled Framework for Calibration and Stateless Real-Time Multi-View Point Cloud Fusion." pith.science (2026). https://pith.science/paper/MD7DMBKP

@misc{pith2026260604376,
  author       = {Pith},
  title        = {Pith review of: FUSE-Flow: A Decoupled Framework for Calibration and Stateless Real-Time Multi-View Point Cloud Fusion},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/MD7DMBKP}},
  note         = {Machine review of arXiv:2606.04376}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Real-time multi-camera 3D reconstruction is a key foundation for immersive media, remote interaction and spatial computing. While synchronized camera arrays are widely adopted, achieving geometrically consistent and scalable real-time reconstruction remains challenging. A key challenge is the close linkage among extrinsic calibration, multi-view fusion and global optimization, which causes fluctuating reconstruction results, cumulative errors and poor system expandability. We propose a decoupled framework for calibration and stateless real-time multi-view point cloud fusion (FUSE-Flow), a framework with two collaborative components: geometry-aligned multi-view extrinsic calibration (GMAC) and reliability-guided multi-view point cloud fusion (FUSE). This split design avoids conflicting optimization objectives for targeted improvement. The GMAC module refines camera extrinsics via geometric constraints and multi-view reconstruction transformers, enabling accurate sparse-view calibration without calibration targets, dense images or global bundle adjustment. The FUSE module integrates confidence weighting and adaptive spatial hashing for stateless fusion, ensuring linear time and memory consumption. The two modules mutually reinforce each other: accurate camera poses boost fusion accuracy, and confidence-aware fusion corrects calibration biases. Validated on public datasets and real camera setups, FUSE-Flow outperforms mainstream real-time reconstruction methods in visual effect, dynamic stability and scalability, providing a practical solution for large-scale real-time 3D reconstruction.

Review history (2 revisions) →

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

35 extracted references · 1 linked inside Pith

  1. [1]

    M. Wang, L. Weng, and F. Gao, ―An automatic extrinsic calibration method for LiDAR -camera fusion via combining semantic and geometric features,‖ in Proc. IEEE Int. Conf. Acoust., Speech Signal Process. (ICASSP) , Hyderabad, India, 2025, pp. 1–5

    2025

  2. [2]

    S. Hu, A. Goldwurm, M. Mujica, S. Cadou, and F. Lerasle, ―A universal framework for extrinsic calibration of camera, radar, and LiDAR,‖ IEEE Robot. Autom. Lett., vol. 11, no. 2, pp. 1842– 1849, 2026

    2026

  3. [3]

    S. Wang, V. Leroy, Y. Cabon, B. Chidlovskii, and J. Revaud, ―DUSt3R: Geometric 3D vision made easy,‖ in Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit. (CVPR), Seattle, WA, USA, 2024, pp. 20697–20709

    2024

  4. [4]

    Campos, R

    C. Campos, R. Elvira, J. J. G. Rodrí guez, J. M. M. Montiel, and J. D. Tardós, ―ORB-SLAM3: An accurate open -source library for visual, visual –inertial, and multimap SLAM,‖ IEEE Trans. Robot., vol. 37, no. 6, pp. 1874–1890, 2021

    2021

  5. [5]

    R. Hai, Y. Shen, Y. Yan, S. Chen, J. Xin, and N. Zheng, ―FlowCalib: Targetless infrastructure LiDAR -camera extrinsic calibration based on optical flow and scene flow,‖ IEEE Trans. Intell. Transp. Syst., vol. 27, no. 1, pp. 1565–1577, 2026

    2026

  6. [6]

    J. Wang, N. Karaev, C. Rupprecht, and D. Novotny, ―VGGSfM: Visual geometry grounded deep structure from motion,‖ in Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit. (CVPR), Seattle, WA, USA, 2024, pp. 21686–21697

    2024

  7. [7]

    X. He, J. Sun, Y. Wang, S. Peng, Q. Huang, H. Bao, and X. Zhou, ―Detector-free structure from motion,‖ in Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit. (CVPR), Seattle, WA, USA, 2024, pp. 21138–21147

    2024

  8. [8]

    J. Wang, C. Rupprecht, and D. Novotny, ―PoseDiffusion: Solving pose estimation via diffusion-aided bundle adjustment,‖ in Proc. IEEE/CVF Int. Conf. Comput. Vis. (ICCV) , Paris, France, 2023, pp. 10245–10255

    2023

  9. [9]

    J. Wang, M. Chen, N. Karaev, A. Vedaldi, C. Rupprecht, and D. Novotny, ―VGGT: Visual geometry grounded transformer,‖ in Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit. (CVPR) , Nashville, TN, USA, 2025, pp. 5294–5306

    2025

  10. [10]

    Keetha et al., ―MapAnything: Universal feed-forward metric 3D reconstruction,‖ arXiv preprint arXiv:2509.13414, 2025

    N. Keetha et al., ―MapAnything: Universal feed-forward metric 3D reconstruction,‖ arXiv preprint arXiv:2509.13414, 2025

    Pith/arXiv arXiv 2025

  11. [11]

    J. Zhao, Q. Zhu, Y. Wang, W. Peng, H. Zhang, and J. Mao, ―Registration of multiview point clouds with unknown overlap,‖ IEEE Trans. Multimedia, vol. 27, pp. 804–819, 2025

    2025

  12. [12]

    Zhang, K

    X. Zhang, K. Bi, S. Chan, S. Lu, and X. Zhou, ―SyNet: A synergistic network for 3D object detection through geometric-semantic-based multi-interaction fusion,‖ IEEE Trans. Multimedia, vol. 27, pp. 4950–4960, 2025

    2025

  13. [13]

    F. Chu, Y. Cong, Y. Wang, and R. Chen, ―DetailRecon: Focusing on detailed regions for online monocular 3D reconstruction,‖ IEEE Trans. Multimedia, vol. 27, pp. 3266–3278, 2025

    2025

  14. [14]

    R. A. Newcombe et al., ―KinectFusion: Real -time dense surface mapping and tracking,‖ in Proc. 10th IEEE Int. Symp. Mixed Augmented Reality (ISMAR), Basel, Switzerland, 2011, pp. 127– 136

    2011

  15. [15]

    Y. Yao, Z. Luo, S. Li, T. Fang, and L. Quan, ―MVSNet: Depth inference for unstructured multi-view stereo,‖ in Proc. Eur. Conf. Comput. Vis. (ECCV), 2018, pp. 767–783

    2018

  16. [16]

    H. Li, X. Meng, X . Zuo, Z. Liu, H. Wang, and D. Cremers, ―PG-SLAM: Photorealistic and geometry -aware RGB-D SLAM in dynamic environments,‖ IEEE Trans. Robot. , vol. 41, pp. 6084–6101, 2025

    2025

  17. [17]

    Schöps, T

    T. Schöps, T. Sattler, and M. Pollefeys, ―SurfelMeshing: Online surfel-based mesh reconstruction,‖ IEEE Trans. Pattern Anal. Mach. Intell., vol. 42, no. 10, pp. 2494–2507, 2020

    2020

  18. [18]

    Whelan, S

    T. Whelan, S. Leutenegger, R. F. Salas-Moreno, B. Glocker, and A. J. Davison, ―ElasticFusion: Dense SLAM without a pose graph,‖ in Proc. Robotics: Sci. Syst. (RSS), Rome, Italy, 2015

    2015

  19. [19]

    Zhu et al., ―NICE -SLAM: Neural implicit scalable encoding for SLAM,‖ in Proc

    Z. Zhu et al., ―NICE -SLAM: Neural implicit scalable encoding for SLAM,‖ in Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit. (CVPR) , New Orleans, LA, USA, 2022, pp. 12776 – 12786

    2022

  20. [20]

    Mescheder, M

    L. Mescheder, M. Oechsle , M. Niemeyer, S. Nowozin, and A. Geiger, ―Occupancy networks: Learning 3D reconstruction in function space,‖ in Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit. (CVPR), Long Beach, CA, USA, 2019, pp. 4455–4465

    2019

  21. [21]

    Schmied, T

    A. Schmied, T. Fischer, M. Danelljan, M. Pollefeys, and F. Yu, ―R3D3: Dense 3D reconstruction of dynamic scenes from multiple cameras,‖ in Proc. IEEE/CVF Int. Conf. Comput. Vis. (ICCV), Paris, France, 2023, pp. 3193–3203

    2023

  22. [22]

    Mildenhall, P

    B. Mildenhall, P. P. Srinivasan, M. Tancik, J. T. Barron, R. Ramamoorthi, and R. Ng, ―NeRF: Representing scenes as neural radiance fields for view synthesis,‖ in Proc. Eur. Conf. Comput. Vis. (ECCV), 2020, pp. 405–421

    2020

  23. [23]

    Kerbl, G

    B. Kerbl, G. Kopanas, T. Leimkuehler, and G. Drettakis , ―3D Gaussian splatting for real -time radiance field rendering,‖ ACM Trans. Graph., vol. 42, no. 4, pp. 1–14, 2023

    2023

  24. [24]

    R. A. Newcombe, D. Fox, and S. M. Seitz, ―DynamicFusion: Reconstruction and tracking of non-rigid scenes in real-time,‖ in Proc. IEEE Conf. Comput. Vis. Pattern Recognit. (CVPR) , Boston, MA, USA, 2015, pp. 343–352

    2015

  25. [25]

    Reijgwart, A

    V. Reijgwart, A. Millane, H. Oleynikova, R. Siegwart, C. Cadena, and J. Nieto, ―Voxgraph: Globally consistent, volumetric mapping using signed distance function submaps,‖ IEEE Robot. Autom. Lett., vol. 5, no. 1, pp. 227–234, 2020

    2020

  26. [26]

    A. Dai, M. Nieß ner, M. Zollhö fer, S. Izadi, and C. Theobalt, ―BundleFusion: Real-time globally consistent 3D reconstruction using on-the-fly surface re-integration,‖ ACM Trans. Graph., vol. 36, 2017, Art. no. 168

    2017

  27. [27]

    Su et al., ―RobustFusion: Robust volumetric performance reconstruction under human -object interactions from monocular RGBD stream,‖ IEEE Trans

    Z. Su et al., ―RobustFusion: Robust volumetric performance reconstruction under human -object interactions from monocular RGBD stream,‖ IEEE Trans. Pattern Anal. Mach. Intell., vol. 45, no. 5, pp. 6196–6213, 2023

    2023

  28. [28]

    Endres, J

    F. Endres, J. Hess, J. Sturm, D. C remers, and W. Burgard, ―3 -D mapping with an RGB -D camera,‖ IEEE Trans. Robot., vol. 30, no. 1, pp. 177–187, 2014

    2014

  29. [29]

    Liu et al., ―WorldMirror: Universal 3D world reconstruction with any -prior prompting,‖ arXiv preprint arXiv:2510.10726, 2025

    Y. Liu et al., ―WorldMirror: Universal 3D world reconstruction with any -prior prompting,‖ arXiv preprint arXiv:2510.10726, 2025

    arXiv 2025

  30. [30]

    W. Guo, B. Wang, and L. Chen, ―NeuV -SLAM: Fast neural multiresolution voxel optimization for RGBD dense SLAM,‖ IEEE Trans. Multimedia, vol. 27, pp. 7546–7546, 2025

    2025

  31. [31]

    Y. Xie, H. Xiao, and W. Kang, ―Tri2plane: Advancing neural implicit surface reconstruction for indoor scenes,‖ IEEE Trans. Multimedia, vol. 27, pp. 4910–4923, 2025

    2025

  32. [32]

    Y. Xiao, T. Zhang, J. Cao, and Z. Chen, ―Accelerated Lloyd’s method for resampling 3D point clouds,‖ IEEE Trans. Multimedia, vol. 27, pp. 1033–1046, 2025

    2025

  33. [33]

    F. Wang, S. Galliani, C. Vogel, P. Speciale, and M. Pollefeys, ―PatchmatchNet: Learned multi -view patchmatch stereo,‖ in Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit. (CVPR) , Nashville, TN, USA, 2021, pp. 14189–14198

    2021

  34. [34]

    J. L. Schö nberger and J. -M. Frahm, ―Structure -from-motion revisited,‖ in Proc. IEEE Conf. Comput. Vis. Pattern Recognit. (CVPR), Las Vegas, NV, USA, 2016, pp. 4104–4113

    2016

  35. [35]

    Zhang, N

    T. Zhang, N. Li, G. Gong, B. Tian, and X. Lin, ―LOD-InfiniTAM: Adaptive depth sampling for accurate RGB -D SLAM,‖ in Proc. 7th Int. Conf. Robot. Comput. Vis. (ICRCV), Hong Kong, China, 2025, pp. 243–247

    2025