pith. sign in

arxiv: 2604.02996 · v2 · submitted 2026-04-03 · 💻 cs.CV

Rendering Multi-Human and Multi-Object with 3D Gaussian Splatting

Weiquan Wang , Jun Xiao , Feifei Shao , Yi Yang , Yueting Zhuang , Long Chen This is my paper

Pith reviewed 2026-05-13 20:35 UTC · model grok-4.3

classification 💻 cs.CV
keywords 3D Gaussian SplattingMulti-Human Multi-ObjectSparse-View ReconstructionScene GraphInstance InteractionDynamic Scene RenderingOcclusion HandlingDigital Twins
0
0 comments X

The pith

A hierarchical 3D Gaussian Splatting framework reconstructs multi-human multi-object dynamic scenes from sparse views by fusing per-instance data and modeling interactions on a scene graph.

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

The paper targets the reconstruction of dynamic scenes containing multiple humans and objects that interact, using inputs from only a few camera views. Severe mutual occlusions make it hard to keep each instance consistent across angles, while their contacts create complex dependencies that standard methods miss. MM-GS solves this with a Per-Instance Multi-View Fusion module that pools information across views for each entity, then a Scene-Level Instance Interaction module that uses a global scene graph to adjust attributes and capture contact effects. A reader would care because the result supports accurate digital models of crowded real-world situations for robotics, VR, and AR without dense camera rigs or manual 3D supervision.

Core claim

MM-GS is a hierarchical framework built on 3D Gaussian Splatting. It first uses Per-Instance Multi-View Fusion to aggregate visual information across all available views and create robust consistent representations for each instance despite occlusion. It then applies a Scene-Level Instance Interaction module on a global scene graph to reason about relationships among all participants and refine their attributes to capture interaction effects, producing state-of-the-art high-fidelity details and plausible inter-instance contacts on challenging datasets.

What carries the argument

The MM-GS hierarchical framework with its Per-Instance Multi-View Fusion module for aggregating cross-view information into consistent per-instance representations and its Scene-Level Instance Interaction module that operates on a global scene graph to refine attributes and model combinatorial interaction dependencies.

Load-bearing premise

The Per-Instance Multi-View Fusion and Scene-Level Instance Interaction modules can reliably resolve severe mutual occlusions and combinatorial interaction effects from sparse views without additional supervision or explicit 3D priors.

What would settle it

A sparse-view capture of heavily overlapping and contacting humans and objects where the output shows view-inconsistent instance details or physically implausible contacts would show that the two modules do not handle the stated challenges.

Figures

Figures reproduced from arXiv: 2604.02996 by Feifei Shao, Jun Xiao, Long Chen, Weiquan Wang, Yi Yang, Yueting Zhuang.

Figure 1
Figure 1. Figure 1: Core challenges in Multi-Human Multi-Object (MHMO) rendering. From sparse views, rendering complex interactions involves overcoming two key challenges: ensuring cross-view consistency under severe occlusion (top) and modeling the mutual influence between instances at contact regions (bottom). Our MM-GS is designed to address both. [18]. A critical bottleneck, however, remains in faithfully capturing and re… view at source ↗
Figure 2
Figure 2. Figure 2: Overview of MM-GS pipeline. Our method refines initial 3D Gaussian representations through three main stages. (a) Human-Object Deformation: We initialize the scene by deforming canonical human and object models to their target poses and representing them as collections of 3D Gaussians. (b) Per-Instance Multi-View Fusion: A Cross-View Fusion network refines each instance’s appearance and local geometry by a… view at source ↗
Figure 3
Figure 3. Figure 3: Qualitative comparison on the HOI-M3 dataset. We highlight specific regions with colored dashed circles to illustrate the differences. Note that our MM-GS generates significantly sharper details and more plausible contact regions. In contrast, the NeRF-based NeuralHOIFVV-MM tends to produce overly smooth or blurry results, while the 3DGS-based GTU-MM suffers from floating artifacts and geometric inconsiste… view at source ↗
Figure 4
Figure 4. Figure 4: Qualitative results of our ablation study. Removing both modules (w/o Both) leads to blurry results. Adding the View Fusion module (+ View Fusion) significantly improves sharpness. Further incorporating the Interaction network (+ View Fusion + Interaction) resolves ambiguities at contact regions, resulting in cleaner boundaries. baseline rendering is noticeably blurry and lacks detail. After incorporating … view at source ↗
read the original abstract

Reconstructing dynamic scenes with multiple interacting humans and objects from sparse-view inputs is a critical yet challenging task, essential for creating high-fidelity digital twins for robotics and VR/AR. This problem, which we term Multi-Human Multi-Object (MHMO) rendering, presents two significant obstacles: achieving view-consistent representations for individual instances under severe mutual occlusion, and explicitly modeling the complex and combinatorial dependencies that arise from their interactions. To overcome these challenges, we propose MM-GS, a novel hierarchical framework built upon 3D Gaussian Splatting. Our method first employs a Per-Instance Multi-View Fusion module to establish a robust and consistent representation for each instance by aggregating visual information across all available views. Subsequently, a Scene-Level Instance Interaction module operates on a global scene graph to reason about relationships between all participants, refining their attributes to capture subtle interaction effects. Extensive experiments on challenging datasets demonstrate that our method significantly outperforms strong baselines, producing state-of-the-art results with high-fidelity details and plausible inter-instance contacts.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 2 minor

Summary. The manuscript presents MM-GS, a hierarchical framework extending 3D Gaussian Splatting for reconstructing and rendering dynamic multi-human multi-object (MHMO) scenes from sparse-view inputs. It first applies a Per-Instance Multi-View Fusion module to aggregate multi-view information into consistent per-instance Gaussian representations, then uses a Scene-Level Instance Interaction module on a global scene graph to model and refine inter-instance relationships and capture interaction effects. The work claims state-of-the-art performance on challenging datasets, with high-fidelity details and plausible inter-instance contacts.

Significance. If the empirical claims hold under rigorous validation, the contribution would be significant for dynamic scene reconstruction tasks in robotics, VR/AR, and digital twin applications. The hierarchical separation of per-instance consistency from scene-level interaction modeling directly targets two core obstacles (severe mutual occlusion and combinatorial dependencies) that standard 3DGS extensions have not fully resolved, offering a practical, additive framework without additional supervision.

major comments (3)
  1. [Abstract] Abstract: The central claims that the method 'significantly outperforms strong baselines' and produces 'state-of-the-art results with ... plausible inter-instance contacts' are stated without any quantitative metrics (e.g., PSNR, SSIM, contact error), baseline specifications, ablation results, or error analysis. This absence makes it impossible to verify whether the data support the claims or whether improvements can be attributed to the proposed modules.
  2. [§3.1] §3.1 (Per-Instance Multi-View Fusion): The module is described as 'aggregating visual information across all available views' with no additional supervision or explicit 3D priors. In standard 3DGS optimization, occluded regions receive only indirect photometric gradients; without depth, silhouette, or correspondence constraints, the optimization can assign arbitrary positions/scales/opacities to unobserved Gaussians. This directly risks view-inconsistent representations that would propagate errors into the Scene-Level Instance Interaction module, undermining the claim of plausible contacts.
  3. [§4] §4 (Experiments): No ablation isolating the Per-Instance Multi-View Fusion under controlled occlusion levels (e.g., varying numbers of mutually occluding instances) is reported, nor are qualitative visualizations or quantitative metrics for occluded-region fidelity provided. Without these, the load-bearing assumption that the fusion reliably resolves combinatorial occlusion cases cannot be evaluated, weakening attribution of any SOTA results to the hierarchical design.
minor comments (2)
  1. [§3.2] The notation and variable definitions for the scene graph (e.g., node/edge attributes, refinement operations) should be introduced with a clear table or equation block in §3.2 to aid readability.
  2. [Figures] Figure captions for qualitative results should explicitly label which method corresponds to each column and note the view count and occlusion severity for each example.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive and detailed feedback. We address each major comment point-by-point below. Where revisions are warranted, we have updated the manuscript to strengthen the presentation of results and clarify methodological details.

read point-by-point responses

  1. Referee: [Abstract] The central claims that the method 'significantly outperforms strong baselines' and produces 'state-of-the-art results with ... plausible inter-instance contacts' are stated without any quantitative metrics (e.g., PSNR, SSIM, contact error), baseline specifications, ablation results, or error analysis. This absence makes it impossible to verify whether the data support the claims or whether improvements can be attributed to the proposed modules.

    Authors: The abstract is intentionally concise as a high-level overview. All quantitative metrics (PSNR, SSIM, LPIPS, contact error), baseline names, and ablation results are reported in full in Section 4 and the supplementary material. To better anchor the claims, we have revised the abstract to include specific quantitative highlights, such as the average PSNR improvement of 1.8 dB over the strongest baseline. revision: yes

  2. Referee: [§3.1] §3.1 (Per-Instance Multi-View Fusion): The module is described as 'aggregating visual information across all available views' with no additional supervision or explicit 3D priors. In standard 3DGS optimization, occluded regions receive only indirect photometric gradients; without depth, silhouette, or correspondence constraints, the optimization can assign arbitrary positions/scales/opacities to unobserved Gaussians. This directly risks view-inconsistent representations that would propagate errors into the Scene-Level Instance Interaction module, undermining the claim of plausible contacts.

    Authors: The Per-Instance Multi-View Fusion jointly optimizes each instance's Gaussians against photometric losses from every available view. Overlapping visible regions across views provide direct gradient signals that constrain the optimization of partially occluded Gaussians, preventing arbitrary assignments. We have expanded §3.1 with a paragraph detailing this multi-view gradient flow and added supplementary visualizations demonstrating view-consistent geometry in heavily occluded areas. revision: partial

  3. Referee: [§4] §4 (Experiments): No ablation isolating the Per-Instance Multi-View Fusion under controlled occlusion levels (e.g., varying numbers of mutually occluding instances) is reported, nor are qualitative visualizations or quantitative metrics for occluded-region fidelity provided. Without these, the load-bearing assumption that the fusion reliably resolves combinatorial occlusion cases cannot be evaluated, weakening attribution of any SOTA results to the hierarchical design.

    Authors: We agree that a controlled occlusion ablation strengthens attribution. We have added new experiments in the revised Section 4 that vary the number of mutually occluding instances from 2 to 5 while keeping total scene complexity fixed. We report masked PSNR on occluded regions and include qualitative renderings of contact areas. These results show consistent gains attributable to the fusion module. revision: yes

Circularity Check

0 steps flagged

No significant circularity; method is additive framework on existing 3DGS

full rationale

The paper presents MM-GS as a hierarchical extension of 3D Gaussian Splatting using two modules: Per-Instance Multi-View Fusion (aggregating views for consistent instance representations) and Scene-Level Instance Interaction (reasoning via scene graph). No equations, derivations, or fitted parameters are described that reduce any prediction to its own inputs by construction. Claims rest on empirical outperformance on datasets rather than self-definitional or self-cited uniqueness theorems. No load-bearing self-citations or ansatz smuggling appear in the provided text; the approach is presented as a practical combination of existing techniques with new modules whose validity is tested externally.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract supplies no explicit free parameters, axioms, or invented entities; the method is described only at the level of two high-level modules built on standard 3D Gaussian Splatting.

pith-pipeline@v0.9.0 · 5487 in / 1150 out tokens · 40531 ms · 2026-05-13T20:35:17.582221+00:00 · methodology

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

60 extracted references · 60 canonical work pages · 1 internal anchor

  1. [1]

    Navigation for human-robot interaction tasks,

    P. Althaus, H. Ishiguro, T. Kanda, T. Miyashita, and H. I. Christensen, “Navigation for human-robot interaction tasks,” inICRA, 2004

    work page 2004

  2. [2]

    Behave: Dataset and method for tracking human object interactions,

    B. L. Bhatnagar, X. Xie, I. A. Petrov, C. Sminchisescu, C. Theobalt, and G. Pons-Moll, “Behave: Dataset and method for tracking human object interactions,” inCVPR, 2022

    work page 2022

  3. [3]

    Human-robot perception in industrial environments: A survey,

    A. Bonci, P. D. Cen Cheng, M. Indri, G. Nabissi, and F. Sibona, “Human-robot perception in industrial environments: A survey,”Sen- sors, 2021

    work page 2021

  4. [4]

    Human-in-the-loop robot learning for smart manufacturing: A human-centric perspective,

    H. Chen, S. Li, J. Fan, A. Duan, C. Yang, D. Navarro-Alarcon, and P. Zheng, “Human-in-the-loop robot learning for smart manufacturing: A human-centric perspective,”IEEE TASE, 2025

    work page 2025

  5. [5]

    H-rssg: High-fidelity robotic surgical scene generation with implicit deformable neural radiance field,

    Q. Chen, K. Qian, Z. Hu, Y . Tai, and Z. Yu, “H-rssg: High-fidelity robotic surgical scene generation with implicit deformable neural radiance field,”IEEE TASE, 2025

    work page 2025

  6. [6]

    A multimode navigation system for an assistive robotics project,

    A. Cherubini, G. Oriolo, F. Macr ´ı, F. Aloise, F. Cincotti, and D. Mattia, “A multimode navigation system for an assistive robotics project,” Autonomous Robots, vol. 25, no. 4, pp. 383–404, 2008

    work page 2008

  7. [7]

    High-quality streamable free- viewpoint video,

    A. Collet, M. Chuang, P. Sweeney, D. Gillett, D. Evseev, D. Calabrese, H. Hoppe, A. Kirk, and S. Sullivan, “High-quality streamable free- viewpoint video,”ACM TOG, 2015

    work page 2015

  8. [8]

    Motion2fusion: Real-time volu- metric performance capture,

    M. Dou, P. Davidson, S. R. Fanello, S. Khamis, A. Kowdle, C. Rhe- mann, V . Tankovich, and S. Izadi, “Motion2fusion: Real-time volu- metric performance capture,”ACM TOG, 2017

    work page 2017

  9. [9]

    Mps-nerf: Generalizable 3d human rendering from multiview images,

    X. Gao, J. Yang, J. Kim, S. Peng, Z. Liu, and X. Tong, “Mps-nerf: Generalizable 3d human rendering from multiview images,”IEEE TPAMI, 2022

    work page 2022

  10. [10]

    Romeo: Revisiting optimization methods for reconstructing 3d human-object interaction models from images,

    A. Gavryushin, Y . Liu, D. Huang, Y .-L. Kuo, J. Valentin, L. Van Gool, O. Hilliges, and X. Wang, “Romeo: Revisiting optimization methods for reconstructing 3d human-object interaction models from images,” inECCV, 2024

    work page 2024

  11. [11]

    Sherf: Generalizable human nerf from a single image,

    S. Hu, F. Hong, L. Pan, H. Mei, L. Yang, and Z. Liu, “Sherf: Generalizable human nerf from a single image,” inICCV, 2023

    work page 2023

  12. [12]

    Arch: Animatable reconstruction of clothed humans,

    Z. Huang, Y . Xu, C. Lassner, H. Li, and T. Tung, “Arch: Animatable reconstruction of clothed humans,” inCVPR, 2020

    work page 2020

  13. [13]

    Neuralhofusion: Neural volumetric rendering under human-object interactions,

    Y . Jiang, S. Jiang, G. Sun, Z. Su, K. Guo, M. Wu, J. Yu, and L. Xu, “Neuralhofusion: Neural volumetric rendering under human-object interactions,” inCVPR, 2022

    work page 2022

  14. [14]

    Hifi4g: High-fidelity human performance rendering via compact gaussian splatting,

    Y . Jiang, Z. Shen, P. Wang, Z. Su, Y . Hong, Y . Zhang, J. Yu, and L. Xu, “Hifi4g: High-fidelity human performance rendering via compact gaussian splatting,” inCVPR, 2024

    work page 2024

  15. [15]

    Instant-nvr: Instant neural volumetric rendering for human-object interactions from monocular rgbd stream,

    Y . Jiang, K. Yao, Z. Su, Z. Shen, H. Luo, and L. Xu, “Instant-nvr: Instant neural volumetric rendering for human-object interactions from monocular rgbd stream,” inCVPR, 2023

    work page 2023

  16. [16]

    Transferring policy of deep reinforcement learning from simulation to reality for robotics,

    H. Ju, R. Juan, R. Gomez, K. Nakamura, and G. Li, “Transferring policy of deep reinforcement learning from simulation to reality for robotics,”NMI, 2022

    work page 2022

  17. [17]

    A compact dynamic 3d gaussian representation for real-time dynamic view synthesis,

    K. Katsumata, D. M. V o, and H. Nakayama, “A compact dynamic 3d gaussian representation for real-time dynamic view synthesis,” in ECCV, 2024

    work page 2024

  18. [18]

    Interact: Trans- former models for human intent prediction conditioned on robot actions,

    K. Kedia, A. Bhardwaj, P. Dan, and S. Choudhury, “Interact: Trans- former models for human intent prediction conditioned on robot actions,” inICRA, 2024

    work page 2024

  19. [19]

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

    B. Kerbl, G. Kopanas, T. Leimkuehler, and G. Drettakis, “3d gaussian splatting for real-time radiance field rendering,”ACM TOG, 2023

    work page 2023

  20. [20]

    Human-centered robot navigation—towards a harmoniously human–robot coexisting environment,

    C.-P. Lam, C.-T. Chou, K.-H. Chiang, and L.-C. Fu, “Human-centered robot navigation—towards a harmoniously human–robot coexisting environment,”IEEE Transactions on Robotics, 2010

    work page 2010

  21. [21]

    Guess the unseen: Dynamic 3d scene reconstruction from partial 2d glimpses,

    I. Lee, B. Kim, and H. Joo, “Guess the unseen: Dynamic 3d scene reconstruction from partial 2d glimpses,” inCVPR, 2024

    work page 2024

  22. [22]

    Uncer- tainty guided policy for active robotic 3d reconstruction using neural radiance fields,

    S. Lee, L. Chen, J. Wang, A. Liniger, S. Kumar, and F. Yu, “Uncer- tainty guided policy for active robotic 3d reconstruction using neural radiance fields,”IEEE RAL, 2022

    work page 2022

  23. [23]

    Deformnet: Latent space modeling and dynamics prediction for deformable object manipulation,

    C. Li, Z. Ai, T. Wu, X. Li, W. Ding, and H. Xu, “Deformnet: Latent space modeling and dynamics prediction for deformable object manipulation,” inICRA, 2024

    work page 2024

  24. [24]

    Gp-nerf: Generalized perception nerf for context-aware 3d scene understanding,

    H. Li, D. Zhang, Y . Dai, N. Liu, L. Cheng, J. Li, J. Wang, and J. Han, “Gp-nerf: Generalized perception nerf for context-aware 3d scene understanding,” inCVPR, 2024

    work page 2024

  25. [25]

    Gaufre: Gaussian deformation fields for real-time dynamic novel view synthesis,

    Y . Liang, N. Khan, Z. Li, T. Nguyen-Phuoc, D. Lanman, J. Tompkin, and L. Xiao, “Gaufre: Gaussian deformation fields for real-time dynamic novel view synthesis,” inWACV, 2025

    work page 2025

  26. [26]

    Learning implicit templates for point-based clothed human modeling,

    S. Lin, H. Zhang, Z. Zheng, R. Shao, and Y . Liu, “Learning implicit templates for point-based clothed human modeling,” inECCV, 2022

    work page 2022

  27. [27]

    Hosnerf: Dynamic human-object-scene neural radiance fields from a single video,

    J.-W. Liu, Y .-P. Cao, T. Yang, Z. Xu, J. Keppo, Y . Shan, X. Qie, and M. Z. Shou, “Hosnerf: Dynamic human-object-scene neural radiance fields from a single video,” inICCV, 2023

    work page 2023

  28. [28]

    Humangaussian: Text-driven 3d human generation with gaussian splatting,

    X. Liu, X. Zhan, J. Tang, Y . Shan, G. Zeng, D. Lin, X. Liu, and Z. Liu, “Humangaussian: Text-driven 3d human generation with gaussian splatting,” inCVPR, 2024

    work page 2024

  29. [29]

    Citygaussian: Real-time high-quality large-scale scene rendering with gaussians,

    Y . Liu, C. Luo, L. Fan, N. Wang, J. Peng, and Z. Zhang, “Citygaussian: Real-time high-quality large-scale scene rendering with gaussians,” in ECCV, 2025

    work page 2025

  30. [30]

    Core4d: A 4d human-object-human interaction dataset for collaborative object rearrangement,

    Y . Liu, C. Zhang, R. Xing, B. Tang, B. Yang, and L. Yi, “Core4d: A 4d human-object-human interaction dataset for collaborative object rearrangement,” inCVPR, 2025

    work page 2025

  31. [31]

    Smpl: a skinned multi-person linear model,

    M. Loper, N. Mahmood, J. Romero, G. Pons-Moll, and M. J. Black, “Smpl: a skinned multi-person linear model,”ACM TOG, 2015

    work page 2015

  32. [32]

    Himo: A new benchmark for full-body human interacting with multiple objects,

    X. Lv, L. Xu, Y . Yan, X. Jin, C. Xu, S. Wu, Y . Liu, L. Li, M. Bi, W. Zeng,et al., “Himo: A new benchmark for full-body human interacting with multiple objects,” inECCV, 2024

    work page 2024

  33. [33]

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

    B. Mildenhall, P. P. Srinivasan, M. Tancik, J. T. Barron, R. Ramamoor- thi, and R. Ng, “Nerf: Representing scenes as neural radiance fields for view synthesis,”Communications of the ACM, pp. 99–106, 2021

    work page 2021

  34. [34]

    Nerfies: Deformable neural radiance fields,

    K. Park, U. Sinha, J. T. Barron, S. Bouaziz, D. B. Goldman, S. M. Seitz, and R. Martin-Brualla, “Nerfies: Deformable neural radiance fields,” inICCV, 2021

    work page 2021

  35. [35]

    Animatable neural radiance fields for modeling dynamic human bodies,

    S. Peng, J. Dong, Q. Wang, S. Zhang, Q. Shuai, X. Zhou, and H. Bao, “Animatable neural radiance fields for modeling dynamic human bodies,” inICCV, 2021

    work page 2021

  36. [36]

    D- nerf: Neural radiance fields for dynamic scenes,

    A. Pumarola, E. Corona, G. Pons-Moll, and F. Moreno-Noguer, “D- nerf: Neural radiance fields for dynamic scenes,” inCVPR, 2021

    work page 2021

  37. [37]

    3dgs- avatar: Animatable avatars via deformable 3d gaussian splatting,

    Z. Qian, S. Wang, M. Mihajlovic, A. Geiger, and S. Tang, “3dgs- avatar: Animatable avatars via deformable 3d gaussian splatting,” in CVPR, 2024

    work page 2024

  38. [38]

    Path planning for autonomous mobile robots: A review,

    J. R. S ´anchez-Ib´a˜nez, C. J. P ´erez-del Pulgar, and A. Garc ´ıa-Cerezo, “Path planning for autonomous mobile robots: A review,”Sensors, 2021

    work page 2021

  39. [39]

    Cooperative navigation for mixed human–robot teams using haptic feedback,

    S. Scheggi, M. Aggravi, and D. Prattichizzo, “Cooperative navigation for mixed human–robot teams using haptic feedback,”IEEE Transac- tions on Human-Machine Systems, vol. 47, no. 4, pp. 462–473, 2016

    work page 2016

  40. [40]

    Structure-from-motion revisited,

    J. L. Schonberger and J.-M. Frahm, “Structure-from-motion revisited,” inCVPR, 2016, pp. 4104–4113

    work page 2016

  41. [41]

    Modeling ambient scene dynamics for free-view synthesis,

    M.-L. Shih, J.-B. Huang, C. Kim, R. Shah, J. Kopf, and C. Gao, “Modeling ambient scene dynamics for free-view synthesis,” inACM SIGGRAPH, 2024

    work page 2024

  42. [42]

    Robustfusion: Robust volumetric performance reconstruction under human-object interactions from monocular rgbd stream,

    Z. Su, L. Xu, D. Zhong, Z. Li, F. Deng, S. Quan, and L. Fang, “Robustfusion: Robust volumetric performance reconstruction under human-object interactions from monocular rgbd stream,”IEEE TPAMI, 2022

    work page 2022

  43. [43]

    Neural free-viewpoint performance rendering under complex human-object interactions,

    G. Sun, X. Chen, Y . Chen, A. Pang, P. Lin, Y . Jiang, L. Xu, J. Yu, and J. Wang, “Neural free-viewpoint performance rendering under complex human-object interactions,” inACM MM, 2021

    work page 2021

  44. [44]

    Graph attention networks,

    P. Veli ˇckovi´c, G. Cucurull, A. Casanova, A. Romero, P. Li `o, and Y . Bengio, “Graph attention networks,” inICLR, 2018

    work page 2018

  45. [45]

    Multimodal human– robot interaction for human-centric smart manufacturing: a survey,

    T. Wang, P. Zheng, S. Li, and L. Wang, “Multimodal human– robot interaction for human-centric smart manufacturing: a survey,” Advanced Intelligent Systems, 2024

    work page 2024

  46. [46]

    Physically Plausible Human-Object Rendering from Sparse Views via 3D Gaussian Splatting

    W. Wang, J. Xiao, Y . Zhuang, and L. Chen, “Physics-aware human- object rendering from sparse views via 3d gaussian splatting,”arXiv preprint arXiv:2503.09640, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  47. [47]

    Image quality assessment: from error visibility to structural similarity,

    Z. Wang, A. C. Bovik, H. R. Sheikh, and E. P. Simoncelli, “Image quality assessment: from error visibility to structural similarity,”IEEE TIP, 2004

    work page 2004

  48. [48]

    Humannerf: Free-viewpoint rendering of moving people from monocular video,

    C.-Y . Weng, B. Curless, P. P. Srinivasan, J. T. Barron, and I. Kemelmacher-Shlizerman, “Humannerf: Free-viewpoint rendering of moving people from monocular video,” inCVPR, 2022

    work page 2022

  49. [49]

    Space-time neural irradiance fields for free-viewpoint video,

    W. Xian, J.-B. Huang, J. Kopf, and C. Kim, “Space-time neural irradiance fields for free-viewpoint video,” inCVPR, 2021

    work page 2021

  50. [50]

    An assistive navigation framework for the visually impaired,

    J. Xiao, S. L. Joseph, X. Zhang, B. Li, X. Li, and J. Zhang, “An assistive navigation framework for the visually impaired,”IEEE transactions on human-machine systems, vol. 45, no. 5, pp. 635–640, 2015

    work page 2015

  51. [51]

    Visibility aware human- object interaction tracking from single rgb camera,

    X. Xie, B. L. Bhatnagar, and G. Pons-Moll, “Visibility aware human- object interaction tracking from single rgb camera,” inCVPR, 2023

    work page 2023

  52. [52]

    Nerf-ds: Neural radiance fields for dynamic specular objects,

    Z. Yan, C. Li, and G. H. Lee, “Nerf-ds: Neural radiance fields for dynamic specular objects,” inCVPR, 2023

    work page 2023

  53. [53]

    Cpf: Learning a contact potential field to model the hand-object interaction,

    L. Yang, X. Zhan, K. Li, W. Xu, J. Li, and C. Lu, “Cpf: Learning a contact potential field to model the hand-object interaction,” inICCV, 2021

    work page 2021

  54. [54]

    Cor- gs: sparse-view 3d gaussian splatting via co-regularization,

    J. Zhang, J. Li, X. Yu, L. Huang, L. Gu, J. Zheng, and X. Bai, “Cor- gs: sparse-view 3d gaussian splatting via co-regularization,” inECCV, 2024

    work page 2024

  55. [55]

    Neuraldome: A neural modeling pipeline on multi-view human-object interactions,

    J. Zhang, H. Luo, H. Yang, X. Xu, Q. Wu, Y . Shi, J. Yu, L. Xu, and J. Wang, “Neuraldome: A neural modeling pipeline on multi-view human-object interactions,” inCVPR, 2023

    work page 2023

  56. [56]

    Hoi-mˆ 3: Capture multiple humans and objects interaction within contextual environment,

    J. Zhang, J. Zhang, Z. Song, Z. Shi, C. Zhao, Y . Shi, J. Yu, L. Xu, and J. Wang, “Hoi-mˆ 3: Capture multiple humans and objects interaction within contextual environment,” inCVPR, 2024

    work page 2024

  57. [57]

    The unreasonable effectiveness of deep features as a perceptual metric,

    R. Zhang, P. Isola, A. A. Efros, E. Shechtman, and O. Wang, “The unreasonable effectiveness of deep features as a perceptual metric,” in CVPR, 2018

    work page 2018

  58. [58]

    Nerf in the palm of your hand: Corrective augmentation for robotics via novel- view synthesis,

    A. Zhou, M. J. Kim, L. Wang, P. Florence, and C. Finn, “Nerf in the palm of your hand: Corrective augmentation for robotics via novel- view synthesis,” inCVPR, 2023

    work page 2023

  59. [59]

    The nerfect match: Exploring nerf features for visual localization,

    Q. Zhou, M. Maximov, O. Litany, and L. Leal-Taix ´e, “The nerfect match: Exploring nerf features for visual localization,” inECCV, 2025

    work page 2025

  60. [60]

    Fsgs: Real-time few-shot view synthesis using gaussian splatting,

    Z. Zhu, Z. Fan, Y . Jiang, and Z. Wang, “Fsgs: Real-time few-shot view synthesis using gaussian splatting,” inECCV, 2025

    work page 2025