pith. sign in

arxiv: 2606.30014 · v1 · pith:RSYBAL7Onew · submitted 2026-06-29 · 💻 cs.CV

Shell-Supervised Gaussian Splatting for Urban Real-to-Sim Reconstruction

Yuan Yang , Peijun Lu , Fangzhou Lu , Sai Fan , Siqi Yan , Chenyuan Zhang , Haobo Liang , Yichen Wang This is my paper

Pith reviewed 2026-06-30 06:39 UTC · model grok-4.3

classification 💻 cs.CV
keywords Gaussian SplattingUrban ReconstructionReal-to-SimGeometric SupervisionFacade ReconstructionNovel View SynthesisEmbodied AISurface Regularization
0
0 comments X

The pith

An external facade shell supplies geometric supervision to stabilize Gaussian Splatting on urban video sequences.

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

The paper shows how an aligned exterior structural shell can be rendered into per-view depth, normal, and mask maps that are then used as mask-gated losses during Gaussian optimization. This regularizes only the visible shell-supported regions while leaving RGB-driven appearance optimization untouched. The resulting surfaces exhibit improved orientation and point-cloud consistency on close-range urban facades compared with photo-only or monocular-cue baselines. A reader would care because embodied-AI simulation requires stable geometry for collision and navigation, not merely plausible novel views.

Core claim

Shell-supervised Gaussian Splatting aligns an exterior facade shell to the video reconstruction frame, renders per-view depth, camera-space normal, and valid-mask maps, and applies these cues through mask-gated losses during Gaussian optimization. This design preserves RGB-driven appearance while regularizing only visible shell-supported facade regions, yielding improved facade orientation and visible-surface point-cloud consistency over photo-only, monocular-cue, and surface-oriented Gaussian baselines while maintaining comparable held-out rendering quality.

What carries the argument

The external facade structural shell rendered into per-view depth, normal, and valid-mask maps that drive mask-gated losses during Gaussian optimization.

If this is right

  • Geometry suitable for collision and navigation reasoning becomes available directly from video without post-processing surface fitting.
  • Visible-surface point clouds become more consistent across views, reducing artifacts in agent-environment interaction tests.
  • Held-out novel-view rendering quality remains comparable to unsupervised baselines, so appearance fidelity is not traded for geometry.
  • The supervision applies selectively through valid masks, avoiding over-regularization on non-facade or occluded regions.

Where Pith is reading between the lines

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

  • The same shell-supervision pattern could be tested on indoor corridors or vehicle exteriors where CAD shells are already available.
  • If shell alignment can be updated from new scans, the method might support incremental reconstruction of changing urban environments.
  • Downstream simulation tasks such as path planning could be run on the output point clouds to quantify whether the measured orientation gains translate into fewer collision failures.

Load-bearing premise

The external facade structural shell can be accurately aligned to the video reconstruction frame and supplies reliable geometric cues only for visible shell-supported regions without introducing systematic bias.

What would settle it

If the measured improvement in facade orientation and point-cloud consistency disappears when the shell is removed or when its alignment is deliberately offset by a few centimeters on the same test scenes, the claim that shell supervision is the source of the geometric gain is falsified.

Figures

Figures reproduced from arXiv: 2606.30014 by Chenyuan Zhang, Fangzhou Lu, Haobo Liang, Peijun Lu, Sai Fan, Siqi Yan, Yichen Wang, Yuan Yang.

Figure 1
Figure 1. Figure 1: Overview of the proposed shell-supervised Gaussian reconstruction framework. A monocular facade video is first reconstructed [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Representative shell supervision on the primary scene: [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Qualitative geometry comparison on a held-out test view of the primary scene. The first column provides the RGB context [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figure 2
Figure 2. Figure 2: Secondary-scene qualitative geometry comparison on a [PITH_FULL_IMAGE:figures/full_fig_p012_2.png] view at source ↗
read the original abstract

Real-to-sim reconstruction for embodied AI requires geometry that is useful for collision reasoning, navigation, and agent-environment interaction, not only photorealistic novel-view synthesis. However, close-range urban facades are difficult for video-to-3D reconstruction: glass, reflections, repeated windows, and weak texture can produce visually plausible renderings with unstable surface geometry. We introduce shell-supervised Gaussian Splatting, a reconstruction-stage framework that uses an external facade structural shell as lightweight geometric supervision for video-driven Gaussian reconstruction. The method aligns an exterior shell to the video reconstruction frame, renders per-view depth, camera-space normal, and valid-mask maps, and applies these cues through mask-gated losses during Gaussian optimization. This design preserves RGB-driven appearance while regularizing only visible shell-supported facade regions. Experiments on anonymized close-range urban facade scenes show improved facade orientation and visible-surface point-cloud consistency over photo-only, monocular-cue, and surface-oriented Gaussian baselines, while maintaining comparable held-out rendering quality.

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

2 major / 2 minor

Summary. The paper proposes shell-supervised Gaussian Splatting, a reconstruction-stage framework that aligns an external facade structural shell to video-driven 3D Gaussian optimization and applies mask-gated losses on rendered per-view depth, camera-space normals, and valid masks. This is intended to regularize only visible shell-supported regions of close-range urban facades while preserving RGB appearance, with the central claim being improved facade orientation and visible-surface point-cloud consistency over photo-only, monocular-cue, and surface-oriented baselines, while maintaining comparable held-out rendering quality.

Significance. If the alignment and bias concerns are resolved, the approach offers a lightweight, practical mechanism for stabilizing geometry in real-to-sim urban reconstruction where standard photometric optimization fails due to reflections, glass, and weak texture; this could be relevant for embodied AI applications requiring reliable collision and navigation geometry.

major comments (2)
  1. [Abstract] Abstract: the claim of improved facade orientation and visible-surface point-cloud consistency over baselines is stated without any quantitative metrics, error bars, dataset details, ablation studies, or alignment accuracy numbers, leaving the central experimental claim unsupported.
  2. [Abstract] Abstract (method description): the load-bearing assumption that the external shell supplies reliable geometric cues only for visible regions without systematic bias from misalignment or shell deviations is asserted but not validated by any quantitative alignment error, ground-truth comparison, or ablation on misalignment effects.
minor comments (2)
  1. Clarify the construction and sourcing of the 'exterior shell' and the exact alignment procedure for reproducibility.
  2. The phrase 'anonymized close-range urban facade scenes' would benefit from additional context on scene scale, number of views, and baseline implementations.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback on the abstract. We address the two major comments point-by-point below and will revise the manuscript accordingly.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the claim of improved facade orientation and visible-surface point-cloud consistency over baselines is stated without any quantitative metrics, error bars, dataset details, ablation studies, or alignment accuracy numbers, leaving the central experimental claim unsupported.

    Authors: We agree that the abstract presents the central claim at a high level without quantitative support. The full manuscript reports these results (including facade orientation and point-cloud consistency metrics, dataset details, and ablations) in the Experiments section. To strengthen the abstract, we will revise it to include representative quantitative findings from the evaluations while remaining within length limits. revision: yes

  2. Referee: [Abstract] Abstract (method description): the load-bearing assumption that the external shell supplies reliable geometric cues only for visible regions without systematic bias from misalignment or shell deviations is asserted but not validated by any quantitative alignment error, ground-truth comparison, or ablation on misalignment effects.

    Authors: The manuscript describes the shell alignment procedure and the use of mask-gating to restrict supervision to visible regions. We acknowledge that the abstract (and manuscript) does not include explicit quantitative alignment error, ground-truth comparisons, or misalignment ablations. We will revise the method and discussion sections to provide additional detail on the alignment process and any available alignment quality measures; a full ablation on misalignment effects would require new experiments and is noted as a limitation. revision: partial

Circularity Check

0 steps flagged

No circularity; external shell provides independent supervision

full rationale

The derivation chain defines mask-gated losses on depth/normal/valid maps rendered from an externally supplied and aligned facade shell. These quantities are not fitted from the Gaussian parameters or defined in terms of the target metrics; the shell originates outside the optimization loop. Experiments compare against independent baselines on held-out rendering and point-cloud consistency, with no reduction of claimed improvements to quantities defined by the inputs. No self-citation is invoked as a uniqueness theorem or load-bearing premise for the method. The central claim therefore remains self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The approach rests on standard computer-vision assumptions plus one domain-specific premise about shell alignment; no new entities are postulated and only minor tunable loss weights are expected.

free parameters (1)
  • mask-gated loss weights
    Weights balancing RGB, depth, and normal terms are expected to be chosen or tuned on the target scenes.
axioms (1)
  • domain assumption An external structural shell can be aligned to the video reconstruction frame with sufficient accuracy to provide useful per-view depth and normal cues.
    The method description states that the shell is aligned to the video reconstruction frame before rendering supervision maps.

pith-pipeline@v0.9.1-grok · 5717 in / 1253 out tokens · 29105 ms · 2026-06-30T06:39:54.749377+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

51 extracted references · 2 canonical work pages · 1 internal anchor

  1. [1]

    Barron, Ben Mildenhall, Dor Verbin, Pratul P

    Jonathan T. Barron, Ben Mildenhall, Dor Verbin, Pratul P. Srinivasan, and Peter Hedman. Mip-NeRF 360: Unbounded anti-aliased neural radiance fields. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 5470–5479, 2022

    2022

  2. [2]

    Matterport3D: Learning from RGB- D data in indoor environments

    Angel Chang, Angela Dai, Thomas Funkhouser, Maciej Hal- ber, Matthias Nießner, Manolis Savva, Shuran Song, Andy Zeng, and Yinda Zhang. Matterport3D: Learning from RGB- D data in indoor environments. InProceedings of the Inter- national Conference on 3D Vision (3DV), pages 667–676, 2017

    2017

  3. [3]

    Depth-supervised NeRF: Fewer views and faster training for free

    Kangle Deng, Andrew Liu, Jun-Yan Zhu, and Deva Ra- manan. Depth-supervised NeRF: Fewer views and faster training for free. InProceedings of the IEEE/CVF Confer- ence on Computer Vision and Pattern Recognition (CVPR), pages 12882–12891, 2022

    2022

  4. [4]

    L´opez, and Vladlen Koltun

    Alexey Dosovitskiy, German Ros, Felipe Codevilla, Anto- nio M. L´opez, and Vladlen Koltun. CARLA: An open urban driving simulator. InProceedings of the 1st Annual Confer- ence on Robot Learning (CoRL), pages 1–16. PMLR, 2017

    2017

  5. [5]

    SuGaR: Surface- aligned gaussian splatting for efficient 3D mesh reconstruc- tion and high-quality mesh rendering

    Antoine Gu ´edon and Vincent Lepetit. SuGaR: Surface- aligned gaussian splatting for efficient 3D mesh reconstruc- tion and high-quality mesh rendering. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 5354–5363, 2024

    2024

  6. [6]

    Neural 3D scene reconstruction with the manhattan-world assumption

    Haoyu Guo, Sida Peng, Haotong Lin, Qianqian Wang, Guofeng Zhang, Hujun Bao, and Xiaowei Zhou. Neural 3D scene reconstruction with the manhattan-world assumption. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 5511–5520, 2022

    2022

  7. [7]

    arXiv:2306.04988 (2023)

    Jianfei Guo, Nianchen Deng, Xinyang Li, Yeqi Bai, Bo- tian Shi, Chiyu Wang, Chenjing Ding, Dongliang Wang, and Yikang Li. Streetsurf: Extending multi-view im- plicit surface reconstruction to street views.arXiv preprint arXiv:2306.04988, 2023

    work page arXiv 2023

  8. [8]

    2D gaussian splatting for geometrically ac- curate radiance fields

    Binbin Huang, Zehao Yu, Anpei Chen, Andreas Geiger, and Shenghua Gao. 2D gaussian splatting for geometrically ac- curate radiance fields. InACM SIGGRAPH 2024 Conference Papers, pages 1–11, 2024

    2024

  9. [9]

    Geometry field splatting with gaussian sur- fels

    Kaiwen Jiang, Venkataram Sivaram, Cheng Peng, and Ravi Ramamoorthi. Geometry field splatting with gaussian sur- fels. InProceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition (CVPR), pages 5752– 5762, 2025

    2025

  10. [10]

    3D gaussian splatting for real-time radiance field rendering.ACM Transactions on Graphics, 42 (4):1–14, 2023

    Bernhard Kerbl, Georgios Kopanas, Thomas Leimk ¨uhler, and George Drettakis. 3D gaussian splatting for real-time radiance field rendering.ACM Transactions on Graphics, 42 (4):1–14, 2023

    2023

  11. [11]

    Tanks and temples: Benchmarking large-scale scene reconstruction.ACM Transactions on Graphics, 36(4):1–13, 2017

    Arno Knapitsch, Jaesik Park, Qian-Yi Zhou, and Vladlen Koltun. Tanks and temples: Benchmarking large-scale scene reconstruction.ACM Transactions on Graphics, 36(4):1–13, 2017

    2017

  12. [12]

    AI2-THOR: An Interactive 3D Environment for Visual AI

    Eric Kolve, Roozbeh Mottaghi, Winson Han, Eli VanderBilt, Luca Weihs, Alvaro Herrasti, Matt Deitke, Kiana Ehsani, Daniel Gordon, Yuke Zhu, Aniruddha Kembhavi, Abhinav Gupta, and Ali Farhadi. AI2-THOR: An interactive 3D en- vironment for visual AI.arXiv preprint arXiv:1712.05474, 2017

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  13. [13]

    Tay- lor, Mathias Unberath, Ming-Yu Liu, and Chen-Hsuan Lin

    Zhaoshuo Li, Thomas M ¨uller, Alex Evans, Russell H. Tay- lor, Mathias Unberath, Ming-Yu Liu, and Chen-Hsuan Lin. Neuralangelo: High-fidelity neural surface reconstruction. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 8456–8465, 2023

    2023

  14. [14]

    Scaffold-GS: Structured 3D gaussians for view-adaptive rendering

    Tao Lu, Mulin Yu, Linning Xu, Yuanbo Xiangli, Limin Wang, Dahua Lin, and Bo Dai. Scaffold-GS: Structured 3D gaussians for view-adaptive rendering. InProceedings of the IEEE/CVF Conference on Computer Vision and Pat- tern Recognition (CVPR), pages 20654–20664, 2024

    2024

  15. [15]

    Srinivasan, Matthew Tancik, Jonathan T

    Ben Mildenhall, Pratul P. Srinivasan, Matthew Tancik, Jonathan T. Barron, Ravi Ramamoorthi, and Ren Ng. NeRF: Representing scenes as neural radiance fields for view syn- thesis. InProceedings of the European Conference on Com- puter Vision (ECCV), pages 405–421, 2020

    2020

  16. [16]

    Instant neural graphics primitives with a multires- olution hash encoding.ACM Transactions on Graphics, 41 (4):1–15, 2022

    Thomas M ¨uller, Alex Evans, Christoph Schied, and Alexan- der Keller. Instant neural graphics primitives with a multires- olution hash encoding.ACM Transactions on Graphics, 41 (4):1–15, 2022

    2022

  17. [17]

    Barron, Ben Mildenhall, Mehdi S

    Michael Niemeyer, Jonathan T. Barron, Ben Mildenhall, Mehdi S. M. Sajjadi, Andreas Geiger, and Noha Radwan. RegNeRF: Regularizing neural radiance fields for view syn- thesis from sparse inputs. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 5480–5490, 2022

    2022

  18. [18]

    Open3D raycastingscene docu- mentation.https : / / www

    Open3D Developers. Open3D raycastingscene docu- mentation.https : / / www . open3d . org / docs / latest / python _ api / open3d . t . geometry . RaycastingScene.html, 2026. Accessed: 2026-06- 17

    2026

  19. [19]

    Scanning technologies to building information modelling: A review.Infrastructures, 7(4):49, 2022

    Rabia Rashdi, Joaqu ´ın Mart´ınez-S´anchez, Pedro Arias, and Zhouyan Qiu. Scanning technologies to building information modelling: A review.Infrastructures, 7(4):49, 2022

    2022

  20. [20]

    Srini- vasan, Jonathan T

    Konstantinos Rematas, Andrew Liu, Pratul P. Srini- vasan, Jonathan T. Barron, Andrea Tagliasacchi, Thomas Funkhouser, and Vittorio Ferrari. Urban radiance fields. In Proceedings of the IEEE/CVF Conference on Computer Vi- sion and Pattern Recognition (CVPR), pages 12932–12942, 2022

    2022

  21. [21]

    Habitat: A platform for embodied AI research

    Manolis Savva, Abhishek Kadian, Oleksandr Maksymets, Yili Zhao, Erik Wijmans, Bhavana Jain, Julian Straub, Jia Liu, Vladlen Koltun, Jitendra Malik, Devi Parikh, and Dhruv Batra. Habitat: A platform for embodied AI research. In Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), pages 9339–9347, 2019

    2019

  22. [22]

    Sch ¨onberger and Jan-Michael Frahm

    Johannes L. Sch ¨onberger and Jan-Michael Frahm. Structure- from-motion revisited. InProceedings of the IEEE Confer- ence on Computer Vision and Pattern Recognition (CVPR), pages 4104–4113, 2016

    2016

  23. [23]

    igibson 1.0: A simulation environment for interactive tasks in large realistic scenes

    Bokui Shen, Fei Xia, Chengshu Li, Roberto Mart ´ın-Mart´ın, Linxi Fan, Guanzhi Wang, Claudia P ´erez-D’Arpino, Shya- mal Buch, Sanjana Srivastava, Lyne Tchapmi, Micael Tchapmi, Kent Vainio, Josiah Wong, Fei-Fei Li, and Sil- vio Savarese. igibson 1.0: A simulation environment for interactive tasks in large realistic scenes. InProceedings of the IEEE/RSJ In...

    2021

  24. [24]

    Neuralrecon: Real-time coherent 3D reconstruc- tion from monocular video

    Jiaming Sun, Yiming Xie, Linghao Chen, Xiaowei Zhou, and Hujun Bao. Neuralrecon: Real-time coherent 3D reconstruc- tion from monocular video. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 15598–15607, 2021

    2021

  25. [25]

    Srinivasan, Jonathan T

    Matthew Tancik, Vincent Casser, Xinchen Yan, Sabeek Pradhan, Ben Mildenhall, Pratul P. Srinivasan, Jonathan T. Barron, and Henrik Kretzschmar. Block-NeRF: Scalable large scene neural view synthesis. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 8248–8258, 2022

    2022

  26. [26]

    Mega-NeRF: Scalable construction of large- scale NeRFs for virtual fly-throughs

    Haithem Turki, Deva Ramanan, and Mahadev Satya- narayanan. Mega-NeRF: Scalable construction of large- scale NeRFs for virtual fly-throughs. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 12922–12931, 2022

    2022

  27. [27]

    DN-splatter: Depth and normal priors for gaussian splatting and meshing

    Matias Turkulainen, Xuqian Ren, Iaroslav Melekhov, Otto Seiskari, Esa Rahtu, and Juho Kannala. DN-splatter: Depth and normal priors for gaussian splatting and meshing. InPro- ceedings of the IEEE/CVF Winter Conference on Applica- tions of Computer Vision (WACV), pages 2421–2431, 2025

    2025

  28. [28]

    BIMCaP: BIM-based AI- supported LiDAR-camera pose refinement

    Miguel Arturo Vega-Torres, Anna Ribic, Borja Garc ´ıa de Soto, and Andr ´e Borrmann. BIMCaP: BIM-based AI- supported LiDAR-camera pose refinement. InProceedings of the 31st International Workshop on Intelligent Computing in Engineering (EG-ICE 2024), pages 423–432, 2024

    2024

  29. [29]

    NeuRIS: Neural reconstruction of indoor scenes using nor- mal priors

    Jiepeng Wang, Peng Wang, Xiaoxiao Long, Christian Theobalt, Taku Komura, Lingjie Liu, and Wenping Wang. NeuRIS: Neural reconstruction of indoor scenes using nor- mal priors. InProceedings of the European Conference on Computer Vision (ECCV), pages 139–155, 2022

    2022

  30. [30]

    NeuS: Learning neural im- plicit surfaces by volume rendering for multi-view recon- struction

    Peng Wang, Lingjie Liu, Yuan Liu, Christian Theobalt, Taku Komura, and Wenping Wang. NeuS: Learning neural im- plicit surfaces by volume rendering for multi-view recon- struction. InAdvances in Neural Information Processing Systems (NeurIPS), pages 27171–27183, 2021

    2021

  31. [31]

    Bovik, Hamid R

    Zhou Wang, Alan C. Bovik, Hamid R. Sheikh, and Eero P. Simoncelli. Image quality assessment: From error visibility to structural similarity.IEEE Transactions on Image Pro- cessing, 13(4):600–612, 2004

    2004

  32. [32]

    Lovegrove, Ju- lian Straub, Simon Green, Richard Szeliski, Steven Butter- field, Shobhit Verma, and Richard Newcombe

    Thomas Whelan, Michael Goesele, Steven J. Lovegrove, Ju- lian Straub, Simon Green, Richard Szeliski, Steven Butter- field, Shobhit Verma, and Richard Newcombe. Reconstruct- ing scenes with mirror and glass surfaces.ACM Transactions on Graphics, 37(4):1–11, 2018

    2018

  33. [33]

    Zamir, Zhi-Yang He, Alexander Sax, Ji- tendra Malik, and Silvio Savarese

    Fei Xia, Amir R. Zamir, Zhi-Yang He, Alexander Sax, Ji- tendra Malik, and Silvio Savarese. Gibson env: Real-world perception for embodied agents. InProceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 9068–9079, 2018

    2018

  34. [34]

    Video2game: Real-time, interactive, realistic and browser-compatible environment from a single video

    Hongchi Xia, Zhi-Hao Lin, Wei-Chiu Ma, and Shenlong Wang. Video2game: Real-time, interactive, realistic and browser-compatible environment from a single video. In Proceedings of the IEEE/CVF Conference on Computer Vi- sion and Pattern Recognition (CVPR), pages 4578–4588, 2024

    2024

  35. [35]

    S-NeRF: Neural radiance fields for street views

    Ziyang Xie, Junge Zhang, Wenye Li, Feihu Zhang, and Li Zhang. S-NeRF: Neural radiance fields for street views. InInternational Conference on Learning Representations (ICLR), 2023

    2023

  36. [36]

    Vid2sim: Realistic and interactive simu- lation from video for urban navigation

    Ziyang Xie, Zhizheng Liu, Zhenghao Peng, Wayne Wu, and Bolei Zhou. Vid2sim: Realistic and interactive simu- lation from video for urban navigation. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 1581–1591, 2025

    2025

  37. [37]

    Street gaussians: Modeling dynamic urban scenes with gaussian splatting

    Yunzhi Yan, Haotong Lin, Chenxu Zhou, Weijie Wang, Haiyang Sun, Kun Zhan, Xianpeng Lang, Xiaowei Zhou, and Sida Peng. Street gaussians: Modeling dynamic urban scenes with gaussian splatting. InProceedings of the Euro- pean Conference on Computer Vision (ECCV), pages 156– 173, 2024

    2024

  38. [38]

    V ol- ume rendering of neural implicit surfaces

    Lior Yariv, Jiatao Gu, Yoni Kasten, and Yaron Lipman. V ol- ume rendering of neural implicit surfaces. InAdvances in Neural Information Processing Systems (NeurIPS), 2021

    2021

  39. [39]

    MonoSDF: Exploring monoc- ular geometric cues for neural implicit surface reconstruc- tion

    Zehao Yu, Songyou Peng, Michael Niemeyer, Torsten Sat- tler, and Andreas Geiger. MonoSDF: Exploring monoc- ular geometric cues for neural implicit surface reconstruc- tion. InAdvances in Neural Information Processing Systems (NeurIPS), 2022

    2022

  40. [40]

    Mip-splatting: Alias-free 3D gaussian splat- ting

    Zehao Yu, Anpei Chen, Binbin Huang, Torsten Sattler, and Andreas Geiger. Mip-splatting: Alias-free 3D gaussian splat- ting. InProceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition (CVPR), pages 19447– 19456, 2024

    2024

  41. [41]

    Gaussian opacity fields: Efficient adaptive surface reconstruction in unbounded scenes.ACM Transactions on Graphics, 43(6): 1–13, 2024

    Zehao Yu, Torsten Sattler, and Andreas Geiger. Gaussian opacity fields: Efficient adaptive surface reconstruction in unbounded scenes.ACM Transactions on Graphics, 43(6): 1–13, 2024

    2024

  42. [42]

    Zanjani, Hong Cai, Hanno Ackermann, Leila Mir- vakhabova, and Fatih Porikli

    Farhad G. Zanjani, Hong Cai, Hanno Ackermann, Leila Mir- vakhabova, and Fatih Porikli. Planar gaussian splatting. In Proceedings of the IEEE/CVF Winter Conference on Ap- plications of Computer Vision (WACV), pages 8887–8896, 2025

    2025

  43. [43]

    Qilin Zhang, Olaf Wysocki, and Boris Jutzi. GS4Buildings: Prior-guided gaussian splatting for 3D building reconstruc- tion.ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences, X-4/W6-2025:249–256, 2025

    2025

  44. [44]

    Efros, Eli Shecht- man, and Oliver Wang

    Richard Zhang, Phillip Isola, Alexei A. Efros, Eli Shecht- man, and Oliver Wang. The unreasonable effectiveness of deep features as a perceptual metric. InProceedings of the IEEE Conference on Computer Vision and Pattern Recogni- tion (CVPR), pages 586–595, 2018

    2018

  45. [45]

    Pixel-GS: Density control with pixel-aware gradient for 3D gaussian splatting

    Zheng Zhang, Wenbo Hu, Yixing Lao, Tong He, and Heng- shuang Zhao. Pixel-GS: Density control with pixel-aware gradient for 3D gaussian splatting. InProceedings of the Eu- ropean Conference on Computer Vision (ECCV), pages 326– 342, 2024. Supplementary Material: Shell-Supervised Gaussian Splatting for Urban Real-to-Sim Reconstruction Yuan Yang1 Peijun Lu2 ...

    2024

  46. [46]

    All scene names, file paths, and site identifiers are omitted to preserve anonymity

    Evaluation Protocol and Alignment Valida- tion This supplementary material provides additional evaluation details and results for the main paper. All scene names, file paths, and site identifiers are omitted to preserve anonymity. We use the same terminology as the main paper and refer to the two evaluated sequences as the primary scene and the secondary ...

  47. [47]

    Depth-only supervision im- proves the scale-aligned depth diagnostic but does not cor- rect facade orientation

    Primary-scene Ablation Table 1 reports the complete ablation of shell-guided loss terms on the primary scene. Depth-only supervision im- proves the scale-aligned depth diagnostic but does not cor- rect facade orientation. In contrast, shell-rendered normal supervision is the dominant cue for reducing unsigned an- gular normal error. Combining depth and no...

  48. [48]

    all regions

    Spatial-region Holdout Stress Test Because our geometry metrics use the exterior facade shell as the visible-structure reference, we additionally conduct a spatial-region holdout stress test on the primary scene. We split shell-supported facade pixels into alternating vertical 1 Table 2. Spatial-region holdout stress test on the primary scene. Ours-A rece...

  49. [49]

    The visualization includes the RGB frame, shell depth, camera-space shell normal, and valid mask

    Secondary-scene Shell Supervision Figure 1 shows a representative shell supervision example from the secondary scene. The visualization includes the RGB frame, shell depth, camera-space shell normal, and valid mask. White pixels in the valid mask denote shell- supported regions used for geometric supervision. Table 3. Secondary-scene geometry validation. ...

  50. [50]

    The secondary scene is used as a validation case rather than a large-scale generaliza- tion benchmark

    Secondary-scene Quantitative Validation Table 3 reports secondary-scene geometry metrics on shell- supported held-out test views. The secondary scene is used as a validation case rather than a large-scale generaliza- tion benchmark. Consistent with the primary scene, the strongest improvement appears in facade orientation: shell- guided supervision reduce...

  51. [51]

    To avoid street-level iden- tifying content, the figure compares only rendered normal maps and angular-error maps on shell-supported regions

    Secondary-scene Qualitative Geometry Figure 2 provides a qualitative geometry comparison on a held-out secondary-scene view. To avoid street-level iden- tifying content, the figure compares only rendered normal maps and angular-error maps on shell-supported regions. The shell-guided model produces more coherent facade ori- entation and lower unsigned angu...