Pith. sign in

REVIEW 2 major objections 5 minor 54 references

Gaussian Splatting for Efficient Satellite Image Photogrammetry

T0 review · 2 major / 5 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read Gaussian splatting with affine satellite cameras reconstructs DSM accuracy comparable to EO-NeRF in minutes instead of a day.

desk verdict EOGS is the first credible 3DGS pipeline for satellite photogrammetry, hitting EO-NeRF-comparable MAE at ~300× lower cost, but the missing shadow-decay coefficient ρ is a concrete reproducibility gap. read the letter →

arxiv 2412.13047 v2 pith:JKSRVZFG submitted 2024-12-17 cs.CV

Luca Savant Aira , Gabriele Facciolo , Thibaud Ehret This is my paper · ORCID
classification cs.CV
keywords 3DGaussiansplattingsatellitephotogrammetrydigitalsurfacemodelradiancefieldsshadowmappingaffinecameraremotesensingmulti-viewstereo
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

EOGS adapts 3D Gaussian splatting to satellite photogrammetry, recovering digital surface models from a small set of multi-date satellite images. The paper's central claim is that this approach reaches elevation accuracy comparable to EO-NeRF, the leading NeRF-based method for this task, while training about 300 times faster: roughly 3 minutes versus 15 hours. The method works by approximating each satellite's RPC pushbroom camera with a per-scene affine camera, so the splatting projection becomes exact and cheap, and by adding a shadow-mapping step based on the sun camera plus three regularizers that promote sparse, view-consistent, opaque geometry. If this holds, satellite DSM production shifts from day-long optimization to a practical interactive tool, matching the accuracy of the slow state of the art at a fraction of the cost.

What carries the argument

The load-bearing object is the per-scene affine camera model $A(x) = Ax + a$ that approximates the full composition of world-to-UTM-to-RPC-to-NDC transformations, with a reported mean reprojection error of about 0.012 pixels; this makes the splatting projection exact, removing the first-order Jacobian approximation of the original 3DGS. The second mechanism is the shadow-mapping variant: the sun is treated as a directional light and represented by an affine sun camera $S$, the elevation render is produced from both the satellite and sun cameras, and the shadow coefficient $s_{A,S}(u) = \min(\exp(-\rho\Delta h_{A,S}(u)),1)$ is applied per pixel, with an ambient-light term $\psi_A$ and a camera-specific color correction $\phi_A$. Three regularizers — L1 opacity sparsity, local view-consistency on albedo and elevation, and an entropy penalty on shadow values — keep the optimization stable and the geometry hard-surfaced. Together, these components make the entire pipeline run in about 3 minutes on datasets where EO-NeRF needs 15 hours.

What would settle it

Compute the full RPC projection for a test scene and compare it to the affine fit; if the mean reprojection error exceeds a few hundredths of a pixel at the scene edges or for strong relief, then extrapolate the effect on the elevation renders. Concretely, run EOGS on a larger AOI or one with steeper terrain and compare the lidar-aligned MAE against EO-NeRF; if EOGS no longer stays within roughly 0.1-0.2 m of EO-NeRF's MAE, the central accuracy claim fails.

Watch

Extended reading notes

Core claim

The central discovery is that the bottleneck of NeRF-based Earth observation — the day-long optimization needed to render shadows and geometry — is not intrinsic: a Gaussian-splatting representation can deliver the same DSM quality in minutes. The key step is replacing the full RPC camera model with a per-scene affine approximation, turning the splatting projection into an exact linear map and removing the need for the first-order Jacobian used in standard 3DGS. Shadows are then rendered by a custom shadow map: the scene's elevation is rendered from the sun's viewpoint, resampled at homologous points, and pixels whose sun-view elevation is higher are darkened with an exponential coefficient derived from a homogeneous-medium model. On the DFC2019 and IARPA2016 benchmarks, EOGS reports elevation MAE within the same range as EO-NeRF (about 1.2-1.6 m depending on scene and foliage masking) while training in about 3 minutes, making it the Pareto-optimal method in accuracy-versus-time among the evaluated approaches.

Load-bearing premise

The method's accuracy rests on the per-scene affine camera being a faithful stand-in for the true RPC pushbroom projection; if that local linearization degrades over larger scenes, higher relief, or more oblique views, the reported 0.012-pixel error and the resulting elevation accuracy no longer hold.

Editorial extensions

If this is right

  • If the paper is right, day-long NeRF training for satellite photogrammetry can be replaced by a 3-minute Gaussian-splatting optimization on the same hardware, enabling near-real-time DSM updates from growing satellite image archives.
  • The affine camera approximation broadens 3DGS to pushbroom sensors with only a small projection error, suggesting that other non-pinhole sensors could be handled in the same way.
  • The shadow-mapping formulation removes the ray-marching requirement of EO-NeRF, making physically grounded shadow modeling compatible with rasterized Gaussian splatting.
  • On foliage-masked evaluation, EOGS matches EO-NeRF's accuracy and in some scenes exceeds it, indicating that structural detail such as buildings and thin vertical structures is recovered at least as well as the state of the art.

Reading between the lines

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

  • Inference: The 0.012-pixel affine approximation error suggests the method's accuracy ceiling is tied to scene extent and relief; extending to large AOIs or steep terrain would likely require piecewise or higher-order camera approximations.
  • Inference: Because performance degrades in low-visibility regions, combining EOGS with a stereo or multi-view depth prior could close the gap with NeRF in those areas while retaining speed.
  • Inference: The shadow-map entropy regularizer hints at a more general principle: Gaussian-splatting optimizers can be steered toward hard-surface geometry by penalizing fractional shadow values, which may transfer to other non-Lambertian or transient scenes.
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

2 major / 5 minor

Summary. The paper introduces EOGS, a 3D Gaussian splatting pipeline adapted to multi-date satellite photogrammetry. It replaces the RPC pushbroom camera model with a per-scene affine approximation, renders shadows through a shadow-mapping variant, and adds sparsity, view-consistency, and opaqueness regularizers. On the DFC2019 and IARPA2016 benchmarks, EOGS reports elevation MAE within about 0.1 m of EO-NeRF without foliage masking (1.46 m vs. 1.35 m) and slightly better with foliage masking (1.37 m vs. 1.38 m), while training in about 3 minutes compared with 15 hours. The authors claim an approximately 300x speedup and a Pareto-optimal accuracy/time trade-off.

Significance. If the reported numbers hold, this is a meaningful advance: EOGS appears to be the first Gaussian-splatting method for satellite photogrammetry, and it achieves accuracy comparable to a strong NeRF baseline at a fraction of the training cost. The evaluation is grounded in external lidar ground truth and compares against independent baselines (EO-NeRF, SAT-NGP, Sat-Mesh, S2P), and the ablation in Table 2 gives quantitative support for each proposed component. The main weaknesses are a missing numerical specification of the shadow decay coefficient and a limited discussion of the robustness of the affine camera approximation, both of which are directly relevant to the central accuracy claim.

major comments (2)
  1. [§3.2, Eq. (11)] The numerical value of the shadow decay coefficient ρ is never given in the main text or in the implementation details. Equation (11) defines the shadow darkening as exp(-ρΔh), and the ablation in Table 2 identifies shadow mapping as the single largest accuracy contributor (a 3.16 m MAE gain). Without the value of ρ, the reported MAE numbers cannot be reproduced, and the reader cannot assess whether the method is sensitive to this parameter. Please report the value used in all experiments and, ideally, a short sensitivity study over a range of ρ.
  2. [§3.1] The per-scene affine approximation of the RPC pushbroom model is justified only by a mean projection error of about 0.012 pixels. Since this affine model is used for every Gaussian projection and for the shadow comparisons, a mean value alone does not establish that the approximation is safe for scenes with larger relief, more oblique views, or larger footprints. Please report the maximum and distribution of the approximation error, and if possible test the method on a scene with greater relief or viewing diversity, so that the generalization of the central accuracy claim is better supported.
minor comments (5)
  1. [Figure 3 caption] The caption contains a typo: "corresponds to a the the 3D point" should be "corresponds to the 3D point".
  2. [§3.4, Eq. (21)] Equation (21) writes "min" without explicitly stating the optimization variables; please specify that the minimization is over the Gaussian primitive parameters and the camera-dependent correction parameters.
  3. [§4.1, Table 1] The training-time comparison mixes numbers from different papers and likely different hardware; a short sentence describing the hardware and noting this caveat would make the 300x speedup claim more precise.
  4. [§4.2, Table 2] The ablation table reports training time for only three of the nine configurations; reporting times for all configurations would make the efficiency contribution of each component clearer.
  5. [§3.2] The phrase "homogeneous medium of density ρ" would benefit from specifying the units of ρ and clarifying how it relates to the attenuation coefficient in the cited volume-rendering model.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: EOGS accuracy is judged against external lidar and independent baselines; self-citations are not load-bearing.

full rationale

EOGS's derivation chain is self-contained against external ground truth. The optimization (Eq. 6) minimizes a photometric loss between the synthesized views from Eq. (13) and the input satellite images; geometry is encoded in the Gaussian centers/elevations and is not supervised by lidar. Accuracy is then measured by MAE between the elevation render (Eq. 7) and lidar scans from DFC2019/IARPA2016, an external reference not used in training. The affine camera approximation (Section 3.1) and shadow-mapping shading (Eqs. 10-13) are model components evaluated through this external metric; the per-scene affine residual (~0.012 px) is an approximation-error statement, not a fitting of the reported MAE. Regularization coefficients were tuned on a single scene and applied uniformly; this is standard hyperparameter selection, and the reported performance is measured on held-out scene-level MAE. Self-citations to EO-NeRF [32] describe prior work and a baseline that is run, not a premise that forces the new result. The undisplayed value of rho in Eq. (11) is a reproducibility gap, not circularity: rho is a fixed decay hyperparameter, and the shadow map's entropy regularizer (Eq. 20) explicitly steers it toward hard shadows rather than fitting the evaluation lidar. No load-bearing step reduces by construction to its inputs.

Assumptions & free parameters 9 free parameters · 5 assumptions · 0 invented entities

The method rests on standard 3DGS optimization machinery, an affine approximation of pushbroom RPC cameras, a directional-sun shadow model, and the assumption that alpha-composited elevation renders represent surface height. Regularization weights are hand-tuned on one scene, and the shadow coefficient rho is not specified in the main text.

free parameters (9)
  • Sparsity regularization weight lambda_o = 0.1
    Eq. (21); found experimentally on a single scene and rounded to the nearest power of ten (Section 3.4).
  • View-consistency color weight lambda_cc = 0.1
    Eq. (21); same tuning procedure as the sparsity weight.
  • View-consistency altitude weight lambda_ac = 0.01
    Eq. (21); same tuning procedure.
  • Opaqueness entropy weight lambda_s = 0.01
    Eq. (21); same tuning procedure.
  • Shadow decay coefficient rho = not stated in main text
    Eq. (11) relies on rho; authors note large rho is unstable but do not give the value used. Likely in the supplement, but missing from the main text.
  • Initial Gaussian density = 0.13 Gaussians per m^3
    Section 3.4; hand-set initialization density controlling the number of primitives K.
  • Opacity pruning threshold alpha_min = 0.0025
    Section 3.3; any primitive with alpha below alpha_min is discarded.
  • View-consistency altitude threshold delta_h_min = 30 cm
    Section 3.3; mask selection for the consistency constraints.
  • Virtual camera perturbation scale = 0.05
    Eq. (19); magnitude of the random perturbation applied to camera A to obtain camera B.
assumptions (5)
  • domain assumption Per-scene affine approximation of the RPC pushbroom camera transformation is accurate (mean error about 0.012 pixels).
    Invoked in Section 3.1 to justify replacing the RPC mapping with an affine camera A; if this approximation fails for larger or more oblique scenes, all splatting projections are biased.
  • domain assumption The sun is the only light source and can be modeled as a directional light; shadows are determined by comparing elevation renders from the sun camera.
    Section 3.2; the shadow mapping pipeline assumes a single directional light and ignores atmospheric scattering beyond a per-camera ambient term psi_A.
  • domain assumption Elevation can be rendered via alpha compositing of Gaussian altitudes (Eq. 7), i.e., the rendered elevation is a consistent surface height.
    The shadow mapping and altitude-consistency losses depend on E_A(u) being a meaningful single-surface elevation; with semi-transparent Gaussians this is only approximate until opaqueness is enforced.
  • standard math Standard 3DGS differentiable rasterization and optimization machinery (from Kerbl et al. [23]) works as claimed.
    The paper builds on the original 3DGS codebase and inherits its projection, densification, and optimization assumptions.
  • domain assumption Bundled-adjusted RPC coefficients and provided sun directions are accurate enough for the pipeline.
    Experiments use the bundled-adjusted RPCs from EO-NeRF (Section 4); errors in these would directly affect geometry recovery.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Gaussian Splatting for Efficient Satellite Image Photogrammetry." pith.science (2026). https://pith.science/paper/JKSRVZFG

@misc{pith2026241213047,
  author       = {Pith},
  title        = {Pith review of: Gaussian Splatting for Efficient Satellite Image Photogrammetry},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/JKSRVZFG}},
  note         = {Machine review of arXiv:2412.13047}
}
read the original abstract

Recently, Gaussian splatting has emerged as a strong alternative to NeRF, demonstrating impressive 3D modeling capabilities while requiring only a fraction of the training and rendering time. In this paper, we show how the standard Gaussian splatting framework can be adapted for remote sensing, retaining its high efficiency. This enables us to achieve state-of-the-art performance in just a few minutes, compared to the day-long optimization required by the best-performing NeRF-based Earth observation methods. The proposed framework incorporates remote-sensing improvements from EO-NeRF, such as radiometric correction and shadow modeling, while introducing novel components, including sparsity, view consistency, and opacity regularizations.

Figures

Figures reproduced from arXiv: 2412.13047 by the authors.

Figure 1
Figure 1. Using a limited number of satellite images of a given [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Summary of the transformation from world-space to NDC-space and its affine approximation. The affine approximation is com [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Shadow mapping illustration. The point u in the satel￾lite image (affine camera A) corresponds to a the the 3D point x = locA(u) on the vertical wall. Projecting x to the sun camera (affine camera S), u˜ = Sx is obtained. Then y = locS (u˜) is obtained localizing u˜. The point x and its pixel u are in shadow because the elevation of y is greater than the elevation of x. In￾deed, all and only the points where the sat… view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: From top-left to bottom-right, shadow maps of EO [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 6
Figure 6. Figure 6: View consistency regularization parameter ablation [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]
Figure 7
Figure 7. Figure 7: Impact of the visibility on the performance (using foliage [PITH_FULL_IMAGE:figures/full_fig_p008_7.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

54 extracted references · 36 canonical work pages

  1. [1]

    Metric infor- mation extraction from spot images and the role of polyno- mial mapping functions

    Emmanuel P Baltsavias and Dirk Stallmann. Metric infor- mation extraction from spot images and the role of polyno- mial mapping functions. In XVII ISPRS Congress, Commis- sion IV. Swiss Federal Institute of Technology, Institute of Geodesy and Photogrammetry, 1992. 2

    work page 1992

  2. [2]

    3d gaussian splatting: Survey, technologies, challenges, and opportunities

    Yanqi Bao, Tianyu Ding, Jing Huo, Yaoli Liu, Yuxin Li, Wenbin Li, Yang Gao, and Jiebo Luo. 3d gaussian splatting: Survey, technologies, challenges, and opportunities. arXiv preprint arXiv:2407.17418, 2024. 3

    arXiv 2024

  3. [3]

    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. 6

    2022

  4. [4]

    The Ames Stereo Pipeline: NASA’s open source software for de- riving and processing terrain data

    Ross A Beyer, Oleg Alexandrov, and Scott McMichael. The Ames Stereo Pipeline: NASA’s open source software for de- riving and processing terrain data. Earth and Space Science, 5(9):537–548, 2018. 2

    work page 2018

  5. [5]

    Sat-ngp : Unleashing neural graphics primitives for fast relightable transient-free 3d reconstruction from satellite imagery

    Camille Billouard, Dawa Derksen, Emmanuelle Sarrazin, and Bruno Vallet. Sat-ngp : Unleashing neural graphics primitives for fast relightable transient-free 3d reconstruction from satellite imagery. In IGARSS 2024 - 2024 IEEE Inter- national Geoscience and Remote Sensing Symposium, pages 8749–8753, 2024. 2, 7, 8

    work page 2024

  6. [6]

    Semantic stereo for incidental satellite images

    Marc Bosch, Kevin Foster, Gordon Christie, Sean Wang, Gregory D Hager, and Myron Brown. Semantic stereo for incidental satellite images. In 2019 IEEE Winter Conference on Applications of Computer Vision (WACV) , pages 1524– 1532, 2019. 7

    work page 2019

  7. [7]

    Revising densification in gaussian splatting

    Samuel Rota Bul `o, Lorenzo Porzi, and Peter Kontschieder. Revising densification in gaussian splatting. arXiv preprint arXiv:2404.06109, 2024. 2, 5

    arXiv 2024

  8. [8]

    Pyramid stereo matching network

    Jia-Ren Chang and Yong-Sheng Chen. Pyramid stereo matching network. In 2018 IEEE/CVF Conference on Com- puter Vision and Pattern Recognition (CVPR), pages 5410– 5418, 2018. 2

    work page 2018

Show all 54 references
  1. [9]

    Earth observing satellites online com- pendium

    Jeff Clauson, Simon Cantrell, Jim Vrabel, John Oeding, Bhaskar Ranjitkar, Tom Rusten, Shankar Ramaseri, and Kimberly Casey. Earth observing satellites online com- pendium. U.S. Geological Survey, 2024. 1

    2024

  2. [10]

    Improving neural im- plicit surfaces geometry with patch warping

    Franc ¸ois Darmon, B´en´edicte Bascle, Jean-Cl ´ement Devaux, Pascal Monasse, and Mathieu Aubry. Improving neural im- plicit surfaces geometry with patch warping. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pat- tern Recognition, pages 6260–6269, 2022. 6

    2022

  3. [11]

    An automatic and modular stereo pipeline for pushbroom images

    Carlo de Franchis, Enric Meinhardt-Llopis, Julien Michel, Jean-Michel Morel, and Gabriele Facciolo. An automatic and modular stereo pipeline for pushbroom images. ISPRS Annals of Photogrammetry, Remote Sensing and Spatial In- formation Sciences, 2-3:49–56, 2014. 2, 7

    2014

  4. [12]

    Shadow neural radiance fields for multi-view satellite photogrammetry

    Dawa Derksen and Dario Izzo. Shadow neural radiance fields for multi-view satellite photogrammetry. In Proceed- ings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 1152–1161, 2021. 2

    2021

  5. [13]

    WorldView Legion, 2024

    eoPortal. WorldView Legion, 2024. https : / / www . eoportal . org / satellite - missions / worldview-legion. 1

    2024

  6. [14]

    MGM: A significantly more global matching for stereovi- sion

    Gabriele Facciolo, Carlo de Franchis, and Enric Meinhardt. MGM: A significantly more global matching for stereovi- sion. In Proceedings of the British Machine Vision Confer- ence (BMVC), number 90, pages 1–12, 2015. 2

    2015

  7. [15]

    Automatic 3D reconstruction from multi-date satel- lite images

    Gabriele Facciolo, Carlo de Franchis, and Enric Meinhardt- Llopis. Automatic 3D reconstruction from multi-date satel- lite images. In 2017 IEEE Conference on Computer Vision and Pattern Recognition Workshops (CVPRW), pages 1542– 1551, 2017. 7

    2017

  8. [16]

    Efros, and Xiaolong Wang

    Yang Fu, Sifei Liu, Amey Kulkarni, Jan Kautz, Alexei A. Efros, and Xiaolong Wang. Colmap-free 3d gaussian splat- ting. In Proceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition (CVPR), pages 20796– 20805, 2024. 3

    2024

  9. [17]

    Incorporating season and solar specificity into renderings made by a nerf archi- tecture using satellite images

    Michael Gableman and Avinash Kak. Incorporating season and solar specificity into renderings made by a nerf archi- tecture using satellite images. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2024. 2

    2024

  10. [18]

    Relightable 3d gaussian: Real-time point cloud relighting with brdf decomposition and ray trac- ing

    Jian Gao, Chun Gu, Youtian Lin, Hao Zhu, Xun Cao, Li Zhang, and Yao Yao. Relightable 3d gaussian: Real-time point cloud relighting with brdf decomposition and ray trac- ing. arXiv:2311.16043, 2023. 3

    2023 arXiv

  11. [19]

    Relationship be- tween photogrammmetry and computer vision

    Richard I Hartley and Joseph L Mundy. Relationship be- tween photogrammmetry and computer vision. Integrating photogrammetric techniques with scene analysis and ma- chine vision, 1944:92–105, 1993. 1

    1944

  12. [20]

    Stereo processing by semiglobal match- ing and mutual information

    Heiko Hirschmuller. Stereo processing by semiglobal match- ing and mutual information. IEEE Transactions on Pattern Analysis and Machine Intelligence, 30(2):328–341, 2007. 2

    2007

  13. [21]

    On the error analysis of 3d gaussian splat- ting and an optimal projection strategy

    Letian Huang, Jiayang Bai, Jie Guo, Yuanqi Li, and Yan- wen Guo. On the error analysis of 3d gaussian splat- ting and an optimal projection strategy. arXiv preprint arXiv:2402.00752, 2024. 3

    2024 arXiv

  14. [22]

    Gaussian- shader: 3d gaussian splatting with shading functions for re- flective surfaces

    Yingwenqi Jiang, Jiadong Tu, Yuan Liu, Xifeng Gao, Xi- aoxiao Long, Wenping Wang, and Yuexin Ma. Gaussian- shader: 3d gaussian splatting with shading functions for re- flective surfaces. In Proceedings of the IEEE/CVF Confer- ence on Computer Vision and Pattern Recognition , p...

    2024

  15. [23]

    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 Transactions on Graphics, 42 (4), 2023. 2, 3

    2023

  16. [24]

    3d gaussian splatting as markov chain monte carlo

    Shakiba Kheradmand, Daniel Rebain, Gopal Sharma, Wei- wei Sun, Jeff Tseng, Hossam Isack, Abhishek Kar, An- drea Tagliasacchi, and Kwang Moo Yi. 3d gaussian splatting as markov chain monte carlo. arXiv preprint arXiv:2404.09591, 2024. 5

    2024 arXiv

  17. [25]

    The fully automatic optical processing system CATENA at DLR

    Thomas Krauß, Pablo d’Angelo, Mathias Schneider, and Veronika Gstaiger. The fully automatic optical processing system CATENA at DLR. ISPRS International Archives of the Photogrammetry, Remote Sensing and Spatial Informa- tion Sciences, 40-1/W1:177–181, 2013. 2

    2013

  18. [26]

    2019 data fusion contest [technical committees]

    Bertrand Le Saux, Naoto Yokoya, Ronny Hansch, Myron Brown, and Greg Hager. 2019 data fusion contest [technical committees]. IEEE Geoscience and Remote Sensing Maga- zine, 7(1):103–105, 2019. 7

    2019

  19. [27]

    Co3d, a worldwide one one-meter accuracy dem for 2025

    Laurent Leb `egue, Eric Cazala-Hourcade, Florie Languille, St´ephanie Artigues, and Olivier Melet. Co3d, a worldwide one one-meter accuracy dem for 2025. The International Archives of the Photogrammetry, Remote Sensing and Spa- tial Information Sciences, XLIII-B1-2020:299–304, 2020. 1

    2025

  20. [28]

    Gs-ir: 3d gaussian splatting for inverse rendering

    Zhihao Liang, Qi Zhang, Ying Feng, Ying Shan, and Kui Jia. Gs-ir: 3d gaussian splatting for inverse rendering. arXiv preprint arXiv:2311.16473, 2023. 3

    2023 arXiv

  21. [29]

    Fisheye-gs: Lightweight and extensible gaus- sian splatting module for fisheye cameras

    Zimu Liao, Siyan Chen, Rong Fu, Yi Wang, Zhongling Su, Hao Luo, Linning Xu, Bo Dai, Hengjie Li, Zhilin Pei, et al. Fisheye-gs: Lightweight and extensible gaus- sian splatting module for fisheye cameras. arXiv preprint arXiv:2409.04751, 2024. 3

    2024 arXiv

  22. [30]

    Disparity estimation networks for aerial and high-resolution satellite images: A review

    Roger Mar ´ı, Thibaud Ehret, and Gabriele Facciolo. Disparity estimation networks for aerial and high-resolution satellite images: A review. Image Processing On Line, 12:501–526,

  23. [31]

    Sat-nerf: Learning multi-view satellite photogrammetry with transient objects and shadow modeling using rpc cameras

    Roger Mar ´ı, Gabriele Facciolo, and Thibaud Ehret. Sat-nerf: Learning multi-view satellite photogrammetry with transient objects and shadow modeling using rpc cameras. In Pro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 1311–1321, 2022. 2

    2022

  24. [32]

    Multi- date earth observation nerf: The detail is in the shadows

    Roger Mar ´ı, Gabriele Facciolo, and Thibaud Ehret. Multi- date earth observation nerf: The detail is in the shadows. In Proceedings of the IEEE/CVF Conference on Computer Vi- sion and Pattern Recognition, pages 2035–2045, 2023. 1, 2, 3, 4, 7, 8

    2023

  25. [33]

    Nerf in the wild: Neural radiance fields for uncon- strained photo collections

    Ricardo Martin-Brualla, Noha Radwan, Mehdi SM Sajjadi, Jonathan T Barron, Alexey Dosovitskiy, and Daniel Duck- worth. Nerf in the wild: Neural radiance fields for uncon- strained photo collections. In Proceedings of the IEEE/CVF conference on computer vision and pattern recogn...

    2021

  26. [34]

    N. Max. Optical models for direct volume rendering. IEEE Transactions on Visualization and Computer Graphics, 1(2): 99–108, 1995. 4

    1995

  27. [35]

    A new satellite imagery stereo pipeline designed for scal- ability, robustness and performance

    Julien Michel, Emmanuelle Sarrazin, David Youssefi, Myr- iam Cournet, Fabrice Buffe, Jean-Marc Delvit, Aur ´elie Em- ilien, Julien Bosman, Olivier Melet, and C ´eline L’Helguen. A new satellite imagery stereo pipeline designed for scal- ability, robustness and performance. ISP...

    2020

  28. [36]

    Nerf: Representing scenes as neural radiance fields for view syn- thesis

    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. Communications of the ACM , 65(1):99–106, 2021. 2

    2021

  29. [37]

    Reg- nerf: Regularizing neural radiance fields for view synthesis from sparse inputs

    Michael Niemeyer, Jonathan T Barron, Ben Mildenhall, Mehdi SM Sajjadi, Andreas Geiger, and Noha Radwan. Reg- nerf: Regularizing neural radiance fields for view synthesis from sparse inputs. In Proceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition,...

    2022

  30. [38]

    Pseudo pansharpening nerf for satellite image collec- tions

    Emilie Pic, Thibaud Ehret, Gabriele Facciolo, and Roger Mar´ı. Pseudo pansharpening nerf for satellite image collec- tions. In IGARSS 2024-2024 IEEE International Geoscience and Remote Sensing Symposium , pages 2650–2655. IEEE,

    2024

  31. [39]

    Theory iiib: General- ization in deep networks

    Tomaso Poggio, Qianli Liao, Brando Miranda, Andrzej Ban- burski, Xavier Boix, and Jack Hidary. Theory iiib: General- ization in deep networks. arXiv preprint arXiv:1806.11379,

    arXiv

  32. [40]

    Sat-mesh: Learning neural im- plicit surfaces for multi-view satellite reconstruction.Remote Sensing, 15(17):4297, 2023

    Yingjie Qu and Fei Deng. Sat-mesh: Learning neural im- plicit surfaces for multi-view satellite reconstruction.Remote Sensing, 15(17):4297, 2023. 2, 7

    2023

  33. [41]

    On the frequency-bias of coordinate-mlps

    Sameera Ramasinghe, Lachlan E MacDonald, and Simon Lucey. On the frequency-bias of coordinate-mlps. Ad- vances in Neural Information Processing Systems , 35:796– 809, 2022. 5

    2022

  34. [42]

    Lolnerf: Learn from one look

    Daniel Rebain, Mark Matthews, Kwang Moo Yi, Dmitry La- gun, and Andrea Tagliasacchi. Lolnerf: Learn from one look. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 1558–1567, 2022. 6

    2022

  35. [43]

    MicMac–a free, open-source solution for pho- togrammetry

    Ewelina Rupnik, Mehdi Daakir, and Marc Pierrot- Deseilligny. MicMac–a free, open-source solution for pho- togrammetry. Open Geospatial Data, Software and Stan- dards, 2(14), 2017. 2

    2017

  36. [44]

    Gir: 3d gaussian inverse ren- dering for relightable scene factorization

    Yahao Shi, Yanmin Wu, Chenming Wu, Xing Liu, Chen Zhao, Haocheng Feng, Jingtuo Liu, Liangjun Zhang, Jian Zhang, Bin Zhou, et al. Gir: 3d gaussian inverse ren- dering for relightable scene factorization. arXiv preprint arXiv:2312.05133, 2023. 3

    2023 arXiv

  37. [45]

    Srinivasan, Ben Mildenhall, Sara Fridovich-Keil, Nithin Raghavan, Utkarsh Singhal, Ravi Ra- mamoorthi, Jonathan T

    Matthew Tancik, Pratul P. Srinivasan, Ben Mildenhall, Sara Fridovich-Keil, Nithin Raghavan, Utkarsh Singhal, Ravi Ra- mamoorthi, Jonathan T. Barron, and Ren Ng. Fourier fea- tures let networks learn high frequency functions in low di- mensional domains. NeurIPS, 2020. 5

    2020

  38. [46]

    A comprehensive study of the rational function model for photogrammetric processing

    C Vincent Tao and Yong Hu. A comprehensive study of the rational function model for photogrammetric processing. Photogrammetric engineering and remote sensing , 67(12): 1347–1358, 2001. 2

    2001

  39. [47]

    Regression shrinkage and selection via the lasso

    Robert Tibshirani. Regression shrinkage and selection via the lasso. Journal of the Royal Statistical Society Series B: Statistical Methodology, 58(1):267–288, 1996. 5

    1996

  40. [48]

    Deep image prior

    Dmitry Ulyanov, Andrea Vedaldi, and Victor Lempitsky. Deep image prior. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 9446–9454,

  41. [49]

    Geological Survey

    U.S. Geological Survey. The universal transverse mercator (utm) grid. Technical report, U.S. Geological Survey, Re- ston, V A, 2001. Report. 3

    2001

  42. [50]

    Casting curved shadows on curved surfaces

    Lance Williams. Casting curved shadows on curved surfaces. In Proceedings of the 5th annual conference on Computer graphics and interactive techniques , pages 270–274, 1978. 4

    1978

  43. [51]

    Recent advances in 3d gaussian splatting

    Tong Wu, Yu-Jie Yuan, Ling-Xiao Zhang, Jie Yang, Yan- Pei Cao, Ling-Qi Yan, and Lin Gao. Recent advances in 3d gaussian splatting. Computational Visual Media, 10(4):613– 642, 2024. 3

    2024

  44. [52]

    Sparsegs: Real- time 360 {\deg} sparse view synthesis using gaussian splat- ting

    Haolin Xiong, Sairisheek Muttukuru, Rishi Upadhyay, Pradyumna Chari, and Achuta Kadambi. Sparsegs: Real- time 360 {\deg} sparse view synthesis using gaussian splat- ting. arXiv preprint arXiv:2312.00206, 2023. 3

    2023 arXiv

  45. [53]

    Hierarchical deep stereo matching on high- resolution images

    Gengshan Yang, Joshua Manela, Michael Happold, and Deva Ramanan. Hierarchical deep stereo matching on high- resolution images. In 2019 IEEE/CVF Conference on Com- puter Vision and Pattern Recognition (CVPR), pages 5510– 5519, 2019. 2

    2019

  46. [54]

    Feihu Zhang, Victor Prisacariu, Ruigang Yang, and Philip H.S. Torr. GA-Net: Guided aggregation net for end- to-end stereo matching. In 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , pages 185–194, 2019. 2

    2019

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.