Pith. sign in

REVIEW 3 major objections 4 minor 51 references

Construction of Digital Terrain Maps from Multi-view Satellite Imagery using Neural Volume Rendering

T0 review · 3 major / 4 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read A neural volume field optimized against multi-view photometric consistency can recover a digital terrain model directly from satellite imagery, at precision near image resolution.

desk verdict Useful application of neural volume rendering to satellite DTM generation, but the headline accuracy claim is only validated against MVS reference DTMs and the supplied full text is unreadable, so treat the numbers cautiously. read the letter →

arxiv 2508.01386 v1 pith:2HGMBN3M submitted 2025-08-02 cs.CV

Josef X. Biberstein , Guilherme Cavalheiro , Juyeop Han , Sertac Karaman This is my paper
classification cs.CV
keywords neuralterrainmapsdigitalmodelsvolumerenderingmulti-viewsatelliteimageryphotometricconsistencyMarsexplorationreconstruction
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

This paper aims to show that digital terrain models can be built directly from multi-view satellite imagery by adapting neural volume rendering, without the depth priors or manual preprocessing that multi-view stereo pipelines require. The method, called neural terrain maps (NTM), optimizes a neural volume field against the photometric consistency of the observed image pixels, using only the ray locus of each pixel as geometric input. If the claim holds, high-resolution terrain maps for Earth and planetary surfaces could be produced more simply, which matters for spacecraft landing, rover navigation, and planetary exploration. The paper demonstrates the approach on synthetic scenes and on real Earth and Mars imagery covering roughly 100 square kilometers, reporting terrain precision near the satellite image resolution even with imperfect camera intrinsics and extrinsics.

What carries the argument

The central object is the neural terrain map itself: a neural volume field in which the scene is represented so that volume rendering along each pixel's viewing ray produces the pixel's color, and the field is optimized by comparing rendered colors with the observed multi-view images. Because the optimization is driven purely by photometric consistency, the field must simultaneously settle on plausible surface heights and surface texture. The paper emphasizes that the only geometric input is the locus of each image pixel, which is what lets the method tolerate imperfect camera intrinsics and extrinsics and avoids depth or structural priors. The rendered-versus-observed color comparison is the mechanism that carries the argument from images to terrain.

What would settle it

On a synthetic scene with known ground truth and large textureless regions, generate several different height fields that produce identical rendered images under the same cameras; if NTM picks one of the wrong fields while matching photometric error, the central claim is falsified. A real-world check is to compare NTM terrain against LiDAR ground truth over sand, ice, or smooth regolith, where color alone gives little height information and the near-image-resolution accuracy prediction should fail.

Watch

Extended reading notes

Core claim

The paper's central claim is that a neural volume field optimized only against multi-view photometric consistency can recover a textured digital terrain map from satellite images. NTM does not compute depth, run stereo matching, or rely on structural priors; it takes the ray locus for each image pixel and learns the terrain surface whose rendered appearance matches the observed views. Evaluated against existing multi-view-stereo DTMs, the method's terrain predictions reach precision almost equal to the resolution of the satellite imagery, and this accuracy persists when camera intrinsics and extrinsics are imperfect. The result is a terrain map that is effectively a by-product of a differentiable rendering optimization rather than of a classical multi-view stereo pipeline.

Load-bearing premise

The load-bearing premise is that matching pixel colors across satellite views uniquely determines ground height, so a wrong but photometrically plausible surface cannot fool the optimization.

Editorial extensions

If this is right

  • DTM production no longer needs manual image preprocessing or depth priors; a single photometric optimization over ray loci replaces the stereo pipeline.
  • Terrain precision at near image resolution means small surface features visible in satellite pixels transfer to the terrain map, which is useful for landing-site selection and rover navigation.
  • The method applies to planetary bodies where only multi-view satellite images exist and no high-quality ground-truth DTM is available, as demonstrated on Mars.
  • Imperfect camera intrinsics and extrinsics do not destroy accuracy, suggesting the approach tolerates realistic errors in orbital imaging metadata.

Reading between the lines

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

  • The paper leaves untested whether the same photometric-only optimization would recover terrain under varying illumination or temporally changing surface albedo; a natural extension is to make the neural volume field time- or illumination-dependent.
  • A further extension would jointly refine camera poses along with the terrain field, potentially removing the last need for precise orbital metadata.
  • The method's limit should appear on textureless terrain such as smooth sand, ice, or regolith, where color error underdetermines height; a synthetic ablation with uniform albedo would test that boundary.
  • Because the terrain map is a learned continuous field, downstream users could query elevation at arbitrary resolutions instead of resampling a fixed grid DTM, though the paper does not demonstrate this.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 4 minor

Summary. The paper proposes Neural Terrain Maps (NTM), a method that adapts neural volume rendering to reconstruct digital terrain models (DTMs) directly from multi-view satellite imagery. The method requires only the ray locus for each image pixel and does not use depth or other structural priors. Experiments are reported on synthetic and real satellite data from Earth and Mars covering roughly 100 km^2 scenes, and the recovered terrain is compared to DTMs produced by traditional multi-view stereo pipelines. The abstract claims that terrain prediction precision is almost equal to the image resolution, even under imperfect camera intrinsics and extrinsics.

Significance. If the precision claim is established, NTM would be a useful alternative to conventional multi-view stereo for planetary DTM production, with the potential to reduce manual preprocessing and avoid explicit depth priors. The demonstrated applicability to Earth and Mars at the 100 km^2 scale is relevant to planetary exploration and terrain-relative navigation. The method's conceptual simplicity and reliance on photometric consistency alone are attractive. However, the evaluation standard used in the abstract—comparison against MVS-derived reference DTMs—leaves the absolute accuracy claim unverified, because both the method and the reference share the same photometric-consistency objective and could share systematic biases from camera model errors, atmospheric effects, and narrow-baseline geometry.

major comments (3)
  1. [Abstract, §5 (Experiments)] The headline claim that terrain prediction precision is 'almost equal to the resolution of the satellite images' is supported only by comparison with 'existing high-quality DTMs produced using traditional multi-view stereo pipelines.' These references are not independent ground truth: any systematic height bias caused by shared pose errors, radiometric calibration, atmospheric scattering, or narrow-baseline geometry will be present in both the NTM output and the reference DTM, inflating the apparent accuracy. The paper must report errors against independent ground truth (e.g., lidar, radar altimetry, or synthetic scenes with known height) and must state the known accuracy or error bars of the reference DTMs. Without this, the central accuracy claim is not established.
  2. [§3 (Method)] The method relies solely on multi-view photometric consistency and explicitly forgoes depth or structural priors, but the paper does not analyze the identifiability of terrain height from the photometric loss. A neural volume field can compensate for an incorrect surface by adjusting view-dependent radiance, and errors in camera poses can be traded against height. The claimed robustness to imperfect intrinsics and extrinsics is therefore meaningful only if evaluated against absolute ground truth. The authors should provide a synthetic experiment with known ground truth and controlled perturbations of camera parameters, demonstrating that height errors remain near image resolution under these conditions.
  3. [§5 (Experiments)] The supplied text does not allow extraction of numerical results, error bars, or ablations. To substantiate the central claim and enable reproducibility, the paper should include per-scene error metrics (e.g., mean and RMSE in meters relative to ground truth and to reference DTMs), statistics across multiple scenes, and ablations of key components (e.g., removing view-dependent color, varying the number of input views, varying camera pose noise). It should also state explicitly whether any hyperparameters were tuned on the evaluation datasets, since the abstract leaves this risk unresolved.
minor comments (4)
  1. [Full text] The manuscript text as supplied to the referee is heavily corrupted by character-encoding errors, rendering equations, tables, and most experimental details unreadable. The authors should ensure a clean PDF or TeX source is provided for review.
  2. [Abstract] The phrase 'precision of terrain prediction almost equal to the resolution of the satellite images' should be defined precisely in the abstract or introduction, e.g., as RMSE in meters relative to the ground-sample distance (GSD), so that the claim is falsifiable.
  3. [Introduction] The paper should compare against prior neural rendering approaches to terrain or height-field reconstruction (e.g., Sat-NeRF and related work) rather than only traditional multi-view stereo, to place the contribution in context.
  4. [§3 (Method)] The statement that the method 'only requires the locus for each image pixel' should explain how this locus is obtained and whether it is assumed to be known exactly or estimated; this is important for assessing the impact of camera calibration errors.

Circularity Check

0 steps flagged · score 0.0 of 10

No load-bearing circularity: the terrain field is optimized from multi-view photometric consistency, and the MVS reference DTMs enter only as an evaluation comparison, not as fitted inputs or defining equations.

full rationale

The paper's derivation chain is the standard neural volume-rendering one: a height field is optimized by comparing rendered pixel colors with the observed multi-view satellite images under a photometric loss, using the usual transmittance-weighted volume rendering equations. The abstract's evaluation statement, 'We evaluate the accuracy of our output terrain maps by comparing with existing high-quality DTMs produced using traditional multi-view stereo pipelines,' places the MVS DTMs strictly on the evaluation side: they are not used in the training loss, nor does any equation in the supplied extract define the terrain in terms of those DTMs. The claim that terrain precision is near image resolution is therefore a measured agreement with an external product, not a fitted parameter renamed as a prediction. The skeptical concern that NTM and MVS reference DTMs may share photometric or camera-model biases is a correctness and external-validity limitation, not a circularity: even if both outputs share a bias, the NTM output is not forced to equal the reference by construction. I also found no load-bearing self-citation chain or imported uniqueness theorem in the supplied text; the volume-rendering formulation is presented directly and does not reduce to a prior result by the same authors. The abstract's silence on hyperparameter tuning is a reporting gap, not a circular step. Accordingly, no significant circularity is identified.

Assumptions & free parameters 0 free parameters · 3 assumptions · 0 invented entities

The abstract does not disclose enough detail to enumerate specific free parameters. The method uses standard volume rendering and a neural network whose weights are optimized against photometric loss; no new physical entities are introduced.

assumptions (3)
  • domain assumption The multi-view photometric loss is a sufficient training signal to recover terrain height.
    The abstract states the method 'only requires the locus for each image pixel and does not rely on depth or any other structural priors', implying this assumption.
  • domain assumption Terrain can be represented as an opaque, single-valued surface within the neural volume field.
    The output is a DTM, a single height per cell, which requires the learned density field to concentrate on one surface.
  • standard math Standard neural volume rendering equations apply to satellite imagery.
    The method is an adaptation of neural volume rendering, relying on the standard camera ray integration model.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Construction of Digital Terrain Maps from Multi-view Satellite Imagery using Neural Volume Rendering." pith.science (2026). https://pith.science/paper/2HGMBN3M

@misc{pith2026250801386,
  author       = {Pith},
  title        = {Pith review of: Construction of Digital Terrain Maps from Multi-view Satellite Imagery using Neural Volume Rendering},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/2HGMBN3M}},
  note         = {Machine review of arXiv:2508.01386}
}
abstract

Digital terrain maps (DTMs) are an important part of planetary exploration, enabling operations such as terrain relative navigation during entry, descent, and landing for spacecraft and aiding in navigation on the ground. As robotic exploration missions become more ambitious, the need for high quality DTMs will only increase. However, producing DTMs via multi-view stereo pipelines for satellite imagery, the current state-of-the-art, can be cumbersome and require significant manual image preprocessing to produce satisfactory results. In this work, we seek to address these shortcomings by adapting neural volume rendering techniques to learn textured digital terrain maps directly from satellite imagery. Our method, neural terrain maps (NTM), only requires the locus for each image pixel and does not rely on depth or any other structural priors. We demonstrate our method on both synthetic and real satellite data from Earth and Mars encompassing scenes on the order of $100 \textrm{km}^2$. We evaluate the accuracy of our output terrain maps by comparing with existing high-quality DTMs produced using traditional multi-view stereo pipelines. Our method shows promising results, with the precision of terrain prediction almost equal to the resolution of the satellite images even in the presence of imperfect camera intrinsics and extrinsics.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

51 extracted references · 39 canonical work pages

  1. [1]

    write newline

    " write newline "" initialize.prev.this.status FUNCTION begin.bib " write newline preamble empty 'skip preamble write newline if " thebibliography " longest.label * " " * write newline " [1] #1 " write newline " url@samestyle " write newline " " write newline " [2] #2 " write newline " =0pt " write newline " " ALTinterwordstretchfactor * " " * write newli...

  2. [2]

    Q a 2q #B 3R b Cr j4ݞù 2K xP

    4000 4000 100 4000 4000 500 `\.=1000 = #1 9 12 #1 * [1] 0pt [0pt][0pt] #1 * \| ** #1 \@IEEEauthorblockNstyle \@IEEEcompsocnotconfonly \@IEEEcompsocconfonly \@IEEEauthorblockAstyle \@IEEEcompsocnotconfonly \@IEEEcompsocconfonly \@IEEEcompsocconfonly \@IEEEauthordefaulttextstyle \@IEEEcompsocnotconfonly \@IEEEauthorblockconfadjspace -0.25em \@IEEEcompsoccon...

    work page

  3. [3]

    Balaram, M

    J. Balaram, M. Aung, and M. P. Golombek The ingenuity helicopter on the perseverance rover vol. 217, no. 4, p. 56. [Online]. Available: https://doi.org/10.1007/s11214-021-00815-w

    work page doi:10.1007/s11214-021-00815-w

  4. [4]

    R. D. Lorenz, E. P. Turtle, J. W. Barnes, and M. G. Trainer Dragonfly: A rotorcraft lander concept for scientific exploration at titan vol. 34, no. 3

    work page

  5. [5]

    T. S. Vaquero et al. EELS : Autonomous snake-like robot with task and motion planning capabilities for ice world exploration vol. 9, no. 88, p. eadh8332, publisher: American Association for the Advancement of Science. [Online]. Available: https://www.science.org/doi/full/10.1126/scirobotics.adh8332

    doi:10.1126/scirobotics.adh8332

  6. [6]

    de la Croix et al

    J.-P. de la Croix et al. Multi-agent autonomy for space exploration on the CADRE lunar technology demonstration In 2024 IEEE Aerospace Conference , pp. 1--14, ISSN : 1095-323X. [Online]. Available: https://ieeexplore.ieee.org/abstract/document/10521425

    work page arXiv 2024

  7. [7]

    J. Schonfeld Summary of the contracted deliveries of NASA payloads to the moon via commercial lunar payload services ( CLPS ) In 2023 IEEE International Conference on Systems, Man, and Cybernetics ( SMC ) , pp. 863--866, ISSN : 2577-1655. [Online]. Available: https://ieeexplore.ieee.org/abstract/document/10394001

    work page arXiv 2023

  8. [8]

    K. A. Michaelson, F. Wang, and R. Zanetti Terrain-relative navigation with neuro-inspired elevation encoding vol. 60, no. 3, pp. 3368--3378. [Online]. Available: https://ieeexplore.ieee.org/document/10438923

    work page arXiv

Show all 51 references
  1. [9]

    Y. Jung, S. Lee, and H. Bang Digital terrain map based safe landing site selection for planetary landing vol. 56, no. 1, pp. 368--380. [Online]. Available: https://ieeexplore.ieee.org/document/8701525

  2. [10]

    Kim and S

    J. Kim and S. Sukkarieh Autonomous airborne navigation in unknown terrain environments vol. 40, no. 3, pp. 1031--1045. [Online]. Available: https://ieeexplore.ieee.org/document/1337472

  3. [11]

    Bell III et al

    J. Bell III et al. Calibration and performance of the mars reconnaissance orbiter context camera ( CTX ) vol. 8, pp. 1--14

  4. [12]

    Tancik et al

    M. Tancik et al. Nerfstudio: A modular framework for neural radiance field development In Special Interest Group on Computer Graphics and Interactive Techniques Conference Conference Proceedings, pp. 1--12. [Online]. Available: http://arxiv.org/abs/2302.04264

    arXiv

  5. [13]

    P. Wang, L. Liu, Y. Liu, C. Theobalt, T. Komura, and W. Wang NeuS : Learning neural implicit surfaces by volume rendering for multi-view reconstruction . [Online]. Available: http://arxiv.org/abs/2106.10689

    arXiv

  6. [14]

    A. Dai, S. Gupta, and G. Gao Neural elevation models for terrain mapping and path planning . [Online]. Available: http://arxiv.org/abs/2405.15227

    arXiv

  7. [15]

    R. A. Beyer, O. Alexandrov, and S. McMichael The ames stereo pipeline: NASA 's open source software for deriving and processing terrain data vol. 5, no. 9, pp. 537--548, \_eprint: https://onlinelibrary.wiley.com/doi/pdf/10.1029/2018EA000409. [Online]. Available: https://online...

    doi

  8. [16]

    de Franchis, E

    C. de Franchis, E. Meinhardt-Llopis, J. Michel, J.-M. Morel, and G. Facciolo An automatic and modular stereo pipeline for pushbroom images vol. II -3, pp. 49--56, conference Name: ISPRS Technical Commission III Symposium (Volume II -3) - 5–7 September 2014, Zurich, Switz...

    2014

  9. [17]

    J. L. Schönberger, E. Zheng, M. Pollefeys, and J.-M. Frahm Pixelwise view selection for unstructured multi-view stereo In European Conference on Computer Vision ( ECCV )

  10. [18]

    J. L. Schönberger and J.-M. Frahm Structure-from-motion revisited In Conference on Computer Vision and Pattern Recognition ( CVPR )

  11. [19]

    H.-Y. Lee, T. Kim, W. Park, and H. K. Lee Extraction of digital elevation models from satellite stereo images through stereo matching based on epipolarity and scene geometry vol. 21, no. 9, pp. 789--796. [Online]. Available: https://www.sciencedirect.com/science/article/pii/S0...

  12. [20]

    M. Wang, F. Hu, and J. Li Epipolar resampling of linear pushbroom satellite imagery by a new epipolarity model vol. 66, no. 3, pp. 347--355. [Online]. Available: https://www.sciencedirect.com/science/article/pii/S0924271611000050

  13. [21]

    J. Oh, W. Lee, C. Toth, D. Grejner-Brzezinska, and C. Lee A piecewise approach to epipolar resampling of pushbroom satellite images based on RPC vol. 76, pp. 1353--1363

  14. [22]

    Zhang, V

    F. Zhang, V. Prisacariu, R. Yang, and P. H. Torr GA -net: Guided aggregation net for end-to-end stereo matching In 2019 IEEE / CVF Conference on Computer Vision and Pattern Recognition ( CVPR ) , pp. 185--194, ISSN : 2575-7075. [Online]. Available: https://ieeexplore.ieee.org/...

    2019

  15. [23]

    X. Gu, Z. Fan, Z. Dai, S. Zhu, F. Tan, and P. Tan Cascade cost volume for high-resolution multi-view stereo and stereo matching . [Online]. Available: http://arxiv.org/abs/1912.06378

    1912 arXiv

  16. [24]

    Singh and N

    P. Singh and N. Komodakis Cloud-gan: Cloud removal for sentinel-2 imagery using a cyclic consistent generative adversarial networks In IGARSS 2018 - 2018 IEEE International Geoscience and Remote Sensing Symposium . IEEE , pp. 1772--1775. [Online]. Available: https://ieeexplore...

    2018

  17. [25]

    Pan Cloud removal for remote sensing imagery vai spatial attention generative adversarial network

    H. Pan Cloud removal for remote sensing imagery vai spatial attention generative adversarial network

  18. [26]

    P. Wu, Z. Pan, H. Tang, and Y. Hu Cloudformer: A cloud-removal network combining self-attention mechanism and convolution vol. 14, no. 23, p. 6132, number: 23 Publisher: Multidisciplinary Digital Publishing Institute. [Online]. Available: https://www.mdpi.com/2072-4292/14/23/6132

  19. [27]

    Christopoulos, V

    D. Christopoulos, V. Ntouskos, and K. Karantzalos CloudTran ++: Improved cloud removal from multi-temporal satellite images using axial transformer networks vol. 17, no. 1, p. 86, number: 1 Publisher: Multidisciplinary Digital Publishing Institute. [Online]. Available: https:/...

  20. [28]

    Alexandrov and R

    O. Alexandrov and R. A. Beyer Multiview shape-from-shading for planetary images vol. 5, no. 10, pp. 652--666, \_eprint: https://agupubs.onlinelibrary.wiley.com/doi/pdf/10.1029/2018EA000390. [Online]. Available: https://onlinelibrary.wiley.com/doi/abs/10.1029/2018EA000390

    doi

  21. [29]

    Langguth, K

    F. Langguth, K. Sunkavalli, S. Hadap, and M. Goesele Shading-aware multi-view stereo In Computer Vision – ECCV 2016 , B. Leibe, J. Matas, N. Sebe, and M. Welling, Eds. Springer International Publishing, pp. 469--485

    2016

  22. [30]

    Bosch, K

    M. Bosch, K. Foster, G. Christie, S. Wang, G. D. Hager, and M. Brown Semantic stereo for incidental satellite images In 2019 IEEE Winter Conference on Applications of Computer Vision ( WACV ) . IEEE , pp. 1524--1532. [Online]. Available: https://ieeexplore.ieee.org/document/8659180/

    2019

  23. [31]

    Facciolo, C

    G. Facciolo, C. De Franchis, and E. Meinhardt-Llopis Automatic 3d reconstruction from multi-date satellite images In 2017 IEEE Conference on Computer Vision and Pattern Recognition Workshops ( CVPRW ) . IEEE , pp. 1542--1551. [Online]. Available: https://ieeexplore.ieee.org/do...

    2017

  24. [32]

    Mildenhall, P

    B. Mildenhall, P. P. Srinivasan, M. Tancik, J. T. Barron, R. Ramamoorthi, and R. Ng NeRF : Representing scenes as neural radiance fields for view synthesis . [Online]. Available: http://arxiv.org/abs/2003.08934

    2003 arXiv

  25. [33]

    Müller, A

    T. Müller, A. Evans, C. Schied, and A. Keller Instant neural graphics primitives with a multiresolution hash encoding. [Online]. Available: https://arxiv.org/abs/2201.05989v2

    arXiv

  26. [34]

    Lin, W.-C

    C.-H. Lin, W.-C. Ma, A. Torralba, and S. Lucey BARF : Bundle-adjusting neural radiance fields . [Online]. Available: http://arxiv.org/abs/2104.06405

    arXiv

  27. [35]

    Martin-Brualla, N

    R. Martin-Brualla, N. Radwan, M. S. M. Sajjadi, J. T. Barron, A. Dosovitskiy, and D. Duckworth NeRF in the wild: Neural radiance fields for unconstrained photo collections . [Online]. Available: http://arxiv.org/abs/2008.02268

    2008 arXiv

  28. [36]

    Y. Wang, Q. Han, M. Habermann, K. Daniilidis, C. Theobalt, and L. Liu NeuS 2: Fast learning of neural implicit surfaces for multi-view reconstruction . [Online]. Available: http://arxiv.org/abs/2212.05231

    arXiv

  29. [37]

    Logothetis, I

    F. Logothetis, I. Budvytis, and R. Cipolla A neural height-map approach for the binocular photometric stereo problem In 2024 IEEE / CVF Winter Conference on Applications of Computer Vision ( WACV ) . IEEE , pp. 1557--1566. [Online]. Available: https://ieeexplore.ieee.org/docum...

    2024

  30. [38]

    Derksen and D

    D. Derksen and D. Izzo Shadow neural radiance fields for multi-view satellite photogrammetry . [Online]. Available: http://arxiv.org/abs/2104.09877

    arXiv

  31. [39]

    R. Mari, G. Facciolo, and T. Ehret Sat- NeRF : Learning multi-view satellite photogrammetry with transient objects and shadow modeling using RPC cameras In 2022 IEEE / CVF Conference on Computer Vision and Pattern Recognition Workshops ( CVPRW ) . IEEE , pp. 1310--1320. [Onlin...

    2022

  32. [40]

    R. Marí, G. Facciolo, and T. Ehret Multi-date earth observation NeRF : The detail is in the shadows In 2023 IEEE / CVF Conference on Computer Vision and Pattern Recognition Workshops ( CVPRW ) . IEEE , pp. 2035--2045. [Online]. Available: https://ieeexplore.ieee.org/document/10209050/

    2023

  33. [41]

    Prosvetov, A

    A. Prosvetov, A. Govorov, M. Pupkov, A. Andreev, and V. Nazarov Illuminating the moon: Reconstruction of lunar terrain using photogrammetry, neural radiance fields, and gaussian splatting vol. 52, p. 100953. [Online]. Available: https://linkinghub.elsevier.com/retrieve/pii/S22...

  34. [42]

    Hansen, C

    M. Hansen, C. Adams, T. Fong, and D. Wettergreen Analyzing the effectiveness of neural radiance fields for geometric modeling of lunar terrain In 2024 IEEE Aerospace Conference , pp. 1--12, ISSN : 1095-323X. [Online]. Available: https://ieeexplore.ieee.org/abstract/document/10521163

    2024

  35. [43]

    Van Kints Neural radiance field methods for satellite imagery of polar climates ISBN : 9798346895510

    E. Van Kints Neural radiance field methods for satellite imagery of polar climates ISBN : 9798346895510. [Online]. Available: https://www.proquest.com/docview/3152433988/abstract/E723FE7B79FA4F93PQ/1

  36. [44]

    R. Li, M. Tancik, and A. Kanazawa NerfAcc : A general NeRF acceleration toolbox . [Online]. Available: http://arxiv.org/abs/2210.04847

    arXiv

  37. [45]

    Nichols, A

    D. Nichols, A. Kahle, H. Fujisada, H. Tsu, and Y. Yamaguchi ASTER instrument design and science objectives publisher: UNKNOWN . [Online]. Available: https://ntrs.nasa.gov/citations/20210004760

  38. [46]

    [Online]

    Kelvin Rodriguez and Astrogeology Science Center Integrated software for imagers and spectrometers ( ISIS ) 8.3.0 . [Online]. Available: https://code.usgs.gov/astrogeology/isis/-/tags/8.3.0

  39. [47]

    A. C. Paquette Abstraction layer for ephemerides . [Online]. Available: https://code.usgs.gov/astrogeology/ale/-/releases/0.9.0

  40. [48]

    F. J. Calef III and T. Parker MSL gale merged orthophoto mosaic . [Online]. Available: https://astrogeology.usgs.gov/search/map/mars_msl_gale_merged_dem_1m

  41. [49]

    R. L. Fergason, D. M. Galuszka, T. M. Hare, D. P. Mayer, and B. L. Redding Mars 2020 terrain relative navigation CTX DTM mosaic . [Online]. Available: https://astrogeology.usgs.gov/search/map/Mars/Mars2020/JEZ_ctx_B_soc_008_DTM_MOLAtopography_DeltaGeoid_20m_Eqc_latTs0_lon0

    2020

  42. [50]

    [Online]

    NASA/METI/AIST/Japan Spacesystems and U.S./Japan ASTER Science Team ASTER global digital elevation model v003 . [Online]. Available: https://www.earthdata.nasa.gov/data/catalog/lpcloud-astgtm-003

  43. [51]

    Müller tiny-cuda-nn

    T. Müller tiny-cuda-nn . [Online]. Available: https://github.com/NVlabs/tiny-cuda-nn

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.