pith. sign in

arxiv: 2605.22556 · v1 · pith:OZ7D67DWnew · submitted 2026-05-21 · 💻 cs.LG

ImplicitTerrainV2: Wavelet-Guided Spatially Adaptive Neural Terrain Representation

Haoan Feng , Xin Xu , Leila De Floriani This is my paper

Pith reviewed 2026-05-22 07:30 UTC · model grok-4.3

classification 💻 cs.LG
keywords implicit neural representationsdigital elevation modelswavelet complexity fieldspatial adaptivitygradient matchingmodel compressionterrain representationGIS
0
0 comments X

The pith

ImplicitTerrainV2 uses a wavelet complexity field to adapt neural terrain models spatially, achieving higher fidelity with fewer parameters and faster training.

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

The paper tries to establish that prior implicit neural representations for terrain can be made practical by adding explicit frequency control and spatial adaptivity. It does so through a wavelet complexity field that localizes higher network capacity and denser sampling to complex regions, plus gradient matching to enforce smooth manifold structure. These changes, followed by quantization and entropy coding, produce a compact neural format. A sympathetic reader would care because digital elevation models are foundational to GIS analysis, and this format removes the need for interpolation on off-grid points while delivering accurate derivatives at low storage cost. Experiments show the result outperforms earlier implicit terrain models on quality and efficiency metrics.

Core claim

ImplicitTerrainV2 advances terrain INRs toward a compact, efficient neural terrain data format by combining a spectral control mechanism with wavelet-guided spatial adaptivity, derivative-aware supervision, and post-training model compression. At its core, a wavelet complexity field derives spatially-adaptive frequency masks from analytically computed wavelet coefficients, localizing high-frequency capacity to complex terrain regions. The same field guides complexity-aware adaptive sampling that concentrates training in high-complexity regions, while gradient matching applies extra supervision to enforce the smooth manifold structure of terrain DEMs for improved derivative fidelity. Post-

What carries the argument

The wavelet complexity field (WCF), which derives spatially-adaptive frequency masks from analytically computed wavelet coefficients to localize high-frequency model capacity and training samples to complex terrain regions.

If this is right

  • Reaches 66.25 dB end-to-end PSNR on Swiss terrain tiles, a 5.70 dB improvement over prior implicit terrain work.
  • Uses 3.2 times fewer parameters and trains in 55 seconds per tile on a single GPU.
  • Compresses to 1.23 bits per pixel after mixed-precision quantization and entropy coding, with only a 0.28 dB PSNR drop.
  • Supports off-grid point queries, closed-form derivative evaluation, and resolution-independent reconstruction.
  • Performs competitively with established DEM codecs in rate-distortion while adding continuous representation capabilities.

Where Pith is reading between the lines

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

  • The same wavelet-guided adaptivity could be tested on other spatially varying fields such as satellite imagery or fluid flow maps where frequency content is non-uniform.
  • GIS pipelines might replace finite-difference derivative calculations with the closed-form outputs from these models for slope and curvature analysis.
  • Hybrid storage schemes that apply the neural format only to high-complexity subregions and keep raster data elsewhere could be evaluated for further efficiency gains.
  • The method's resolution independence suggests direct use in multi-scale terrain rendering without precomputing separate pyramid levels.

Load-bearing premise

The wavelet complexity field derived from analytically computed wavelet coefficients correctly identifies and localizes high-frequency terrain regions so that adaptive frequency masks and sampling improve fidelity without introducing bias or missing critical features.

What would settle it

If disabling the wavelet-guided adaptive frequency masks and sampling on the 50 Swiss terrain tiles produces no PSNR gain or visible artifacts in high-frequency areas, the contribution of the WCF guidance would be falsified.

Figures

Figures reproduced from arXiv: 2605.22556 by Haoan Feng, Leila De Floriani, Xin Xu.

Figure 1
Figure 1. Figure 1: Overview of the ImplicitTerrainV2 pipeline. A terrain DEM is decomposed into a smooth shape model [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Mask comparison on a terrain tile. Hash-grid masks [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 4
Figure 4. Figure 4: Qualitative comparison on three terrain tiles. The [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Rate-distortion comparison across bitrates. Red [PITH_FULL_IMAGE:figures/full_fig_p009_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Topographic features and topological skeleton derived from ImplicitTerrainV2’s shape model [PITH_FULL_IMAGE:figures/full_fig_p010_6.png] view at source ↗
Figure 1
Figure 1. Figure 1: Convergence on 5 tiles: our single-scale WCF [PITH_FULL_IMAGE:figures/full_fig_p013_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Absolute-error maps on a tile with 155 m relief (shared colorscale in m). [PITH_FULL_IMAGE:figures/full_fig_p014_2.png] view at source ↗
read the original abstract

Digital elevation models (DEMs) underpin terrain analysis in Geographic Information Systems (GIS), but in their common raster form, they rely on interpolation for off-grid sampling and finite-difference operators for derivative-based analysis. Implicit neural representations (INRs) offer a continuous alternative, but prior terrain INRs lack explicit frequency control, neglect the gradient structure of terrain, and remain too large and costly to train for practical deployment. We present ImplicitTerrainV2, which advances terrain INRs toward a compact, efficient neural terrain data format by combining a spectral control mechanism with wavelet-guided spatial adaptivity, derivative-aware supervision, and post-training model compression. At its core, a wavelet complexity field (WCF) derives spatially-adaptive frequency masks from analytically computed wavelet coefficients, localizing high-frequency capacity to complex terrain regions. The same field guides complexity-aware adaptive sampling that concentrates training in high-complexity regions, while gradient matching applies extra supervision to enforce the smooth manifold structure of terrain DEMs for improved derivative fidelity. Post-training mixed-precision quantization and entropy coding reduce storage to 1.23 bpp with a 0.28 dB PSNR drop. On 50 Swiss terrain tiles, ImplicitTerrainV2 reaches 66.25 dB end-to-end PSNR, improving over the prior work by 5.70 dB while using 3.2x fewer parameters and training in 55 s per tile on a single GPU. Our compressed neural format is competitive with several established DEM codecs in rate-distortion performance, while additionally supporting off-grid point queries, closed-form derivative evaluation, and resolution-independent reconstruction, which may benefit many downstream GIS applications.

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

1 major / 2 minor

Summary. The manuscript introduces ImplicitTerrainV2, an implicit neural representation for digital elevation models that incorporates a wavelet complexity field (WCF) derived from analytically computed wavelet coefficients to produce spatially adaptive frequency masks and sampling densities. It adds derivative-aware supervision to enforce terrain manifold structure and applies post-training mixed-precision quantization plus entropy coding. On 50 Swiss terrain tiles the method is reported to reach 66.25 dB end-to-end PSNR (5.70 dB above prior work), 3.2× fewer parameters, 55 s training per tile, and 1.23 bpp storage while remaining competitive with established DEM codecs and supporting off-grid queries, closed-form derivatives, and resolution-independent reconstruction.

Significance. If the central claims are substantiated, the work supplies a compact continuous alternative to raster DEMs that directly addresses frequency control, gradient fidelity, and storage cost. The combination of wavelet-guided adaptivity with derivative matching and compression yields measurable gains in PSNR and parameter count while adding capabilities (off-grid sampling, analytic derivatives) that raster formats lack. Evaluation across 50 tiles and explicit comparison to both prior INRs and traditional codecs strengthens the practical relevance for GIS applications.

major comments (1)
  1. [Abstract and §3] Abstract and §3 (WCF definition): the headline 5.70 dB PSNR improvement, 3.2× parameter reduction, and derivative fidelity rest on the assumption that the wavelet complexity field correctly localizes high-frequency terrain structure. No ablation isolating the WCF contribution, no overlap or precision-recall metrics against ground-truth frequency maps, and no failure-case analysis on the 50 tiles are provided, leaving the load-bearing adaptivity mechanism unverified.
minor comments (2)
  1. [Abstract] The abstract states 'analytically computed wavelet coefficients' without naming the wavelet family, decomposition depth, or exact complexity metric; adding these details would improve reproducibility.
  2. [§4] Table captions and axis labels in the rate-distortion plots should explicitly state whether PSNR is computed on the original raster grid or on off-grid query points.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the positive assessment of the work's significance and for the detailed comment on the wavelet complexity field. We address the concern point by point below and outline the revisions we will make.

read point-by-point responses

  1. Referee: [Abstract and §3] Abstract and §3 (WCF definition): the headline 5.70 dB PSNR improvement, 3.2× parameter reduction, and derivative fidelity rest on the assumption that the wavelet complexity field correctly localizes high-frequency terrain structure. No ablation isolating the WCF contribution, no overlap or precision-recall metrics against ground-truth frequency maps, and no failure-case analysis on the 50 tiles are provided, leaving the load-bearing adaptivity mechanism unverified.

    Authors: We agree that an explicit ablation isolating the WCF would provide stronger direct evidence. The manuscript demonstrates the WCF's role through end-to-end gains over prior INRs (66.25 dB PSNR, 3.2× fewer parameters), where the WCF is defined in §3 as the normalized sum of absolute wavelet coefficients across scales, used to generate per-location frequency masks and adaptive sampling densities. To verify this mechanism, we will add an ablation in the revision that replaces the WCF-guided masks and sampling with uniform counterparts while keeping the INR architecture, derivative supervision, and compression fixed; the resulting PSNR and parameter counts will be reported on the same 50 Swiss tiles. We will also add qualitative overlays of the WCF on representative tiles to illustrate localization of high-frequency structure. Quantitative overlap or precision-recall metrics against a ground-truth frequency map are not feasible because no canonical pixel-wise frequency label exists for terrain; the wavelet coefficients themselves serve as the analytic proxy. We will instead include a short discussion of this design choice. Finally, we will add a brief failure-case analysis in the supplement covering uniformly smooth and highly oscillatory tiles. revision: yes

Circularity Check

0 steps flagged

No significant circularity detected

full rationale

The paper's core mechanism computes the wavelet complexity field directly from analytically derived wavelet coefficients on the input terrain data, which functions as an external signal rather than a self-referential definition. Performance claims (PSNR, parameter counts, training time) arise from supervised neural training and post-training compression, not from any fitted parameter being renamed as a prediction or from a derivation that reduces to its own inputs by construction. No self-citation chains, uniqueness theorems imported from prior author work, or ansatzes smuggled via citation are present in the abstract or described pipeline that would force the results. The derivation remains self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

2 free parameters · 1 axioms · 1 invented entities

Based on the abstract, the method rests on several domain assumptions and introduced constructs whose parameters are not enumerated; full text would be needed to list them exhaustively.

free parameters (2)
  • wavelet complexity thresholds
    Parameters that turn wavelet coefficients into spatially adaptive frequency masks are chosen or tuned per tile.
  • mixed-precision quantization levels
    Bit widths for post-training quantization are selected to reach the reported 1.23 bpp with 0.28 dB drop.
axioms (1)
  • domain assumption Analytically computed wavelet coefficients on terrain data produce a reliable complexity field that guides frequency allocation without artifacts.
    Invoked as the core mechanism for spatial adaptivity.
invented entities (1)
  • Wavelet Complexity Field (WCF) no independent evidence
    purpose: Derives spatially-adaptive frequency masks and guides complexity-aware sampling.
    New construct introduced to localize high-frequency capacity.

pith-pipeline@v0.9.0 · 5836 in / 1448 out tokens · 69779 ms · 2026-05-22T07:30:05.965079+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

79 extracted references · 79 canonical work pages · 2 internal anchors

  1. [1]

    Jason Ansel, Edward Yang, Horace He, Natalia Gimelshein, Animesh Jain, Michael Voznesensky, Bin Bao, Peter Bell, David Berard, Evgeni Burovski, et al . 2024. 10 ImplicitTerrainV2: Wavelet-Guided Spatially Adaptive Neural Terrain Representation Pytorch 2: Faster machine learning through dynamic python bytecode trans- formation and graph compilation. InProc...

    work page 2024

  2. [2]

    Jacobs, Yoni Kasten, and Shira Kritchman

    Ronen Basri, Meirav Galun, Amnon Geifman, David W. Jacobs, Yoni Kasten, and Shira Kritchman. 2020. Frequency Bias in Neural Networks for Input of Non-Uniform Density. InInternational Conference on Machine Learning

    work page 2020

  3. [3]

    Yizhak Ben-Shabat, Chamin Hewa Koneputugodage, Sameera Ramasinghe, and Stephen Gould. 2024. Neural experts: Mixture of experts for implicit neural representations.Advances in Neural Information Processing Systems37 (2024), 101641–101670

    work page 2024

  4. [4]

    Boucheron and Charles D

    Laura E. Boucheron and Charles D. Creusere. 2005. Lossless Wavelet-Based Compression of Digital Elevation Maps for Fast and Efficient Search and Retrieval. IEEE Trans. Geosci. Remote Sens.43, 5 (2005), 1133–1143. doi:10.1109/TGRS.2004. 841477

    work page doi:10.1109/tgrs.2004 2005

  5. [5]

    Franck Cappello, Sheng Di, Sihuan Li, Xin Liang, Ali Murat Gok, Dingwen Tao, Chun Hong Yoon, Xin-Chuan Wu, Yuri Alexeev, and Frederic T. Chong

    work page

  6. [6]

    Use Cases of Lossy Compression for Floating-Point Data in Scientific Data Sets.Int. J. High Perform. Comput. Appl.33, 6 (2019), 1201–1220. doi:10.1177/ 1094342019853336

    work page 2019

  7. [7]

    Yunbo Chen, Xiaojun Li, Fei Yang, and Peng Liu. 2023. BRIEF: Biomedical implicit neural representations for feature extraction.Medical Image Analysis90 (2023), 102960

    work page 2023

  8. [8]

    Collet \ and\ author M

    Yann Collet and Murray Kucherawy. 2018.Zstandard Compression and the ‘application/zstd’ Media Type. RFC RFC 8478. Internet Engineering Task Force (IETF). doi:10.17487/RFC8478

    work page doi:10.17487/rfc8478 2018

  9. [9]

    Wojciech M Czarnecki, Simon Osindero, Max Jaderberg, Grzegorz Swirszcz, and Razvan Pascanu. 2017. Sobolev training for neural networks.Advances in neural information processing systems30 (2017)

    work page 2017

  10. [10]

    Adam Dai, Shubh Gupta, and Grace Gao. 2024. Neural elevation models for terrain mapping and path planning.arXiv preprint arXiv:2405.15227(2024)

    work page arXiv 2024

  11. [11]

    Leila De Floriani, Ulderico Fugacci, Federico Iuricich, and Paola Magillo. 2015. Morse complexes for shape segmentation and homological analysis: discrete models and algorithms. InComputer graphics forum, Vol. 34. Wiley Online Library, 761–785

    work page 2015

  12. [12]

    1996.DEFLATE Compressed Data Format Specification version 1.3

    Peter Deutsch. 1996.DEFLATE Compressed Data Format Specification version 1.3. RFC RFC 1951. Internet Engineering Task Force (IETF). doi:10.17487/RFC1951

    work page doi:10.17487/rfc1951 1996

  13. [13]

    Emilien Dupont, Adam Goli’nski, Milad Alizadeh, Yee Whye Teh, and A. Doucet

    work page

  14. [14]

    COIN: COmpression with Implicit Neural representations.ArXiv abs/2103.03123 (2021)

    work page arXiv 2021

  15. [15]

    Emilien Dupont, Hrushikesh Loya, Milad Alizadeh, Adam Goliński, Yee Whye Teh, and Arnaud Doucet. 2022. Coin++: Neural compression across modalities. arXiv preprint arXiv:2201.12904(2022)

    work page arXiv 2022

  16. [16]

    Esri. 2024. LERC: Limited Error Raster Compression. https://github.com/Esri/lerc. Accessed: 2026-04-16

    work page 2024

  17. [17]

    Amer Essakine, Yanqi Cheng, Chun-Wun Cheng, Lipei Zhang, Zhongying Deng, Lei Zhu, Carola-Bibiane Schönlieb, and Angelica I Aviles-Rivero. 2024. Where Do We Stand with Implicit Neural Representations? A Technical and Performance Survey.arXiv preprint arXiv:2411.03688(2024)

    work page arXiv 2024

  18. [18]

    Morsali et al. 2025. STAF: Sinusoidal Trainable Activation Functions for Implicit Neural Representation.arXiv(2025)

    work page 2025

  19. [19]

    Rizal Fathony, Anit Kumar Sahu, Devin Willmott, and J Zico Kolter. 2020. Multi- plicative filter networks. InInternational Conference on Learning Representations

    work page 2020

  20. [20]

    Riccardo Fellegara, Federico Iuricich, and Leila De Floriani. 2017. Efficient representation and analysis of triangulated terrains. InProceedings of the 25th ACM SIGSPATIAL International Conference on Advances in Geographic Information Systems. 1–4

    work page 2017

  21. [21]

    Riccardo Fellegara, Federico Iuricich, Yunting Song, and Leila De Floriani. 2023. Terrain trees: a framework for representing, analyzing and visualizing triangu- lated terrains.GeoInformatica27, 3 (2023), 525–564

    work page 2023

  22. [22]

    Riccardo Fellegara, Federico Luricich, Leila De Floriani, and Kenneth Weiss. 2014. Efficient computation and simplification of discrete Morse decompositions on triangulated terrains. InProceedings of the 22nd ACM SIGSPATIAL International Conference on Advances in Geographic Information Systems. 223–232

    work page 2014

  23. [23]

    Haoan Feng, Diana Aldana, Tiago Novello, and Leila De Floriani. 2025. SAS- Net: Spatially-adaptive sinusoidal neural networks for INRs.arXiv preprint arXiv:2503.09750(2025)

    work page arXiv 2025

  24. [24]

    Haoan Feng, Yunting Song, and Leila De Floriani. 2024. Critical features tracking on triangulated irregular networks by a scale-space method. InProceedings of the 32nd ACM International Conference on Advances in Geographic Information Systems. 54–66

    work page 2024

  25. [25]

    Haoan Feng, Xin Xu, and Leila De Floriani. 2024. ImplicitTerrain: a Continu- ous Surface Model for Terrain Data Analysis. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshop. 899–909

    work page 2024

  26. [26]

    Robin Forman. 1998. Morse theory for cell complexes.Advances in mathematics 134, 1 (1998), 90–145

    work page 1998

  27. [27]

    Robin Forman. 2002. A user’s guide to discrete Morse theory.Séminaire Lotharingien de Combinatoire48 (2002), B48c

    work page 2002

  28. [28]

    Ulderico Fugacci, Sara Scaramuccia, Federico Iuricich, Leila De Floriani, et al

    work page

  29. [29]

    Persistent Homology: a Step-by-step Introduction for Newcomers.. In STAG. 1–10

    work page

  30. [30]

    Heckbert

    Michael Garland and Paul S. Heckbert. 1995. Fast Triangular Approximation of Terrains and Height Fields. InTechnical Report CMU-CS-95-181. Carnegie Mellon University

    work page 1995

  31. [31]

    Amir Gholami, Sehoon Kim, Zhen Dong, Zhewei Yao, Michael W Mahoney, and Kurt Keutzer. 2022. A survey of quantization methods for efficient neural network inference. InLow-power computer vision. Chapman and Hall/CRC, 291–326

    work page 2022

  32. [32]

    Amos Gropp, Lior Yariv, Niv Haim, Matan Atzmon, and Yaron Lipman

    work page

  33. [33]

    Implicit geometric regularization for learning shapes.arXiv preprint arXiv:2002.10099(2020)

    work page arXiv 2002

  34. [34]

    Zekun Hao, Arun Mallya, Serge Belongie, and Ming-Yu Liu. 2022. Implicit Neural Representations with Levels-of-Experts.Advances in Neural Information Processing Systems35 (2022), 2564–2576

    work page 2022

  35. [35]

    Amir Hertz, Or Perel, Raja Giryes, Olga Sorkine-Hornung, and Daniel Cohen-Or

    work page

  36. [36]

    Advances in Neural Information Processing Systems34 (2021), 8820–8832

    Sape: Spatially-adaptive progressive encoding for neural optimization. Advances in Neural Information Processing Systems34 (2021), 8820–8832

    work page 2021

  37. [37]

    1987.Introduction to numerical analysis

    Francis Begnaud Hildebrand. 1987.Introduction to numerical analysis. Courier Corporation

    work page 1987

  38. [38]

    Berthold K. P. Horn. 1981. Hill shading and the reflectance map.Proc. IEEE69, 1 (1981), 14–47

    work page 1981

  39. [39]

    Amirhossein Kazerouni, Reza Azad, Alireza Hosseini, Dorit Merhof, and Ulas Bagci. 2024. INCODE: Implicit Neural Conditioning with Prior Knowledge Embeddings. InProceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision. 1298–1307

    work page 2024

  40. [40]

    Kidner and Derek H

    David B. Kidner and Derek H. Smith. 2003. Advances in the Data Compression of Digital Elevation Models.Comput. Geosci.29, 8 (2003), 985–1002. doi:10.1016/ S0098-3004(03)00097-9

    work page 2003

  41. [41]

    Diederik P Kingma and Jimmy Ba. 2014. Adam: A method for stochastic opti- mization.arXiv preprint arXiv:1412.6980(2014)

    work page internal anchor Pith review Pith/arXiv arXiv 2014

  42. [42]

    Théo Ladune, Pierrick Philippe, Félix Henry, Erwan Le Pennec, and Clare E. Gordon. 2023. Cool-chic: Coordinate-based Low Complexity Hierarchical Image Codec. InIEEE International Conference on Computer Vision (ICCV)

    work page 2023

  43. [43]

    Shaomeng Li, Peter Lindstrom, and John Clyne. 2023. Lossy Scientific Data Compression With SPERR. InProc. IEEE Int. Parallel Distrib. Process. Symp. (IPDPS). 1007–1017. doi:10.1109/IPDPS54959.2023.00104

    work page doi:10.1109/ipdps54959.2023.00104 2023

  44. [44]

    2004.Digital Terrain Modeling: Principles and Methodology

    Zhilin Li, Qing Zhu, and Christopher Gold. 2004.Digital Terrain Modeling: Principles and Methodology. CRC Press

    work page 2004

  45. [45]

    Xin Liang, Kai Zhao, Sheng Di, Sihuan Li, Robert Underwood, Ali M Gok, Jiannan Tian, Junjing Deng, Jon C Calhoun, Dingwen Tao, et al. 2022. Sz3: A modular framework for composing prediction-based error-bounded lossy compressors. IEEE Transactions on Big Data9, 2 (2022), 485–498

    work page 2022

  46. [46]

    David B Lindell, Dave Van Veen, Jeong Joon Park, and Gordon Wetzstein. 2022. Bacon: Band-limited coordinate networks for multiscale scene representation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recogni- tion. 16252–16262

    work page 2022

  47. [47]

    Peter Lindstrom. 2014. Fixed-Rate Compressed Floating-Point Arrays.IEEE Trans. Vis. Comput. Graph.20, 12 (2014), 2674–2683. doi:10.1109/TVCG.2014.2346458

    work page doi:10.1109/tvcg.2014.2346458 2014

  48. [48]

    Peter Lindstrom and Martin Isenburg. 2006. Fast and Efficient Compression of Floating-Point Data.IEEE Trans. Vis. Comput. Graph.12, 5 (2006), 1245–1250. doi:10.1109/TVCG.2006.143

    work page doi:10.1109/tvcg.2006.143 2006

  49. [49]

    Jinyang Liu, Sheng Di, Kai Zhao, Xin Liang, Sian Jin, Zizhe Jian, Jiajun Huang, Shixun Wu, Zizhong Chen, and Franck Cappello. 2024. High-Performance Effective Scientific Error-Bounded Lossy Compression with Auto-Tuned Multi- Component Interpolation.Proc. ACM Manag. Data2, 1 (2024), 1–27. doi:10.1145/ 3639259

    work page 2024

  50. [50]

    Zhen Liu, Hao Zhu, Qi Zhang, Jingde Fu, Weibing Deng, Zhan Ma, Yanwen Guo, and Xun Cao. 2024. FINER: Flexible spectral-bias tuning in Implicit NEural Representation by Variable-periodic Activation Functions. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2713–2722

    work page 2024

  51. [51]

    Sk Sazid Mahammad and R Ramakrishnan. 2003. GeoTIFF-A standard image file format for GIS applications.Map India(2003), 28–31

    work page 2003

  52. [52]

    Maiya, Max Ehrlich, Hanlin Kim, Jiaqi Ren, Lingjie Li, and Kfir Aberman

    Shishira R. Maiya, Max Ehrlich, Hanlin Kim, Jiaqi Ren, Lingjie Li, and Kfir Aberman. 2023. NIRVANA: Neural Implicit Representations of Videos with Adaptive Networks and Autoregressive Patching.arXiv preprint arXiv:2212.14593 (2023)

    work page arXiv 2023

  53. [53]

    Stephane G. Mallat. 1989. A theory for multiresolution signal decomposition: the wavelet representation.IEEE Trans. Pattern Anal. Mach. Intell.11, 7 (1989), 674–693. doi:10.1109/34.192463

    work page doi:10.1109/34.192463 1989

  54. [54]

    Julien NP Martel, David B Lindell, Connor Z Lin, Eric R Chan, Marco Monteiro, and Gordon Wetzstein. 2021. Acorn: Adaptive coordinate networks for neural scene representation.arXiv preprint arXiv:2105.02788(2021)

    work page arXiv 2021

  55. [55]

    Ishit Mehta, Michaël Gharbi, Connelly Barnes, Eli Shechtman, Ravi Ramamoorthi, and Manmohan Chandraker. 2021. Modulated periodic activations for generaliz- able local functional representations. InProceedings of the IEEE/CVF International Conference on Computer Vision. 14214–14223. 11 Feng et al

    work page 2021

  56. [56]

    Paulius Micikevicius, Sharan Narang, Jonah Alben, Gregory Diamos, Erich Elsen, David Garcia, Boris Ginsburg, Michael Houston, Oleksii Kuchaiev, Ganesh Venkatesh, et al. 2017. Mixed precision training.arXiv preprint arXiv:1710.03740 (2017)

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  57. [57]

    1963.Morse theory

    John Willard Milnor. 1963.Morse theory. Number 51. Princeton university press

    work page 1963

  58. [58]

    Thomas Müller, Alex Evans, Christoph Schied, and Alexander Keller. 2022. In- stant neural graphics primitives with a multiresolution hash encoding.ACM Transactions on Graphics41, 4 (2022), 1–15

    work page 2022

  59. [59]

    Nason and Bernard W

    Guy P. Nason and Bernard W. Silverman. 1995. The Stationary Wavelet Transform and Some Statistical Applications. InWavelets and Statistics. Springer New York, 281–299. doi:10.1007/978-1-4612-2544-7_17

    work page doi:10.1007/978-1-4612-2544-7_17 1995

  60. [60]

    Tiago Novello, Diana Aldana, Andre Araujo, and Luiz Velho. 2025. Tun- ing the Frequencies: Robust Training for Sinusoidal Neural Networks. arXiv:2407.21121 [cs.LG] https://arxiv.org/abs/2407.21121

    work page arXiv 2025

  61. [61]

    Tiago Novello, Guilherme Schardong, Luiz Schirmer, Vinícius da Silva, Hélio Lopes, and Luiz Velho. 2022. Exploring differential geometry in neural implicits. Computers & Graphics108 (2022), 49–60. doi:10.1016/j.cag.2022.09.003

    work page doi:10.1016/j.cag.2022.09.003 2022

  62. [62]

    1999.Discrete-time signal processing

    Alan V Oppenheim. 1999.Discrete-time signal processing. Pearson Education India

    work page 1999

  63. [63]

    Hamprecht, Yoshua Bengio, and Aaron C

    Nasim Rahaman, Aristide Baratin, Devansh Arpit, Felix Dräxler, Min Lin, Fred A. Hamprecht, Yoshua Bengio, and Aaron C. Courville. 2018. On the Spectral Bias of Neural Networks. InInternational Conference on Machine Learning

    work page 2018

  64. [64]

    Sameera Ramasinghe, Lachlan E MacDonald, and Simon Lucey. 2022. On the frequency-bias of coordinate-mlps.Advances in Neural Information Processing Systems35 (2022), 796–809

    work page 2022

  65. [65]

    Vishwanath Saragadam, Daniel LeJeune, Jasper Tan, Guha Balakrishnan, Ashok Veeraraghavan, and Richard G Baraniuk. 2023. Wire: Wavelet implicit neural representations. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 18507–18516

    work page 2023

  66. [66]

    Vishwanath Saragadam, Jasper Tan, Guha Balakrishnan, Richard G Baraniuk, and Ashok Veeraraghavan. 2022. MINER: Multiscale Implicit Neural Representations. InEuropean Conference on Computer Vision

    work page 2022

  67. [67]

    Sara Scaramuccia, Federico Iuricich, Leila De Floriani, and Claudia Landi. 2020. Computing multiparameter persistent homology through a discrete Morse-based approach.Computational Geometry89 (2020), 101623

    work page 2020

  68. [68]

    Kexuan Shi, Xingyu Zhou, and Shuhang Gu. 2024. Improved Implicit Neural Representation with Fourier Reparameterized Training. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 25985–25994

    work page 2024

  69. [69]

    Vincent Sitzmann, Julien Martel, Alexander Bergman, David Lindell, and Gordon Wetzstein. 2020. Implicit neural representations with periodic activation func- tions.Advances in Neural Information Processing Systems33 (2020), 7462–7473

    work page 2020

  70. [70]

    Swisstopo. 2023. SwissALTI3D datasets. https://www.swisstopo.admin.ch/en/ height-model-swissalti3d

    work page 2023

  71. [71]

    Matthew Tancik, Pratul Srinivasan, Ben Mildenhall, Sara Fridovich-Keil, Nithin Raghavan, Utkarsh Singhal, Ravi Ramamoorthi, Jonathan Barron, and Ren Ng

    work page

  72. [72]

    Fourier features let networks learn high frequency functions in low dimen- sional domains.Advances in Neural Information Processing Systems33 (2020), 7537–7547

    work page 2020

  73. [73]

    Dingwen Tao, Sheng Di, Xin Liang, Zizhong Chen, and Franck Cappello. 2019. Optimizing Lossy Compression Rate-Distortion from Automatic Online Selection between SZ and ZFP.IEEE Trans. Parallel Distrib. Syst.30, 8 (2019), 1763–1777. doi:10.1109/TPDS.2019.2894404

    work page doi:10.1109/tpds.2019.2894404 2019

  74. [74]

    Thomas Walker, Octave Mariotti, Amir Vaxman, and Hakan Bilen. 2025. Spatially- Adaptive Hash Encodings for Neural Surface Reconstruction. In2025 IEEE/CVF Winter Conference on Applications of Computer Vision (W ACV). IEEE, 2963–2972

    work page 2025

  75. [75]

    Ian H Witten, Radford M Neal, and John G Cleary. 1987. Arithmetic coding for data compression.Commun. ACM30, 6 (1987), 520–540

    work page 1987

  76. [76]

    Randolph Franklin, and Daniel M

    Zhongyi Xie, W. Randolph Franklin, and Daniel M. Tracy. 2010. Slope Preserving Lossy Terrain Compression.ACM SIGSPATIAL Special2, 1 (2010), 19–24. doi:10. 1145/1953102.1953106

    work page arXiv 2010

  77. [77]

    Wentao Yuan, Qingtian Zhu, Xiangyue Liu, Yikang Ding, Haotian Zhang, and Chi Zhang. 2022. Sobolev training for implicit neural representations with approximated image derivatives. InEuropean Conference on Computer Vision. Springer, 72–88

    work page 2022

  78. [78]

    Gizem Yüce, Guillermo Ortiz-Jiménez, Beril Besbinar, and Pascal Frossard. 2022. A structured dictionary perspective on implicit neural representations. InPro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 19228–19238

    work page 2022

  79. [79]

    Zevenbergen and Colin R

    Lyle W. Zevenbergen and Colin R. Thorne. 1987. Quantitative analysis of land surface topography.Earth Surface Processes and Landforms12, 1 (1987), 47–56. 12 ImplicitTerrainV2: Wavelet-Guided Spatially Adaptive Neural Terrain Representation A Dataset Curation We evaluate on 50 terrain tiles selected from 33,399 candidates from the swissALTI3D DEM [65] via ...

    work page 1987