Pith. sign in

REVIEW 4 major objections 5 minor 29 references

How Usable Are Geospatial Foundation Models? A Systematic Evaluation of 89 Models

T0 review · 4 major / 5 minor · reviewed 2026-08-05 · deepseek-v4-flash

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

Pith's one-line read A seven-dimension usability review of 89 geospatial foundation models finds that no release documents uncertainty quantification—the trust feature domain experts rank first—and that over half the models demand machine-learning expertise jus

desk verdict A genuinely useful first systematic usability rubric for GeoFMs, but the reliability evidence for its headline numbers is shakier than the abstract admits, and the authors' own Tessera is scored top without disclosure: worth refereeing after careful revision. read the letter →

arxiv 2608.03804 v1 pith:NQ6HR4WO submitted 2026-08-04 cs.HC

Robin Young , Artyom Gabtraupov , Kenzy Soror , Srinivasan Keshav This is my paper
classification cs.HC
keywords geospatialfoundationmodelsusabilityevaluationhuman-computerinteractionEarthobservationuncertaintyquantificationexpertelicitationaccessibilitysystematicreview
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 claims that geospatial foundation models—the large AI systems trained on satellite imagery to power environmental monitoring—have been built for machine-learning specialists, not for the ecologists, conservation scientists, and planners who need them. The authors surveyed eleven domain experts, turned their stated priorities into a seven-dimension usability rubric, and applied it to 89 published models. They report that 50.6% of models require users to re-train from scratch or assemble a complex software environment, and that no model provides documented uncertainty quantification—the single feature the experts most wanted for trusting outputs. The paper reads this as a field-level misalignment: development has optimized benchmark accuracy while leaving interfaces, documentation, trust features, and archival permanence underinvested. If correct, the finding reframes geo-AI progress from 'how accurate is the model?' to 'can the intended scientist actually use it?', a prerequisite for these tools to affect climate and biodiversity decisions.

What carries the argument

The carrying mechanism is the seven-dimension framework itself, a rubric turning 'can the intended user actually run this model?' into measurable levels and checklists. Access & Deployment is a 0–5 ladder from 'source code only' to 'graphical interface'; Interaction & Customization is a base level plus a checklist of documented interaction paradigms; Trust & Transparency is a four-attribute checklist including replicable benchmarks and uncertainty quantification; community support, scientific permanence, multilingual support, and offline usability complete the instrument. The framework does two jobs: wide-variance dimensions (access, interaction, community) discriminate between models today,

What would settle it

Two checks would settle it. (1) Independently audit the 63 accessible code repositories for runnable, documented uncertainty-estimation code: finding even one model with per-pixel or per-prediction confidence output would overturn the 0% (T3) headline, converting it from a capability gap to a documentation gap. (2) Rebuild the corpus from a different set of primary sources—industry releases, Chinese-language publications, agency products—and re-run the rubric: if the aggregate percentages (50.6% at Level 0–1, ~95% monolingual) do not survive, the field-level conclusions are an artifact of the

Watch

Extended reading notes

Core claim

The paper claims the geospatial foundation model field has optimized benchmark accuracy while underinvesting in user-centered infrastructure. Grounded in an eleven-expert survey, the authors built a seven-dimension usability rubric and had two raters score 89 models (93% observed agreement). Findings: 50.6% of models require pre-training or a poorly supported environment; only 6.7% reach a hosted API or GUI; zero of 63 accessible models document uncertainty quantification—the trust feature experts ranked first; 96.8% are open-core but only 1.6% archived with a DOI; 95.2% are monolingual. Wide-variance dimensions discriminate models today; near-zero-variance dimensions are field-level diagnos

Load-bearing premise

The 89-model corpus, assembled from four published model surveys, stands in for the entire geospatial foundation model landscape; if large segments (industrial models, non-English venues, or models omitted from those four surveys) differ systematically, the field-level percentages—50.6% inaccessible, 0% with documented uncertainty—would be wrong, and the paper does not test how its numbers respond to corpus composition.

Editorial extensions

If this is right

  • Practitioners get a working shortlist: only 6 of 89 models reach the top access tiers (hosted API or GUI), so any project expecting a field scientist to run a GeoFM today must plan around that short list or budget for ML-engineering support.
  • No ecological conclusion drawn from a GeoFM currently rests on published confidence estimates; the paper's strict threshold suggests the fix is architectural—downstream heads should report calibrated intervals and embedding-level out-of-distribution flags by default.
  • Because the rubric scores documented capability, the result doubles as an incentive: stating a usability profile in a model card would turn invisible gaps into selection pressure and give developers credit for work that already exists but goes unreported.
  • The embedding-as-data-product pattern—precomputing embeddings once and serving them as versioned, downloadable data—is the demonstrated route to closing the access gap, and the paper argues a community norm of wrapping inference code in lightweight web demos could lift most releases several levels at negligible cost.
  • Temporal trends move in one direction only: 2025 releases were more accessible (54% at Level 3+, up from 27% in 2024), but trust, permanence, and multilingual dimensions showed no improvement at all, so recent progress is concentrated in access alone.

Reading between the lines

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

  • The corpus is bounded by four source surveys, so the field-level percentages are only as good as that coverage; an independent corpus built from industry releases, agency products, and non-English venues is the untested check, and until it runs, the 50.6% and 0% figures should be read as claims about the surveyed landscape, not a census.
  • The paper's 'communicated not latent capability' stance yields a sharp test: audit the 63 accessible codebases for runnable, documented uncertainty code. Finding any would convert the headline from 'no model provides UQ' to 'no model documents UQ'—a documentation failure, arguably still a usability failure, but a materially different claim.
  • The expert survey's top-rated AI capability, anomaly detection (6.4/7), points to a product niche the paper does not pursue: embedding-based change-flagging tools aimed at field ecologists rather than ML engineers, which would directly exercise the offline pre-cache-and-analyze workflow the framework defines.
  • The paper's access trend rests on small annual samples (n=5 to n=30) and its own reading is cautious; a sharp prediction follows from its framework—if the 2025 jump to mean level 2.2 is a genuine norm shift, 2026 releases should stay at or above it and the diagnostic dimensions should finally show variance, while a fallback would indicate a selection artifact.
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

4 major / 5 minor

Summary. This paper develops a seven-dimension usability evaluation framework for geospatial foundation models (GeoFMs), motivated by a pilot expert survey (N=11) with ecology and conservation scientists. The framework covers Access & Deployment, Interaction & Customization, Trust & Transparency, Community & Support, Scientific Permanence, Multilingual Support, and Offline Usability. Two raters applied the rubric to 89 GeoFMs drawn from four existing surveys. The main empirical claims are that 50.6% of the 89 models sit at access levels 0–1 (no usable pretrained weights or weights with minimal environment guidance), no accessible model provides documented uncertainty quantification, and documentation/community support is generally thin. The paper distinguishes discriminating dimensions from field-level diagnostics and closes with recommendations for embedding-as-data, usability metadata, and uncertainty-aware design.

Significance. The contribution is potentially valuable: it is one of the first usability-centric, systematic evaluations of GeoFMs, with a transparent rubric and a full per-model rating table (Appendix D). The finding that uncertainty quantification is entirely absent from a large sample of models, and that most models require substantial ML expertise, is a useful challenge to the model-centric evaluation culture in this field. The authors also deserve credit for explicitly bounding their corpus, reporting detailed limitations, and making their rubric and survey instrument available in appendices. However, the central quantitative claims depend on (a) the reliability of the D1 ratings and (b) the representativeness of the corpus, and both need reinforcement before the field-level conclusions can be taken as established.

major comments (4)
  1. [§4.3, §6.6, Appendix C Table C.6] The reliability evidence for D1 is internally inconsistent. Section 4.3 reports overall nominal κ=0.92 and ordinal weighted κ=0.95; §6.6 states D1 nominal κ=0.56 (quadratic-weighted κ=0.91); Appendix C Table C.6 reports D1 AC1=κ=0.912 and labels the row 'Ordinal.' These three statements cannot all describe the same calibration sample. Since D1 is the primary outcome, the abstract's 'highly consistent' claim and the headline 50.6% figure rest on unresolved reporting. Moreover, only 20 models were double-rated; the remaining 69 were coded by one rater after a discussion that resolved the Level 2/3 boundary, but no per-rater distributions or post-discussion reliability are reported. A systematic leniency difference between raters at exactly the Level 0/1 versus 2/3 boundary would directly shift the 50.6% figure. Please reconcile the statistics, report nominal κ for D1 explicitly, and provid
  2. [§4.1, §5.8, §6.5] The corpus is assembled from four secondary surveys, and the paper explicitly notes it is bounded by them, yet the Discussion repeatedly upgrades aggregate percentages to 'field-level diagnostics' and 'field-level norms.' No sensitivity analysis is provided for corpus composition; if the four surveys under-represent industry models, non-English venues, or models released outside those surveys, the estimates (e.g., 0% with UQ, 96.8% open-core-only) could be sample-specific. At minimum, the paper should either restrict the claims to 'the 89 models in these four surveys' or provide a robustness check (e.g., leave-one-survey-out, comparison with a hand-curated list, or a characterization of the excluded margin).
  3. [§2.2.3, §5.3, §6.2] The pilot survey (N=11, single institution) is used to support the central interpretive claim that uncertainty quantification is the feature domain experts prioritize most for trust, and this drives the 'misalignment' conclusion. The survey section itself is carefully hedged as exploratory, but the Results and Discussion treat it as strong evidence ('the feature that our survey found domain experts prioritize most highly,' 'the most direct evidence of misalignment'). With 10 respondents answering the trust question and 5 selecting uncertainty visualization, this is a 50% preference in a very small, homogeneous sample. Please temper the field-level language or supplement with additional evidence.
  4. [§4.2, §5.9.5, §6.1] Two of the four authors are co-authors of Tessera, which is rated D1=5, receives the only full set of interaction paradigms (I1–I5) and one of the two Level-5 community scores, and is used in §6.1 as the 'existence demonstration' for the embedding-as-data-product model. The manuscript does not disclose this relationship in the rating methodology or in a competing-interests statement. For an evaluation study, the absence of disclosure makes it impossible for readers to assess potential bias in the ratings of Tessera and in the archetype comparison of Table 5. Please add a COI statement and, ideally, have an independent rater score Tessera or justify why the co-authorship does not affect the rating.
minor comments (5)
  1. [§4.2] The text says 'two raters independently evaluated each model in the corpus,' but the described procedure has only a 20-model overlap double-rated; the remaining models are single-rated. Please rephrase to avoid implying full double-coding.
  2. [Appendix C Table C.6] The D1 row is typed 'Ordinal' but reports AC1 and κ; Section 4.3 says AC2 is used for ordinal dimensions. Clarify which statistics are computed for ordinal versus nominal dimensions.
  3. [§5.5] Typo: 'reproducibility can remains fragile' should be 'reproducibility can remain fragile.'
  4. [Table D.7] The row 'DOF A (Xiong et al., 2024)' appears to be a typo for 'DOFA' (or possibly duplicates 'OFA-Net'). Please correct the entry and check the model list for duplicates.
  5. [§5.6] The denominator for D6 is '61 models with available data,' whereas §5.2 reports 63 accessible models. Explain why two models lack D6 data (e.g., no documentation located) so the percentages are interpretable.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the evaluation is an empirical measurement with an externally verifiable rubric, not a derivation from its own inputs.

full rationale

The paper makes no mathematical derivation and fits no parameters; its core claims are direct empirical measurements of 89 models against a seven-dimension rubric. The rubric was informed by a pilot survey and HCI theory, but the D1 distribution (50.6% at Levels 0–1), the 0% documented uncertainty-quantification figure, and the other aggregate percentages are observations of public artifacts, not outputs forced by the framework. The pilot survey highlights uncertainty as a priority and the framework includes T3 as a result, but the absence of T3 across the corpus is an independent fact about model documentation. Tessera is rated as the most feature-complete model and used as an existence proof, and two current authors co-authored Tessera; however, the cited properties of Tessera (public code, embeddings, Zulip channel, etc.) are externally verifiable, so this is minor self-reference rather than load-bearing circularity. The paper itself discloses limitations, including the D1 inter-rater disagreement and the snapshot nature of the corpus; these are correctness/reliability concerns, not circularity. The conflicting D1 κ values (nominal 0.56 in Section 6.6 versus AC1/κ 0.912 in Appendix C) undermine confidence in the reliability evidence but do not make any stated result equivalent to its input by definition. The central claims stand as empirical findings rather than as renamed inputs.

Assumptions & free parameters 0 free parameters · 4 assumptions · 0 invented entities

This is an evaluation paper, so there are no fitted constants or invented physical entities. The load-bearing premises are: representativeness of the 89-model corpus, the sufficiency of documented capability as a usability proxy, the construct validity of the unvalidated seven-dimension framework, and the reliability of the author-performed ratings.

assumptions (4)
  • domain assumption The 89-model corpus, drawn from four source surveys (Feng et al., Zhu et al., Lu et al., Xiao et al.), is representative of the GeoFM ecosystem during the evaluation window.
    All field-level percentages (e.g., D3: 0% uncertainty, D5: 97% open-core, D6: 95% monolingual) are computed over this corpus; if it is not representative, the 'field-level diagnostic' conclusions do not generalize (Section 4.1).
  • domain assumption Documented public capability is the correct operationalization of usability; undocumented capabilities are treated as nonexistent.
    The authors state in Section 6.6 that 'if a capability is not documented, it does not exist from a usability perspective'; this modeling choice could undercount the practical usability of models with latent capabilities.
  • domain assumption The seven evaluation dimensions, derived from HCI theory and the N=11 pilot survey, capture the usability attributes that matter to intended GeoFM users.
    The framework's construct validity is assumed; Section 6.7 concedes it has not yet been validated with task-based user studies.
  • domain assumption Inter-rater agreement after calibration is sufficient for the reported ratings to be treated as reliable.
    The paper reports high agreement but also contains internal inconsistencies in the reliability statistics (Appendix C vs Sections 4.3 and 6.6); the assumption that the ratings are consistent underlies every quantitative result.

how reviews work

0 comments
Cite this review

Pith. "Pith review of How Usable Are Geospatial Foundation Models? A Systematic Evaluation of 89 Models." pith.science (2026). https://pith.science/paper/NQ6HR4WO

@misc{pith2026260803804,
  author       = {Pith},
  title        = {Pith review of: How Usable Are Geospatial Foundation Models? A Systematic Evaluation of 89 Models},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/NQ6HR4WO}},
  note         = {Machine review of arXiv:2608.03804}
}
read the original abstract

Geospatial foundation models (GeoFMs) offer transformative potential for environmental monitoring, yet adoption among ecologists is uneven. Most evaluations are model-centric, focusing on architecture and benchmark accuracy, which overlooks whether the systems are usable by their intended audiences. To address this gap, we first conducted a pilot expert elicitation survey with ecology and conservation scientists that helped us identify misalignments between current GeoFM development priorities and their needs. Informed by these findings and based on HCI theory, we created a seven-dimension evaluation covering Access & Deployment, Interaction & Customization, Trust & Transparency, Community & Support, Scientific Permanence, Multilingual Support, and Offline Usability. Then, two raters applied this rubric to 89 GeoFMs. We found distinct accessibility gaps where nearly a third provide no support to practitioners beyond their source code. Dimensions along which ratings were highly consistent function as field-level diagnostics, revealing where there is room for improvement for current GeoFMs.

Figures

Figures reproduced from arXiv: 2608.03804 by the authors.

Figure 1
Figure 1. Distribution of Access & Deployment levels (Dimension 1) across 89 geospatial foundation models. 23 [PITH_FULL_IMAGE:figures/full_fig_p023_1.png] view at source ↗
Figure 2
Figure 2. Distribution of (a) supported interaction paradigms (Dimension 2), (b) trust and transparency attributes (Dimension 3), and (c) community and support levels (Di￾mension 4) across 63 accessible GeoFMs. 24 [PITH_FULL_IMAGE:figures/full_fig_p024_2.png] view at source ↗
Figure 3
Figure 3. Temporal trends in GeoFM accessibility. (a) Raw count of models by access level and release year. (b) Proportion of models at each access level, normalized by year, controlling for the uneven growth in model releases. (c) Mean access level (green) and percentage of models at Level 0 (red) over time. Sample sizes vary across years (shown below axis). 35 [PITH_FULL_IMAGE:figures/full_fig_p035_3.png] view at source ↗

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

29 extracted references · 21 canonical work pages

  1. [1]

    What is your primary field(s) of expertise? (Select all that ap- ply.) Ecology / Conservation / Earth Science / Agriculture / Geogra- phy / Architecture / Computer Science / Other. 57

    work page

  2. [2]

    satel- lite imagery, GIS data, aerial photos) in your research? (Select one.) Daily / Weekly / Monthly / A few times a year / Rarely or never

    How frequently do you work with geospatial data (e.g. satel- lite imagery, GIS data, aerial photos) in your research? (Select one.) Daily / Weekly / Monthly / A few times a year / Rarely or never

    work page

  3. [3]

    How many years of experience do you have working with geospatial data? (Select one.) < 1 / 1–3 / 3–5 / 5–10 / > 10

    work page

  4. [4]

    (Scale: 1-No Experience to 5-Expert, one rating per row.) Desktop GIS Software (e.g

    Please rate your comfort level with the following tools and methods. (Scale: 1-No Experience to 5-Expert, one rating per row.) Desktop GIS Software (e.g. QGIS, ArcGIS) / Cloud Platforms (e.g. Google Earth Engine) / Programming (e.g. Python, R) for data anal- ysis / Command line interfaces and shell scripting. Section 2: Current Workflows & Pain Points

    work page

  5. [5]

    Thinking of a recent geospatial project, how difficult were each of the following steps? (Scale: 1-Trivial to 7-A Major Bar- rier, one rating per row.) Finding the right dataset for your question / Navigating data portals and their access control systems / Down- loading large datasets ( >100 GB) / Converting data between different formats / Integrating datas...

    work page

  6. [6]

    What is the single biggest bottleneck that slows down your research or causes the most frustration? (Open text.)

    work page

  7. [7]

    What tools in your current research workflow work very well? What makes your research easier? (Open text.) Section 3: Perspectives on AI Foundation Models

    work page

  8. [8]

    geospatial founda- tion models

    How familiar are you with the concept of “geospatial founda- tion models” (e.g. Tessera, Prithvi, Clay, Presto, Aurora, or Google Embedding Fields)? (Select one.) I could train or imple- ment one from scratch / I could use a pretrained model for a new task / I could explain the core concepts but not implement it / I have seen the term and have a general i...

    work page

Show all 29 references
  1. [9]

    Imagine you could customize a general AI model for a specific task. What would be the biggest barrier? (Select one.) The technical skill required / The effort of preparing my own data / The time to train or fine-tune / The computational cost / I’m not sure. 58

  2. [10]

    To trust the output of an AI model for a scientific publication or policy recommendation, which is most important to you? (Select one.) High accuracy scores on a standard benchmark / The model’s reasoning aligns with my domain knowledge / A clear visual- ization of where the mo...

  3. [11]

    If you wanted to train an AI model to recognize a specific phenomenon, what would be your biggest concern? (Select one.) Not knowing what data to include / Not knowing if I have enough training data / Not understanding what architecture to choose / Not knowing what hardware I n...

  4. [12]

    What foundation models, if any, have you tried? What hap- pened? (Open text.) Section 4: Your Ideal Tools & Final Thoughts

  5. [13]

    In an ideal world, what kind of tool or AI assistant would make the biggest positive impact on your geospatial research? (Open text.)

  6. [14]

    Please rate the potential usefulness of the following AI ca- pabilities for your research. (Scale: 1-Not Useful to 7-Extremely Useful, one rating per row.) Finding similar regions across the globe / Generatively filling in gaps in data / Generating realistic what-if scenarios /...

  7. [15]

    doi: 10.1007/978-3-031-73242-3_15 . P. Akiva, M. Purri, M. Leotta, Self-supervised material and texture represen- tation learning for remote sensing tasks, in: 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2022, pp. 8193–

    2022 doi

  8. [16]

    GEE, ArcGIS, QGIS)

    If a new, powerful AI model for geospatial analysis was avail- able, in which format would you prefer to access it? (Select one.) Source code on GitHub to adapt and train myself / A pretrained model to deploy and fine-tune / Pre-computed data files (represen- tation vectors) to ...

  9. [17]

    After an AI model provides a result, what methods would you use to validate it? (Select all that apply.) Compare overall ac- curacy with published benchmarks / Visually spot-check against high- resolution imagery / Compare with my own field data or local knowl- edge / Check if ...

  10. [18]

    black box

    What is your biggest concern about the increasing use of com- plex, “black box” AI models in your field? (Open text.) Appendix B. Model Evaluation Rubric The following rubric was used by both raters to evaluate each model. For each model, raters recorded the model name, version...

    2025

  11. [20]

    doi: 10.1007/978-3-031-85187-2_20 . K. Wu, Y. Zhang, L. Ru, B. Dang, J. Lao, L. Yu, J. Luo, Z. Zhu, Y. Sun, J. Zhang, Q. Zhu, J. Wang, M. Yang, J. Chen, Y. Zhang, Y. Li, A semantic-enhanced multi-modal remote sensing foundation model for earth observation, Nature Machine Intel...

    2025

  12. [24]

    doi: 10.1007/978-3-031-73390-1_24 . P. Dias, A. Tsaris, J. Bowman, A. Potnis, J. Arndt, H. L. Yang, D. Lunga, Oreole-fm: successes and challenges toward billion-parameter foundation models for high-resolution satellite imagery, in: Proceedings of the 32nd ACM International Con...

    2024

  13. [26]

    When evaluating an AI model, how important are the fol- lowing properties to you? (Scale: 1-Not Important to 7-Critically Important, one rating per row.) Predictions respect known physical laws / The model understands temporal patterns / Analysis moves flu- idly between scales ...

  14. [142]

    Clay, Clay foundation model, 2024

    doi: 10.1109/IGARSS46834.2022.9883983. Clay, Clay foundation model, 2024. URL: https://huggingface.co/ made-with-clay/Clay. doi: 10.57967/hf/1537. 45 Google Earth Engine, Google DeepMind, Google satellite embedding v1 annual, Earth Engine Data Catalog, 2025. Retrieved July 1, ...

    2022

  15. [331]

    URL: https://link.springer.com/10.1007/978-3-031-85187-2_

    doi

  16. [2023]

    arXiv:2312.16385

    URL: https://arxiv.org/abs/2312.16385. arXiv:2312.16385. G. Astruc, N. Gonthier, C. Mallet, L. Landrieu, Anysat: One earth ob- servation model for many resolutions, scales, and modalities, in: 2025 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2025, pp...

    2025 arXiv

  17. [2024]

    doi: 10.48550/arXiv

    URL: http://arxiv.org/abs/2304.14065. doi: 10.48550/arXiv. 2304.14065, arXiv:2304.14065 [cs]. D. Szwarcman, S. Roy, P. Fraccaro, . E. Gíslason, B. Blumenstiel, R. Ghosal, P. H. de Oliveira, J. L. de Sousa Almeida, R. Sedona, Y. Kang, S. Chakraborty, S. Wang, C. Gomes, A. Kumar...

    2026 doi arXiv

  18. [2025]

    arXiv:2503.16426

    URL: https://arxiv.org/abs/2503.16426. arXiv:2503.16426. M. J. Smith, L. Fleming, J. E. Geach, Earthpt: a time series foundation model for earth observation, 2024. URL: https://arxiv.org/abs/2309. 07207. arXiv:2309.07207. J. Tan, C. Zhang, B. Dang, Y. Li, Towards privacy-prese...

    2024

  19. [4392]

    doi: 10.1038/s41467-021-24638-z . B. Han, S. Zhang, X. Shi, M. Reichstein, Bridging remote sensors with multisensor geospatial foundation models, in: 2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2024, pp. 27852– 27862. doi: 10.1109/CVPR52733.2024...

    2024

  20. [8205]

    doi: 10.1109/CVPR52688.2022.00803. P. Ravirathinam, A. Parthasarathy, A. Khandelwal, R. Ghosh, V. Ku- mar, Towards knowledge guided pretraining approaches for multimodal 50 foundation models: Applications in remote sensing, ACM Trans. Spa- tial Algorithms Syst. (2026). URL: ht...

    2022

  21. [9887]

    doi: 10.1109/ICCV51701.2025.00921. D. Wang, J. Zhang, B. Du, G.-S. Xia, D. Tao, An empirical study of remote sensing pretraining, IEEE Transactions on Geoscience and Remote Sensing 61 (2023a) 1–20. doi: 10.1109/TGRS.2022.3176603. D. Wang, Q. Zhang, Y. Xu, J. Zhang, B. Du, D. T...

    2025

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.