Pith. sign in

REVIEW 2 major objections 5 minor 80 references

Evaluating Semantic and Spatial Guidance for Foundation Model Segmentation of Small-Scale PV in Remote Sensing Imagery

T0 review · 2 major / 5 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read Prompting strategy, not training data volume, is what governs how SAM3 adapts to small solar panel segmentation.

desk verdict Careful empirical study of SAM3 prompting for small PV segmentation, but the headline hierarchy is confounded with which encoders get fine-tuned. read the letter →

arxiv 2608.10801 v1 pith:GQF65IJN submitted 2026-08-11 cs.CV

Roni Blushtein-Livnon , Tal Svoray , Osher Rafaeli , Michael Dorman , Itay Fischhendler , Havazelet Yahel , Emir Galilee This is my paper · ORCID
classification cs.CV
keywords foundationmodelsSAM3promptingstrategiesPVsegmentationremotesensingsmallobjectsemanticguidancespatial
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 asks which way of telling a segmentation foundation model what to look for—a text description, a bounding box, or both—matters most when the target is small, sparse photovoltaic panels in aerial imagery. The authors fine-tune SAM3 under all three prompting strategies while varying supervision scale, transfer learning, spatial resolution, and imaging conditions, and they measure the resulting hierarchy in accuracy and stability. Their central claim is that prompting strategy is the primary factor governing model behavior: hybrid prompts perform best, geometric prompts are intermediate, and text-only prompts are worst and least stable. The finding matters because it shows where adaptation effort should go when building PV maps for off-grid regions, and it indicates that most accuracy gains arrive with only a few hundred annotated samples.

What carries the argument

The load-bearing object is SAM3's modular promptable architecture, which encodes text and geometric prompts into a shared embedding space through a unified perception encoder and fuses them with image features via cross-attention; a presence head estimates whether the concept appears, a query-based transformer decoder localizes instances, and a mask decoder produces pixel masks. The paper exploits this modularity to build three fine-tuning pipelines—text-only, with the text encoder trainable; geometric-only, with the geometry encoder trainable; and hybrid, with both trainable, while the mask decoder is always trainable. This setup, together with ground-truth-derived bounding boxes as a controlled spatial prompt, is what lets the authors attribute performance differences to prompting strategy rather than to detector error.

What would settle it

Run a crossover experiment in which every prompting condition is fine-tuned with the same set of trainable modules—for example, unfreeze both encoders under text-only and geometric-only prompts, and unfreeze only the relevant encoder under hybrid—and check whether the hybrid-over-geometric-over-text ranking and the complementarity effect survive; if they do not, the claim that prompting strategy is the dominant factor would be refuted.

Watch

Extended reading notes

Core claim

Working with SAM3, a promptable vision-language foundation model, the paper reports a consistent and statistically significant performance hierarchy across all imaging conditions and datasets: hybrid prompting (text plus bounding box) achieves the highest F1, IoU, and Recall; geometric prompting (bounding box only) is intermediate; textual prompting (noun phrase only) is lowest and most variable. The paper argues that semantic and spatial guidance are complementary rather than interchangeable: text conveys target identity, boxes supply localization, spatial guidance mainly reduces localization uncertainty, and semantic guidance helps recover pixels that would otherwise be missed. It also reports that spatial guidance substantially reduces sensitivity to spatial resolution and cross-image variability, that performance plateaus after roughly 200–300 training samples for spatially guided strategies, and that transfer learning matters mainly for text-only prompting at low supervision. The same hierarchy reproduces on three external aerial datasets, supporting the generality of the findings.

Load-bearing premise

The comparison treats prompting strategy as if it were separable from which parts of the model get fine-tuned, but text-only, geometric-only, and hybrid each come with a different fixed set of trainable encoders, so the observed ranking could partly reflect adaptation capacity or optimization dynamics rather than the guidance modality itself.

Editorial extensions

If this is right

  • For SAM3-based PV mapping, the prompting pipeline should be treated as a design decision on par with model architecture, because it changes accuracy, robustness, and data requirements.
  • Text-only prompting is the least reliable option for small, sparse targets and needs the most supervision; applications that cannot provide spatial cues should expect lower recall and higher sensitivity to image quality.
  • Hybrid prompting delivers the highest accuracy and the lowest sensitivity to resolution, image variability, and supervision scale, making it the preferred configuration for heterogeneous multi-campaign imagery.
  • A few hundred labeled samples are enough to reach most of the achievable performance for spatially guided prompting, so annotation budgets can be kept small.
  • Transfer learning adds little once explicit spatial guidance is available, so its value is mostly confined to text-driven adaptation under limited supervision.

Reading between the lines

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

  • If prompting strategy is genuinely the dominant control knob, then prompt-level design—such as learning or optimizing text descriptors and box proposals jointly—could be a more cost-effective investment than architectural modification for this class of small-object remote sensing tasks.
  • The paper's use of ground-truth boxes leaves open whether a fully automated pipeline with detector-generated boxes preserves the hybrid advantage; testing that chain would determine operational feasibility.
  • The authors' explanation for the text-only decline at higher supervision (over-specialization to dominant appearances) is plausible but untested; a direct test would compare worst-case and minority-appearance recall as training grows.
  • The same controlled comparison could be applied to other small and imbalanced targets, and to other promptable foundation models, to see whether the hybrid-over-geometric-over-text hierarchy is a SAM3 property or a general one.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

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 systematically evaluates SAM3 (a promptable vision-language foundation model) for pixel-level segmentation of small-scale residential photovoltaic (PV) installations in remote sensing imagery, comparing three prompting strategies: textual only, geometric only (ground-truth bounding boxes), and hybrid (text + boxes). Experiments vary supervision scale (200-700 samples plus zero-shot), training strategy (independent per-image fine-tuning versus sequential transfer learning), spatial resolution (0.25, 0.145, 0.132 m/pixel), and cross-image variability across seven acquisition years in the Northern Negev, with additional validation on three public datasets (France Google, France IGN, Queens NY). The paper reports that prompting strategy is the dominant factor governing model behavior: textual prompting performs worst and is most sensitive to supervision, resolution, and imaging conditions; geometric prompting is substantially better; hybrid prompting achieves the highest and most stable performance, indicating complementarity between semantic and spatial guidance. The authors also report strong data efficiency, with most gains from a few hundred samples, and limited transfer-learning benefit except for textual prompting. The central claim is that the choice of prompt modality is a primary control knob when adapting SAM3 to small-object PV segmentation, and that semantic and spatial guidance are complementary.

Significance. If the central claims hold, this would be a valuable and much-needed systematic benchmark of prompting modalities for a state-of-the-art promptable segmentation foundation model in a challenging, imbalanced remote sensing setting. The study is thoughtfully designed in several respects: held-out validation and test partitions that are fixed across experiments, geographically non-overlapping splits, prompt selection performed on a spatially disjoint subset before the main experiments, external datasets spanning three resolution levels, and appropriate statistical models (fractional logit with HC3 standard errors, Type-II ANOVA with partial eta-squared). The data-efficiency finding (a few hundred annotated samples suffice) has practical importance for mapping PV in off-grid regions. The authors also disclose the use of ground-truth-derived boxes and provide a preliminary detector feasibility experiment, which is an honest limitation statement.

major comments (2)
  1. [III-F] The three prompting pipelines are not just different prompt inputs; they are also different fine-tuning configurations. Textual-only unfreezes the text encoder, geometric-only unfreezes the geometry encoder, and hybrid unfreezes both, with the mask decoder always unfrozen and the vision encoder and DETR components frozen. Thus 'prompting strategy' is perfectly aliased with the set of trainable components. The conclusion that hybrid prompting outperforms geometric prompting because semantic and spatial guidance are complementary (Conclusions 1 and 4) presupposes that the hybrid advantage comes from the text+box input, but it could equally come from having two unfrozen encoders, or from more favorable optimization dynamics when both encoders are jointly adapted. Likewise, the geometric-versus-textual gap could reflect the geometry encoder being easier to adapt than the text encoder rather than a property of the guidance modality. To support the headline claims, this aliasing must be broken. A feasible design is to evaluate each prompt type under multiple fine-tuning configurations (e.g., all pipelines with both encoders unfrozen, or hybrid with only one encoder unfrozen), or to include a matched-capacity control that varies the number of unfrozen components independently of the prompt input. As written, the reported hierarchy is a joint effect of prompt input and trainable components, not an isolated effect of prompting strategy.
  2. [IV (Figures 3-5; Conclusions 2-4)] All experiments appear to be single runs per condition: no seed variation, confidence intervals, or repeated trials are reported. The paper makes load-bearing robustness claims, such as 'lowest sensitivity to changes in supervision scale, training strategy, spatial resolution, and cross-image variability' (Conclusion 4), and 'consistently' observed hierarchies, but with one run per cell the within-condition stochasticity is unmeasured. The Friedman and Wilcoxon tests treat each pooled setting as an observation, which does not quantify run-to-run variability. This is especially important for the reported decline of textual-prompting Recall and IoU beyond 400 training samples (Section IV-C2), which could be within-run noise. I request that the authors add at least three independent seeds for representative conditions (or otherwise quantify variability, e.g., bootstrap confidence intervals on the pooled distributions), and that they explicitly report the number of runs used in each figure and table. Without this, the stability and sensitivity conclusions are not statistically grounded.
minor comments (5)
  1. [Throughout] There are several typographical and formatting issues: 'Isreal' in the author affiliations, inconsistent spelling of the first author's name between the main text ('Blushtein-Livnon') and reference [4] ('Bluestein-Livnon'), and broken LaTeX artifacts such as 'ANOV A' and 'Wald Type-II' spacing in Section IV-D2 and the supplementary tables.
  2. [Figure 6 caption] The caption states that predictions correspond to the 'best-performing model configuration' for each prompting strategy and image, but it does not specify how this configuration was selected (e.g., highest validation F1, or an oracle over test sets). Please clarify the selection criterion so that the qualitative examples cannot be perceived as cherry-picked.
  3. [III-D] The prompt selection procedure evaluates only M=5 candidate noun phrases. The choice of M is not justified, and although the paper notes that exhaustive search was not the goal, a short discussion of how M affects the reliability of the selected 'solar panels' prompt would strengthen the methodology.
  4. [III-G] All predictions are thresholded at 0.5, but the paper does not report any sensitivity analysis to this threshold. A brief check of thresholds in the range 0.3-0.7 would help confirm that the reported F1/IoU hierarchy is not an artifact of the threshold choice.
  5. [Code/data availability] No code or data availability statement is provided. Given that the primary dataset is private, a public release of the code and the defined train/validation/test splits (or a detailed data description) would materially improve reproducibility for the community.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the study is an empirical evaluation whose prompt selection was performed on a spatially disjoint subset, and no fitted parameter is renamed as a prediction.

full rationale

The paper's central claims (prompting strategy dominates, hybrid > geometric > textual, data efficiency) are supported by controlled experiments and external benchmarks rather than by a derivation chain that reduces to its inputs. Prompt selection used a dedicated dataset from a spatially distinct part of the study area that was excluded from all subsequent experiments (Section III-D), with Eq. 1 and Eq. 2 selecting the best candidate before the main evaluation; therefore the main comparisons are not fitted on the same data they predict. The use of ground-truth-derived bounding boxes for geometric and hybrid prompting is an explicitly disclosed controlled comparison intended to isolate prompt modality from detector errors (Section III-E1 and Section V-F), so the resulting advantage for spatial guidance is a design choice clearly labeled as idealized rather than a hidden circular step. The self-citations to Blushtein-Livnon et al. [5] and [6] are motivational or annotation-protocol references and are not load-bearing for the performance hierarchy; the hierarchy is also reproduced on independent public datasets (France, Queens). The alias between prompt type and which encoders are unfrozen in Section III-F is a potential confound for causal interpretation, but it is a validity concern, not a reduction of the conclusion to its own assumptions by construction. No equation in the paper is equivalent to an output by definition, and no fitted parameter is renamed as a prediction. Accordingly, no circular step is identified and the circularity score is 0.

Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

The central claim rests on four documented assumptions: representativeness of the selected text prompt, confound-free fine-tuning configuration, oracle bounding boxes standing in for operational spatial prompts, and reliability of the annotation protocol. No new physical entities or conserved quantities are introduced. The statistical analyses use conventional fractional logit and ANOVA assumptions.

free parameters (2)
  • Text prompt 'solar panels' = solar panels
    Selected as argmax of S(p_i) over M=5 LLM-generated candidate noun phrases on a dedicated prompt-selection subset (Eq. 2). Textual prompting performance, and therefore the reported hierarchy, depends on this choice; the other four candidates are not listed.
  • Fine-tuning hyperparameters = lr=1e-4, 25 epochs, batch size 2
    Chosen by hand and held fixed for all runs (Section III-F); sensitivity of the central comparison to these values is not tested.
assumptions (4)
  • domain assumption SAM3 is a representative promptable foundation model whose behavior generalizes to other promptable FMs.
    Section V-F states that only SAM3 was evaluated and that generalizing across architectures remains unknown, yet the abstract and conclusions frame prompting strategy as a general determinant of FM behavior.
  • ad hoc to paper The fine-tuning configuration, which unfreezes different modules for each prompting pipeline, does not drive the observed hierarchy.
    Section III-F unfreezes the text encoder for textual-only, the geometry encoder for geometric-only, and both for hybrid; no ablation fixes the prompt type and varies the trainable modules.
  • domain assumption Ground-truth-derived bounding boxes are a valid proxy for spatial guidance in operational pipelines.
    Section III-E1 and Section V-F acknowledge that detector-generated prompts have not yet been evaluated end-to-end; only feasibility results for a YOLO detector are reported.
  • domain assumption The annotation protocol of Blushtein-Livnon et al. [6] produces reliable ground-truth masks.
    Section III-B relies on majority voting by three expert annotators validated in prior work by the same authors; this is not independently re-verified in the present paper.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Evaluating Semantic and Spatial Guidance for Foundation Model Segmentation of Small-Scale PV in Remote Sensing Imagery." pith.science (2026). https://pith.science/paper/GQF65IJN

@misc{pith2026260810801,
  author       = {Pith},
  title        = {Pith review of: Evaluating Semantic and Spatial Guidance for Foundation Model Segmentation of Small-Scale PV in Remote Sensing Imagery},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/GQF65IJN}},
  note         = {Machine review of arXiv:2608.10801}
}
read the original abstract

Spatio-temporal PV data are essential for understanding adoption processes in off-grid regions, yet such data remain largely unavailable. Automated segmentation of remote sensing (RS) imagery offers a promising solution; yet, residential PV systems remain challenging targets because of their small size and sparse distribution, resulting in severe target-background imbalance. Vision-language foundation models (FMs) provide a data-efficient paradigm through prompt-based semantic and spatial guidance, but the relative contribution of different prompt types remains unclear. We systematically evaluate SAM3 for small-scale PV segmentation in RS imagery by comparing textual, geometric, and hybrid prompting, under varying supervision levels, training strategies, spatial resolutions, and imaging conditions. Multi-temporal aerial imagery from a large off-grid rural region serves as a study site, with findings validated across three additional datasets. Prompting strategy emerged as the dominant factor governing model behavior. Textual prompting consistently produced the lowest performance and showed the greatest sensitivity to supervision and imaging conditions. In contrast, spatial guidance substantially improved both segmentation accuracy and robustness. Hybrid prompting achieved the highest accuracy and stability, indicating that semantic and spatial guidance provide complementary information. Most performance gains were achieved with only a few hundred annotated samples, demonstrating strong data efficiency. Transfer learning had limited overall impact, with only modest improvements observed for textual prompting under limited supervision. Overall, our findings establish prompting strategy as a key determinant of SAM3 adaptation, robustness, and generalization, highlighting the potential of promptable FMs for scalable PV mapping in data-constrained off-grid regions.

Figures

Figures reproduced from arXiv: 2608.10801 by the authors.

Figure 1
Figure 1. Experimental framework. SAM3 performance is systematically evaluated as a function of RS data, training regime, and inference prompting strategy. Data variation captures differences in imaging conditions. The training regime compares independent per-image training and sequential TL across varying supervision scales. Prompt-conditioned FT configurations define the prompt type and the model’s unfrozen components. The … view at source ↗
Figure 2
Figure 2. reveals clear performance differences between can￾didate prompts. The prompt solar panels achieves the highest overall score, S(pi)= 0.85, and it significantly outperforms all other candidates. Accordingly, it was selected as the optimal prompt popt [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗
Figure 4
Figure 4. Training regime effects. Panels a–c compare TL and independent (No TL) training strategy across supervision scale and prompting type. Panel d shows the mean F1 gain of TL relative to No TL. TL mainly benefited textual prompting at low supervision, while under geometric and hybrid prompting its effects were largely limited to ZS, with larger gains for geometric prompting. Performance under geometric and hybrid prompt… view at source ↗
Figures from the paper (4 more)
Figure 5
Figure 5. Figure 5: Effect of spatial resolution across prompting strategies and supervision scales. Higher resolution generally improved seg￾mentation performance, with textual prompting showing the greatest sensitivity to resolution. Resolution effects were substantially smaller under g…
Figure 6
Figure 6. Figure 6: Qualitative comparison of prompting strategies across spatial resolutions. Representative examples of PV segmentation under textual, geometric, and hybrid prompting across three spatial resolutions. Predicted masks are decomposed into true positives (TP), false positiv…
Figure 7
Figure 7. Figure 7: Performance of prompting strategies across spatial res￾olutions and supervision scales in the benchmark datasets. The performance hierarchy observed in the primary dataset was preserved, with most gains achieved after the first 200-300 annotated samples. Overall perfor…
Figure 8
Figure 8. Figure 8: EMMs of adjusted F1 scores by prompting strategy and dataset after controlling for supervision scale. Error bars denote 95% CI. Different letters denote significant differences within each dataset. The largest differences occurred between textual and the two spatially …

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

80 extracted references · 71 canonical work pages

  1. [1]

    A deep learning based framework for solar panel segmen- tation and fault classification enhanced with explainable ai

    Adib, A.U.R., Islam, M., Abid, M.S., Ahshan, R., 2025. A deep learning based framework for solar panel segmen- tation and fault classification enhanced with explainable ai. Solar Energy 302, 114058

    work page 2025

  2. [2]

    Flamingo: a visual language model for few-shot learning

    Alayrac, J.B., Donahue, J., Luc, P., Miech, A., Barr, I., Hasson, Y ., Lenc, K., Mensch, A., Millican, K., Reynolds, M., et al., 2022. Flamingo: a visual language model for few-shot learning. Advances in neural infor- mation processing systems 35, 23716–23736

    work page 2022

  3. [3]

    Energy transition at local level: Analyzing the role of peer effects and socio-economic factors on uk solar photovoltaic deployment

    Balta-Ozkan, N., Yildirim, J., Connor, P.M., Truckell, I., Hart, P., 2021. Energy transition at local level: Analyzing the role of peer effects and socio-economic factors on uk solar photovoltaic deployment. Energy Policy 148, 112004. doi:10.1016/j.enpol.2020.112004

    work page arXiv 2021

  4. [4]

    Economic aspects of urban green- ness along a dryland rainfall gradient: A time-series analysis

    Bluestein-Livnon, R., Svoray, T., Dorman, M., Van Der Beek, K., 2023. Economic aspects of urban green- ness along a dryland rainfall gradient: A time-series analysis. Urban Forestry & Urban Greening 83, 127915

    work page 2023

  5. [5]

    On the effectiveness of textual prompting with lightweight fine-tuning for sam3 remote sensing segmentation

    Blushtein-Livnon, R., Rafaeli, O., Ioffe, D., Boger, A., Esquenazi, K.S., Svoray, T., 2026. On the effectiveness of textual prompting with lightweight fine-tuning for sam3 remote sensing segmentation. IEEE Geoscience and Remote Sensing Letters

    work page 2026

  6. [6]

    Performance of human annotators in object detection and segmentation of remotely sensed data

    Blushtein-Livnon, R., Svoray, T., Dorman, M., 2025. Performance of human annotators in object detection and segmentation of remotely sensed data. IEEE Transac- tions on Geoscience and Remote Sensing doi:10.1109/ TGRS.2025.3555235

    arXiv 2025

  7. [7]

    Application of a semantic segmentation convolutional neural network for accurate automatic detection and mapping of solar photovoltaic arrays in aerial imagery

    Camilo, J., Wang, R., Collins, L.M., Bradbury, K., Malof, J.M., 2018. Application of a semantic segmentation convolutional neural network for accurate automatic de- tection and mapping of solar photovoltaic arrays in aerial imagery. arXiv preprint:1801.04018

    work page Pith review arXiv 2018

  8. [8]

    Sam 3: Segment anything with concepts

    Carion, N., Gustafson, L., Hu, Y .T., Debnath, S., Hu, R., et al., 2025. Sam 3: Segment anything with concepts. arXiv preprint:2511.16719

    arXiv 2025

Show all 80 references
  1. [9]

    Rsprompter: Learning to prompt for remote sensing instance segmentation based on visual foundation model

    Chen, K., Liu, C., Chen, H., Zhang, H., Li, W., Zou, Z., Shi, Z., 2024. Rsprompter: Learning to prompt for remote sensing instance segmentation based on visual foundation model. IEEE Transactions on Geoscience and Remote Sensing 62, 1–17

    2024

  2. [10]

    Multiscale adapter based on sam for remote sensing semantic segmentation

    Chen, S., Yu, Y ., Li, Y ., Wang, Z., Li, X., Han, J., 2025a. Multiscale adapter based on sam for remote sensing semantic segmentation. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing 18, 6806–6819

  3. [11]

    Dgtrsd and dgtrsclip: A dual-granularity remote sensing image– text dataset and vision–language foundation model for alignment

    Chen, W., Deng, Y ., Jin, W., Chen, J., Chen, J., Feng, Y ., Xi, Z., Liu, D., Li, K., Meng, Y ., 2025b. Dgtrsd and dgtrsclip: A dual-granularity remote sensing image– text dataset and vision–language foundation model for alignment. IEEE Journal of Selected Topics in Applied Ea...

  4. [12]

    Edge-enhanced sam for extracting photovoltaic power plants from remote sensing imagery

    Chen, Y ., Zhou, J., Chen, Y ., Wang, J., Zhang, X., Ge, Y ., Ma, H., 2025c. Edge-enhanced sam for extracting photovoltaic power plants from remote sensing imagery. International Journal of Applied Earth Observation and Geoinformation 140, 104580

  5. [13]

    Masked-attention mask transformer for universal image segmentation, in: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp

    Cheng, B., Misra, I., Schwing, A.G., Kirillov, A., Gird- har, R., 2022. Masked-attention mask transformer for universal image segmentation, in: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp. 1290–1299

    2022

  6. [14]

    Unleashing the potential of sam for medical adaptation via hierarchical decoding, in: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp

    Cheng, Z., Wei, Q., Zhu, H., Wang, Y ., Qu, L., Shao, W., Zhou, Y ., 2024. Unleashing the potential of sam for medical adaptation via hierarchical decoding, in: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp. 3511–3522

    2024

  7. [15]

    Heterogeneity in the adoption of photovoltaic systems in flanders

    De Groote, O., Pepermans, G., Verboven, F., 2016. Heterogeneity in the adoption of photovoltaic systems in flanders. Energy economics 59, 45–57

    2016

  8. [16]

    Op- timizing zero-shot text-based segmentation of remote sensing imagery using sam and grounding dino

    Diab, M., Kolokoussis, P., Brovelli, M.A., 2025. Op- timizing zero-shot text-based segmentation of remote sensing imagery using sam and grounding dino. Artificial Intelligence in Geosciences 6, 100105

    2025

  9. [17]

    Furedi, T., Kimsal, E., Cornejo, S., Liero, N., Ranalli, J.,

  10. [18]

    Generalized deep learning model for photovoltaic module segmentation from satellite and aerial imagery

    Garc ´ıa, G., Aparcedo, A., Nayak, G.K., Ahmed, T., Shah, M., Li, M., 2024. Generalized deep learning model for photovoltaic module segmentation from satellite and aerial imagery. Solar Energy 274, 112539

    2024

  11. [19]

    Global Off-Grid Solar Market Report: Sales and Impact Data (H2 2022)

    GOGLA, 2023. Global Off-Grid Solar Market Report: Sales and Impact Data (H2 2022). Technical Report. GOGLA. Utrecht, The Netherlands. URL: https://www. gogla.org. accessed 20 July 2025

    2023

  12. [20]

    Spatial patterns of solar photovoltaic system adoption: The influence of neighbors and the built environment

    Graziano, M., Gillingham, K., 2015. Spatial patterns of solar photovoltaic system adoption: The influence of neighbors and the built environment. Journal of Eco- nomic Geography 15, 815–839. doi:10.1093/jeg/ lbu036

    2015 doi

  13. [21]

    Transpv: Refining photovoltaic panel detection accuracy through a vision transformer- based deep learning model

    Guo, Z., Lu, J., Chen, Q., Liu, Z., Song, C., Tan, H., Zhang, H., Yan, J., 2024. Transpv: Refining photovoltaic panel detection accuracy through a vision transformer- based deep learning model. Applied energy 355, 122282

    2024

  14. [22]

    Solarnet: a deep learning framework to map solar power plants in china from satellite imagery

    Hou, X., Wang, B., Hu, W., Yin, L., Wu, H., 2019. Solarnet: a deep learning framework to map solar power plants in china from satellite imagery. arXiv preprint:1912.03685

    2019 arXiv

  15. [23]

    Hulusic, V ., Valenzise, G., Debattista, K., Dufaux, F.,

  16. [24]

    When 17 remote sensing meets foundation model: A survey and beyond

    Huo, C., Chen, K., Zhang, S., Wang, Z., Yan, H., Shen, J., Hong, Y ., Qi, G., Fang, H., Wang, Z., 2025. When 17 remote sensing meets foundation model: A survey and beyond. remote sensing 17, 179

    2025

  17. [25]

    Population without electricity access, 2010–

    IEA, 2025. Population without electricity access, 2010–

    2025

  18. [26]

    Tracking SDG 7: The Energy Progress Report

    IRENA, IEA, UNSD, Bank, W., WHO, 2025. Tracking SDG 7: The Energy Progress Report. Technical Report. IRENA. URL: https://trackingsdg7.esmap.org

    2025

  19. [27]

    Multi-resolution dataset for photovoltaic panel segmentation from satellite and aerial imagery

    Jiang, H., Yao, L., Lu, N., Qin, J., Liu, T., Liu, Y ., Zhou, C., 2021. Multi-resolution dataset for photovoltaic panel segmentation from satellite and aerial imagery. Earth System Science Data Discussions 2021, 1–17

    2021

  20. [28]

    Geoseg: Training-free reasoning- driven segmentation in remote sensing imagery

    Jiang, L., Pei, Y ., Zhao, Y ., Wu, T., Yu, S., Zhang, L., Cai, D., et al., 2026. Geoseg: Training-free reasoning- driven segmentation in remote sensing imagery. arXiv preprint:2603.03983

    2026

  21. [29]

    Ultralytics yolo26: Unified real-time end-to-end vision models

    Jocher, G., Qiu, J., Liu, M., Lyu, S., Akyon, F.C., Kalfaoglu, M.E., 2026. Ultralytics yolo26: Unified real-time end-to-end vision models. URL: https://arxiv.org/abs/2606.03748, doi:10.48550/ arXiv.2606.03748,arXiv:2606.03748

    2026 doi arXiv

  22. [30]

    A crowdsourced dataset of aerial images with annotated solar photovoltaic arrays and installation metadata

    Kasmi, G., Saint-Drenan, Y .M., Trebosc, D., Jolivet, R., Leloux, J., Sarr, B., Dubus, L., 2023. A crowdsourced dataset of aerial images with annotated solar photovoltaic arrays and installation metadata. Scientific Data 10, 59. doi:10.1038/s41597-023-01951-4

    2023 doi

  23. [31]

    Geoai for detection of solar photovoltaic installations in the netherlands

    Kausika, B.B., Nijmeijer, D., Reimerink, I., Brouwer, P., Liem, V ., 2021. Geoai for detection of solar photovoltaic installations in the netherlands. Energy and AI 6, 100111

    2021

  24. [32]

    Segment anything, in: Proceedings of the IEEE/CVF international conference on computer vision, pp

    Kirillov, A., Mintun, E., Ravi, N., Mao, H., Rolland, C., Gustafson, L., Xiao, T., Whitehead, S., Berg, A.C., Lo, W.Y ., et al., 2023. Segment anything, in: Proceedings of the IEEE/CVF international conference on computer vision, pp. 4015–4026

    2023

  25. [33]

    Fine-tuning can distort pretrained features and underperform out-of-distribution

    Kumar, A., Raghunathan, A., Jones, R., Ma, T., Liang, P., 2022. Fine-tuning can distort pretrained features and underperform out-of-distribution. arXiv preprint:2202.10054

    2022 arXiv

  26. [34]

    The group robustness is in the details: Revisiting finetuning under spurious correlations

    LaBonte, T., Hill, J.C., Zhang, X., Muthukumar, V ., Kumar, A., 2024. The group robustness is in the details: Revisiting finetuning under spurious correlations. Advances in Neural Information Processing Systems 37, 121598–121629

    2024

  27. [35]

    A review of remote sensing image segmentation by deep learning methods

    Li, J., Cai, Y ., Li, Q., Kou, M., Zhang, T., 2024a. A review of remote sensing image segmentation by deep learning methods. International Journal of Digital Earth 17, 2328827

  28. [36]

    Segearth-ov: Towards training- free open-vocabulary segmentation for remote sensing images, in: Proceedings of the Computer Vision and Pattern Recognition Conference, pp

    Li, K., Liu, R., Cao, X., Bai, X., Zhou, F., Meng, D., Wang, Z., 2025a. Segearth-ov: Towards training- free open-vocabulary segmentation for remote sensing images, in: Proceedings of the Computer Vision and Pattern Recognition Conference, pp. 10545–10556

  29. [37]

    Joint-task learning framework with scale adaptive and position guidance modules for improved household rooftop photovoltaic segmentation in remote sensing image

    Li, L., Lu, N., Qin, J., 2025b. Joint-task learning framework with scale adaptive and position guidance modules for improved household rooftop photovoltaic segmentation in remote sensing image. Applied Energy 377, 124521

  30. [38]

    Understanding rooftop pv panel semantic segmentation of satellite and aerial images for better using machine learning

    Li, P., Zhang, H., Guo, Z., Lyu, S., Chen, J., Li, W., Song, X., Shibasaki, R., Yan, J., 2021. Understanding rooftop pv panel semantic segmentation of satellite and aerial images for better using machine learning. Advances in applied energy 4, 100057

    2021

  31. [39]

    Reobench: Benchmarking robustness of earth observation foundation models

    Li, X., Tao, Y ., Zhang, S., Liu, S., Xiong, Z., Luo, C., Liu, L., Pechenizkiy, M., Zhu, X., Huang, T., 2026. Reobench: Benchmarking robustness of earth observation foundation models. Advances in Neural Information Processing Systems 38

    2026

  32. [40]

    Vision-language models in remote sensing: Current progress and future trends

    Li, X., Wen, C., Hu, Y ., Yuan, Z., Zhu, X.X., 2024b. Vision-language models in remote sensing: Current progress and future trends. IEEE Geoscience and Remote Sensing Magazine 12, 32–66

  33. [41]

    Remoteclip: A vision language foundation model for remote sensing

    Liu, F., Chen, D., Guan, Z., Zhou, X., Zhu, J., Ye, Q., Fu, L., Zhou, J., 2024. Remoteclip: A vision language foundation model for remote sensing. IEEE Transactions on Geoscience and Remote Sensing 62, 1–16

    2024

  34. [42]

    Pointsam: Pointly-supervised segment anything model for remote sensing images

    Liu, N., Xu, X., Su, Y ., Zhang, H., Li, H.C., 2025. Pointsam: Pointly-supervised segment anything model for remote sensing images. IEEE Transactions on Geo- science and Remote Sensing 63, 1–15

    2025

  35. [43]

    Pv identifier: Extraction of small-scale distributed photovoltaics in complex en- vironments from high spatial resolution remote sensing images

    Lu, N., Li, L., Qin, J., 2024. Pv identifier: Extraction of small-scale distributed photovoltaics in complex en- vironments from high spatial resolution remote sensing images. Applied Energy 365, 123311

    2024

  36. [44]

    What drives solar energy adoption in developing countries? evidence from household-level data

    Mahn, D., Kammen, D.M., Hirth, L., 2024. What drives solar energy adoption in developing countries? evidence from household-level data. Energy Economics 138, 107924. doi:10.1016/j.eneco.2024.107924

    2024

  37. [45]

    Malof, J.M., Bradbury, K., Collins, L.M., Newell, R.G.,

  38. [46]

    The segment anything model (sam) for remote sensing applications: From zero to one shot

    Osco, L.P., Wu, Q., De Lemos, E.L., Gonc ¸alves, W.N., Ramos, A.P.M., Li, J., Junior, J.M., 2023. The segment anything model (sam) for remote sensing applications: From zero to one shot. International Journal of Applied Earth Observation and Geoinformation 124, 103540

    2023

  39. [47]

    Prompt-tuning sam: From generalist to specialist with only 2048 pa- rameters and 16 training images, in: Proceedings of the Computer Vision and Pattern Recognition Conference, pp

    Piater, T., Barz, B., Freytag, A., 2025. Prompt-tuning sam: From generalist to specialist with only 2048 pa- rameters and 16 training images, in: Proceedings of the Computer Vision and Pattern Recognition Conference, pp. 4688–4698

    2025

  40. [48]

    Consumer-grade uav imagery facilitates semantic segmentation of species-rich savanna tree layers

    Popp, M.R., Kalwij, J.M., 2023. Consumer-grade uav imagery facilitates semantic segmentation of species-rich savanna tree layers. Scientific Reports 13, 13892

    2023

  41. [49]

    Learning transferable visual models from natural language supervision, in: International conference on machine learning, PmLR

    Radford, A., Kim, J.W., Hallacy, C., Ramesh, A., Goh, G., Agarwal, S., Sastry, G., Askell, A., Mishkin, P., Clark, J., et al., 2021. Learning transferable visual models from natural language supervision, in: International conference on machine learning, PmLR. pp. 8748–8763

    2021

  42. [50]

    Segment anything, from space?, in: Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, pp

    Ren, S., Luzi, F., Lahrichi, S., Kassaw, K., Collins, L.M., Bradbury, K., Malof, J.M., 2024. Segment anything, from space?, in: Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, pp. 8355–8365

    2024

  43. [51]

    Geosam: Fine-tuning sam with multi- modal prompts for mobility infrastructure segmentation

    Sultan, R.I., Li, C., Zhu, H., Khanduri, P., Brocanelli, 18 M., Zhu, D., 2023. Geosam: Fine-tuning sam with multi- modal prompts for mobility infrastructure segmentation. arXiv preprint:2311.11319

    2023 arXiv

  44. [52]

    General generative ai- based image augmentation method for robust rooftop pv segmentation

    Tan, H., Guo, Z., Lin, Z., Chen, Y ., Huang, D., Yuan, W., Zhang, H., Yan, J., 2024. General generative ai- based image augmentation method for robust rooftop pv segmentation. Applied Energy 368, 123554

    2024

  45. [53]

    Enhancing pv panel segmentation in remote sensing images with constraint refinement modules

    Tan, H., Guo, Z., Zhang, H., Chen, Q., Lin, Z., Chen, Y ., Yan, J., 2023. Enhancing pv panel segmentation in remote sensing images with constraint refinement modules. Applied Energy 350, 121757

    2023

  46. [54]

    Wikidata: a free collaborative knowledgebase

    Vrande ˇci´c, D., Kr ¨otzsch, M., 2014. Wikidata: a free collaborative knowledgebase. Communications of the ACM 57, 78–85

    2014

  47. [55]

    Samrs: Scaling-up remote sens- ing segmentation dataset with segment anything model

    Wang, D., Zhang, J., Du, B., Xu, M., Liu, L., Tao, D., Zhang, L., 2023. Samrs: Scaling-up remote sens- ing segmentation dataset with segment anything model. Advances in Neural Information Processing Systems 36, 8815–8827

    2023

  48. [56]

    Pv segmenter: A frequency-guided edge-aware network for distributed photovoltaic segmentation in remote sensing imagery

    Wang, S., Shao, Z., Hou, D., Cai, B., 2025. Pv segmenter: A frequency-guided edge-aware network for distributed photovoltaic segmentation in remote sensing imagery. Applied Energy 393, 126137

    2025

  49. [57]

    Cris: Clip-driven referring image segmentation, in: Proceedings of the IEEE/CVF conference on com- puter vision and pattern recognition, pp

    Wang, Z., Lu, Y ., Li, Q., Tao, X., Guo, Y ., Gong, M., Liu, T., 2022. Cris: Clip-driven referring image segmentation, in: Proceedings of the IEEE/CVF conference on com- puter vision and pattern recognition, pp. 11686–11695

    2022

  50. [58]

    Robust fine-tuning of zero-shot models, in: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp

    Wortsman, M., Ilharco, G., Kim, J.W., Li, M., Kornblith, S., Roelofs, R., Lopes, R.G., Hajishirzi, H., Farhadi, A., Namkoong, H., et al., 2022. Robust fine-tuning of zero-shot models, in: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp. 7959–7971

    2022

  51. [59]

    Segformer: Simple and efficient design for semantic segmentation with transformers

    Xie, E., Wang, W., Yu, Z., Anandkumar, A., Alvarez, J.M., Luo, P., 2021. Segformer: Simple and efficient design for semantic segmentation with transformers. Advances in neural information processing systems 34, 12077–12090

    2021

  52. [60]

    Segearth-r2:towards comprehensive language-guided segmentation for remote sensing images

    Xin, Z., Li, K., Chen, L., Li, W., Xiao, Y ., Qiao, H., Zhang, W., Meng, D., Cao, X., 2025. Segearth-r2:towards comprehensive language-guided segmentation for remote sensing images. arXiv preprint:2512.20013

    2025

  53. [61]

    Remotesam: Towards segment anything for earth observation, in: Proceedings of the 33rd ACM International Conference on Multime- dia, pp

    Yao, L., Liu, F., Chen, D., Zhang, C., Wang, Y ., Chen, Z., Xu, W., Di, S., Zheng, Y ., 2025. Remotesam: Towards segment anything for earth observation, in: Proceedings of the 33rd ACM International Conference on Multime- dia, pp. 3027–3036

    2025

  54. [62]

    Pvsam: Adapting geometric prompts to segment anything model for photovoltaic detection in remote sensing im- agery

    Yao, X., Zhang, S., Liang, Z., Li, J., Liu, C., 2026. Pvsam: Adapting geometric prompts to segment anything model for photovoltaic detection in remote sensing im- agery. Applied Energy 404, 127137

    2026

  55. [63]

    Calibrating multi-modal representations: A pursuit of group robustness without annotations, in: 2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), IEEE

    You, C., Mint, Y ., Dai, W., Sekhon, J.S., Staib, L., Dun- can, J.S., 2024. Calibrating multi-modal representations: A pursuit of group robustness without annotations, in: 2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), IEEE. pp. 26140–26150

    2024

  56. [64]

    Deepsolar: A machine learning framework to efficiently construct a solar deployment database in the united states

    Yu, J., Wang, Z., Majumdar, A., Rajagopal, R., 2018. Deepsolar: A machine learning framework to efficiently construct a solar deployment database in the united states. Joule 2, 2605–2617

    2018

  57. [65]

    Chatearthnet: A global-scale image-text dataset empow- ering vision-language geo-foundation models

    Yuan, Z., Xiong, Z., Mou, L., Zhu, X.X., 2025. Chatearthnet: A global-scale image-text dataset empow- ering vision-language geo-foundation models. Earth System Science Data Discussions 2024, 1–24

    2025

  58. [66]

    Toward global rooftop pv detection with deep active learning

    Zech, M., Tetens, H.P., Ranalli, J., 2024. Toward global rooftop pv detection with deep active learning. Advances in Applied Energy 16, 100191

    2024

  59. [67]

    Rsam-seg: A sam-based model with prior knowledge integration for remote sensing image semantic segmen- tation

    Zhang, J., Li, Y ., Yang, X., Jiang, R., Zhang, L., 2025. Rsam-seg: A sam-based model with prior knowledge integration for remote sensing image semantic segmen- tation. Remote Sensing 17, 590

    2025

  60. [68]

    Zhang, J., Zhou, Z., Mai, G., Hu, M., Guan, Z., Li, S., Mu, L., 2024a. Text2seg: Zero-shot remote sens- ing image semantic segmentation via text-guided visual foundation models, in: Proceedings of the 7th ACM SIGSPATIAL International workshop on AI for geo- graphic knowledge d...

  61. [69]

    Segclip: Multimodal visual-language and prompt learning for high-resolution remote sensing semantic segmentation

    Zhang, S., Zhang, B., Wu, Y ., Zhou, H., Jiang, J., Ma, J., 2024b. Segclip: Multimodal visual-language and prompt learning for high-resolution remote sensing semantic segmentation. IEEE Transactions on Geoscience and Remote Sensing

  62. [70]

    Rs5m and georsclip: A large-scale vision-language dataset and a large vision-language model for remote sensing

    Zhang, Z., Zhao, T., Guo, Y ., Yin, J., 2024c. Rs5m and georsclip: A large-scale vision-language dataset and a large vision-language model for remote sensing. IEEE Transactions on Geoscience and Remote Sensing 62, 1– 23

  63. [71]

    Zhao, Z., Alzubaidi, L., Zhang, J., Duan, Y ., Gu, Y .,

  64. [72]

    Enhancing visual feature constraints in segmentation models for photovoltaic panel recognition

    Zhao, Z., Li, K., Chen, Y ., Wang, J., 2025. Enhancing visual feature constraints in segmentation models for photovoltaic panel recognition. Energy and AI 21, 100544

    2025

  65. [73]

    Geoground: A unified large vision-language model for remote sensing visual grounding

    Zhou, Y ., Lan, M., Li, X., Feng, L., Ke, Y ., Jiang, X., Li, Q., Yang, X., Zhang, W., 2024. Geoground: A unified large vision-language model for remote sensing visual grounding. arXiv preprint:2411.11904

    2024 arXiv

  66. [74]

    Deep learning in remote sensing: A comprehensive review and list of resources

    Zhu, X.X., Tuia, D., Mou, L., Xia, G.S., Zhang, L., Xu, F., Fraundorfer, F., 2017. Deep learning in remote sensing: A comprehensive review and list of resources. IEEE geoscience and remote sensing magazine 5, 8–36. 19 SUPPLEMENTARYTABLES TABLE S1.Type-II ANOV A results for yea...

    2017

  67. [207]

    doi:10.1038/s41597-025-06523-2

    doi

  68. [2016]

    Applied energy 183, 229–240

    Automatic detection of solar photovoltaic arrays in high resolution aerial imagery. Applied energy 183, 229–240

  69. [2017]

    Robust dynamic range computation for high dy- namic range content, in: IS&T International Symposium on Electronic Imaging

  70. [2024]

    Expert Systems with Applications 242, 122807

    A comparison review of transfer learning and self- supervised learning: Definitions, applications, advantages and limitations. Expert Systems with Applications 242, 122807

  71. [2025]

    IEA, Paris

    https://www.iea.org/data-and-statistics/charts/ population-without-electricity-access-2010-2025. IEA, Paris

    2010

  72. [2026]

    Scientific Data 13,

    Labeled photovoltaic installations for orthographic aerial imagery in queens, new york. Scientific Data 13,

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.