Pith. sign in

REVIEW 3 major objections 81 references

Evidential learning makes remote sensing image–text retrieval uncertainty-aware and more reliable under sensor, atmosphere, and vocabulary noise.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · grok-4.5

2026-07-11 01:33 UTC pith:KTBBAMZI

load-bearing objection Solid systems paper that packages EDL, correctness-aligned uncertainty, and selective RS-TTA for noisy CMRSITR; gains are real on their synthetic splits but transfer remains unproven. the 3 major comments →

arxiv 2607.06032 v1 pith:KTBBAMZI submitted 2026-07-07 cs.IR

Uncertainty-Aware Cross-Modal Remote Sensing Image-Text Retrieval via Evidential Learning

Zhuoyue Wang , Xueqian Wang , Gang Li , Chengxi Li , Yongpan Liu , Yifang Ban This is my paper
classification cs.IR
keywords cross-modal retrievalremote sensingevidential learninguncertainty estimationtest-time augmentationimage-text matchingDirichlet distributionCMRSITR
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

Standard remote sensing image–text retrieval systems treat every query as certain, even when haze, radiometric drift, sensor striping, or inconsistent RS vocabulary make the match unreliable. This paper introduces ELC: it models batch-wise image–text correspondences as Dirichlet distributions so each query gets an uncertainty score, trains that score to be high when retrieval would be wrong and low when it would be right, and distills intra-modal similarity structure from frozen mentor encoders so the evidence is more discriminative. At test time a fixed deferral ratio routes only high-uncertainty queries through remote-sensing-aware augmentations (radiometric affine, small rotation, curated RS lexicon) while confident queries return immediately. On clean RSICD and RSITMD the method stays competitive with strong CLIP-based baselines; on synthetically degraded 50/50 noisy splits the advantage widens. The practical claim is a selective retrieval pipeline that can tell when to trust its ranking and when to look harder under RS-specific degradations.

Core claim

Modeling inter-modal RS image–text matches with evidential (Dirichlet) learning, aligning the resulting uncertainty with retrieval correctness, distilling mentor-side intra-modal structure into the trainable encoders, and gating RS-aware test-time augmentation on a fixed high-uncertainty fraction yields competitive clean-set retrieval and stronger robustness than existing CMRSITR methods under the evaluated sensor- and atmosphere-related image perturbations and RS-vocabulary heterogeneity.

What carries the argument

Evidential learning (EDL) that turns in-batch similarity scores into non-negative evidence for a Dirichlet over correspondences, so per-query uncertainty is the reciprocal of total concentration; uncertainty–correctness alignment (UCL) and intra-modal relationship learning (RL) make that uncertainty informative; a fixed-deferral-ratio gate then applies RS-aware test-time augmentation only to high-uncertainty queries.

Load-bearing premise

The synthetic noise pipeline (haze, radiometric affine, readout/stripe noise, and a 174-entry RS lexicon) plus a fixed 10 percent deferral budget produce uncertainty scores and refinement gains that transfer to real operational degradations and query streams.

What would settle it

Train ELC and strong baselines on the same clean RSICD/RSITMD splits, then evaluate both on real satellite acquisitions with documented atmospheric or sensor artifacts and captions from different annotation campaigns; if ELC no longer shows a larger relative RSUM gain under those real degradations, or if its uncertainty no longer ranks noisy queries higher, the robustness claim fails.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

Share X Bluesky LinkedIn Reddit HN

If this is right

  • Low-uncertainty queries can be returned immediately while compute is spent only on the uncertain fraction.
  • Retrieval quality degrades less under haze, radiometric drift, readout/stripe noise, and RS synonym drift than fully confident baselines.
  • Per-query uncertainty separates clean from noisy samples and can serve as an operational reliability signal.
  • RS-specific augmentations outperform generic color jitter, flips, and EDA-style synonym replacement for deferred queries.
  • Uncertainty-gated selective inference extends the conventional CMRSITR pipeline without changing the gallery encoding.

Where Pith is reading between the lines

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

  • The same uncertainty gate could trigger human review in disaster-response or monitoring workflows when scores spike.
  • Adaptive deferral thresholds may be required when the live clean/noisy mix differs from the paper’s 50/50 test construction.
  • Stronger domain-specific mentors could further tighten Dirichlet evidence without redesigning the EDL head.
  • The framework is naturally portable to other cross-modal retrieval settings that face domain-specific sensor and vocabulary noise.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 0 minor

Summary. The paper proposes ELC, an evidential-learning framework for cross-modal remote sensing image–text retrieval that estimates per-query uncertainty and uses it for selective inference. Training combines Dirichlet-based inter-modal evidential learning (with uncertainty-aware weighting), an uncertainty–correctness alignment loss (UCL) that encourages high uncertainty on incorrect retrievals, and intra-modal relationship learning (RL) that distills similarity structure from frozen mentor encoders. At test time, queries whose uncertainty exceeds a threshold set by a fixed deferral ratio are refined by RS-aware test-time augmentation (radiometric affine, small-angle rotation, RS-vocabulary substitution), while low-uncertainty queries are returned directly. On RSICD and RSITMD the method is competitive with strong CLIP-based baselines (including RemoteCLIP); on author-constructed noisy 50/50 clean/noisy splits it reports clearer RSUM gains, supported by ablations, uncertainty histograms, and a comparison of RS-TTA against generic TTA.

Significance. If the robustness results transfer beyond the synthetic noise model, the work is a useful step for operational CMRSITR: most prior methods return fully confident rankings and ignore query-level reliability under sensor/atmosphere and caption heterogeneity. The combination of EDL with an explicit correctness-alignment objective (UCL), mentor-based intra-modal distillation (RL), and uncertainty-gated RS-TTA is a coherent pipeline rather than a single loss tweak. Strengths include a full training/test algorithm, systematic ablations of UW/UCL/RL (Tables III–IV), uncertainty distribution analysis (Fig. 7), hyperparameter and deferral-ratio sensitivity, and a controlled RS-TTA vs generic-TTA comparison (Tables V–VI). The main significance is therefore methodological and empirical within the evaluated degradation protocol, not yet a demonstrated real-world robustness guarantee.

major comments (3)
  1. The central robustness claim (Abstract; Tables I–II) rests entirely on synthetic degradations defined in §IV-A (Koschmieder haze with fixed A=0.85 and t(x)∈[0.55,0.80], per-channel radiometric affine, readout σ=0.02, stripe σ=0.025, and a 174-entry lexicon with p_sub=0.3 on a 50/50 clean/noisy split). RS-TTA operators in §III-E (radiometric affine, ±15° rotation, same lexicon) partially mirror those generators. Tables V–VI show RS-TTA beats generic TTA inside this protocol, but that does not establish transfer to real sensor/atmosphere streams or independent vocabulary drift. Please either (i) add at least one evaluation on real degraded acquisitions or independently sourced caption variants, or (ii) substantially qualify the claim to “under the evaluated synthetic RS-specific degradations” throughout Abstract, Introduction, and Conclusion, and discuss the risk of operator–noise alignmen
  2. §III-E and Algorithm 1 set the deferral threshold τ by ranking the current test-query uncertainties and cutting at fixed ρ=0.1. This is a test-set percentile, not a fixed absolute or online threshold available at deployment when queries arrive sequentially. Fig. 11 shows sensitivity to ρ, but the reported gains assume oracle knowledge of the full query-set uncertainty distribution. Clarify how τ would be chosen in streaming settings (validation calibration, absolute u threshold, adaptive budget) and report performance under a threshold fixed from validation only, without re-ranking the test set.
  3. Tables I–II report single-run point estimates with no error bars, seeds, or significance tests, while several clean-set margins over RemoteCLIP are small (e.g., RSICD RSUM 215.20 vs 211.57). Given free parameters b1,b2,b3 and ρ (§IV-B, Figs. 9–11), the competitive clean-set claim and the ~9-point noisy RSICD gap need multi-seed means±std or a paired significance test so that ranking claims are not driven by a single training trajectory.

Circularity Check

1 steps flagged

No definitional circularity in the method or claims; only a minor conference self-citation of the authors' own prior version and ordinary hyperparameter selection on held-out data.

specific steps
  1. self citation load bearing [Abstract footnote / Introduction / Related Work closing paragraph / Ref. [1]]
    "Part of this paper has been published by International Geoscience and Remote Sensing Symposium 2025 [1]. ... This paper significantly extends the content of its conference version [1] by linking estimated per-query uncertainty to test-time retrieval..."

    The only self-citation is the authors’ own shorter conference precursor. It is not used to justify a uniqueness theorem, forbid alternatives, or supply a load-bearing mathematical premise; the journal text treats [1] as prior work being extended. This is ordinary self-citation of an earlier version and raises the score only to 1, not higher.

full rationale

ELC is an empirical engineering method (EDL Dirichlet modeling of batch similarities + UCL alignment loss + mentor distillation RL + uncertainty-gated RS-TTA). The training losses (Eqs. 5–19) and test procedure (Algorithm 1, fixed deferral ratio ρ) are not algebraic restatements of the R@L/RSUM metrics; ablations (Tables III–IV) and sensitivity plots (Figs. 9–11) treat components as independent design choices whose contribution is measured, not forced by construction. The sole self-citation [1] is the authors’ own IGARSS 2025 conference precursor; the journal version explicitly states it extends that work with the uncertainty-gated test stage and expanded noise experiments, so [1] is not a load-bearing uniqueness theorem or hidden premise. Hyperparameters (b1=40, b2=b3=1, ρ=0.1) and the synthetic noise pipeline are selected/analyzed on the evaluation sets in the usual ML fashion; this is ordinary tuning, not a fitted parameter renamed as a first-principles prediction. No self-definitional loop, no uniqueness imported from the authors, and no ansatz smuggled via citation appears in the derivation chain. The robustness claim remains an empirical comparison against external SOTA baselines under author-defined degradations, which may raise external-validity questions but does not constitute circularity of the claimed derivation.

Axiom & Free-Parameter Ledger

4 free parameters · 4 axioms · 2 invented entities

The claim rests on standard EDL/Dirichlet machinery, CLIP-style dual encoders, hand-chosen loss weights and deferral ratio, and a synthetic noise model whose parameters are set by the authors. No new physical entities; the invented pieces are algorithmic modules whose value is measured only inside the paper’s own benchmarks.

free parameters (4)
  • b1 (KL annealing scale) = 40
    Set to 40; controls strength of KL term in EDL loss (Eq. 8–9).
  • b2, b3 (UCL and RL weights) = 1, 1
    Both set to 1 in the total loss (Eq. 19); chosen by sensitivity plots on validation.
  • deferral ratio ρ = 0.1
    Fixed at 0.1 so that the most-uncertain 10% of queries receive RS-TTA; selected after sweeping {0,0.05,0.1,0.15,0.2} on noisy test sets.
  • noise generation parameters (A, t(x), g_c, β_c, σ_readout, σ_stripe, p_sub) = A=0.85, t∈[0.55,0.80], g∈[0.90,1.10], β∈[-0.03,0.03], σ_r=0.02, σ_s=0.025, p_sub=0.30
    Hand-set values that define the synthetic Noisy RSICD/RSITMD tests (Section IV-A); performance claims under degradation depend on them.
axioms (4)
  • domain assumption Inter-modal similarity scores can be treated as non-negative evidence for a Dirichlet distribution over in-batch correspondences (standard EDL assumption).
    Section III-B, Eqs. 2–4; inherited from Sensoy et al. and applied without new proof that batch-wise Dirichlet is calibrated for retrieval.
  • ad hoc to paper Uncertainty u = K/S derived from total Dirichlet strength is a usable reliability signal once aligned by UCL.
    Section III-C and Fig. 2–3 motivate UCL precisely because raw EDL uncertainty is not automatically aligned with correctness.
  • domain assumption Frozen Unicom and GTE mentor encoders supply useful intra-modal neighborhood structure for RS images and captions.
    Section III-D; mentors are not updated and their quality is taken as given.
  • ad hoc to paper The synthetic sensor/atmosphere and RS-vocabulary noise model is representative of the non-ideal conditions claimed in the introduction.
    Section IV-A; all robustness claims are measured on this constructed distribution.
invented entities (2)
  • Uncertainty–correctness alignment learning (UCL) loss no independent evidence
    purpose: Force high uncertainty on incorrect top-1 matches and low uncertainty on correct ones so that the EDL score can gate TTA.
    Defined in Eqs. 12–14; value demonstrated only via ablation on the paper’s own metrics.
  • RS-aware test-time augmentation (RS-TTA) operators no independent evidence
    purpose: Radiometric affine, ±15° rotation, and curated RS-lexicon substitution applied only to high-uncertainty queries.
    Section III-E; shown better than generic TTA on the authors’ noisy sets, no external validation.

pith-pipeline@v1.1.0-grok45 · 31762 in / 3253 out tokens · 37838 ms · 2026-07-11T01:33:42.219123+00:00 · methodology

0 comments
Cite this review

Pith. "Pith review of Uncertainty-Aware Cross-Modal Remote Sensing Image-Text Retrieval via Evidential Learning." pith.science (2026). https://pith.science/paper/KTBBAMZI

@misc{pith2026260706032,
  author       = {Pith},
  title        = {Pith review of: Uncertainty-Aware Cross-Modal Remote Sensing Image-Text Retrieval via Evidential Learning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/KTBBAMZI}},
  note         = {Machine review of arXiv:2607.06032}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

In cross-modal remote sensing image-text retrieval (CMRSITR), test-time remote sensing (RS) images and textual descriptions may deviate from well-curated benchmark conditions due to sensor- and atmosphere-related image degradations and text-side RS-vocabulary heterogeneity. Under such non-ideal conditions, existing CMRSITR methods may produce unreliable retrieval results because they perform retrieval with full certainty for each query and do not distinguish the varying uncertainty across queries. To address this issue, we propose an evidential learning-based CMRSITR (ELC) method for uncertainty-aware retrieval. During the training phase of ELC, evidential learning (EDL) is employed to model the inter-modal correspondences between RS images and textual descriptions as Dirichlet distributions, from which the uncertainty of each query can be obtained. Based on the EDL outputs, uncertainty-correctness alignment learning (UCL) is introduced to align the estimated uncertainty with retrieval correctness, encouraging high uncertainty for incorrect retrieval and low uncertainty for correct retrieval. Furthermore, intra-modal relationship learning (RL) distills the intra-modal similarity structure from pretrained mentor encoders for the trainable encoders, thereby making the Dirichlet distributions modeled by EDL more discriminative. In the test phase of ELC, the estimated uncertainty is compared with a threshold determined by a fixed deferral ratio, where low-uncertainty queries are directly returned and high-uncertainty queries are refined by RS-aware test-time augmentation (RS-TTA). Experimental results demonstrate that ELC achieves competitive retrieval performance compared with state-of-the-art CMRSITR methods and provides stronger robustness under the evaluated RS-specific degradations, including sensor- and atmosphere-related image perturbations and RS-vocabulary heterogeneity.

Figures

Figures reproduced from arXiv: 2607.06032 by Chengxi Li, Gang Li, Xueqian Wang, Yifang Ban, Yongpan Liu, Zhuoyue Wang.

Figure 3
Figure 3. Figure 3: Illustration of uncertainty–correctness alignment in UCL. The pentagram denotes the correct decision point. (a) For correct decisions, UCL encourages lower uncertainty. (b) For incorrect decisions, UCL encourages higher uncertainty [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗

Review history (2 revisions) →

discussion (0)

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

Reference graph

Works this paper leans on

81 extracted references · 81 canonical work pages · 4 internal anchors

  1. [1]

    Cecmr: Calibrated evidential learning for cross modal remote sensing image -text retrieval,

    Z. Wang, X. Wang, and G. Li, “Cecmr: Calibrated evidential learning for cross modal remote sensing image -text retrieval,” in Proc. IEEE Int. Geosci. Remote Sens. Symp. (IGARSS), 2025, pp. 6345-6349. Fig. 9. Sensitivity analysis of hyperparameters on the RSICD dataset. Fig. 10. Sensitivity analysis of hyperparameters on the RSITMD dataset. (a) (b) Fig. 11...

    work page 2025

  2. [2]

    Big data for remote sensing: Challenges and opportunities,

    M. Chi, A. Plaza, J. A. Benediktsson, Z. Sun, J. Shen, and Y. Zhu, “Big data for remote sensing: Challenges and opportunities,” Proc. IEEE, vol. 104, no. 11, pp. 2207-2219, 2016

    work page 2016

  3. [3]

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

    X. X. Zhu et al. , “Deep learning in remote sensing: A comprehensive review and list of resources,” IEEE Geosci. Remote Sens. Mag., vol. 5, no. 4, pp. 8-36, 2017

    work page 2017

  4. [4]

    Remote sensing image retrieval in the past decade: Achievements, challenges, and future directions,

    W. Zhou, H. Guan, Z. Li, Z. Shao, and M. R. Delavar, “Remote sensing image retrieval in the past decade: Achievements, challenges, and future directions,” IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens., vol. 16, pp. 1447-1473, 2023

    work page 2023

  5. [5]

    Image retrieval from remote sensing big data: A survey,

    Y. Li, J. Ma, and Y. Zhang, “Image retrieval from remote sensing big data: A survey,” Inf. Fusion, vol. 67, pp. 94-115, Mar. 2021

    work page 2021

  6. [6]

    Exploring a fine -grained multiscale method for cross - modal remote sensing image retrieval,

    Z. Yuan et al., “Exploring a fine -grained multiscale method for cross - modal remote sensing image retrieval,” IEEE Trans. Geosci. Remote Sens., vol. 60, pp. 1-19, 2022

    work page 2022

  7. [7]

    Exploring uni -modal feature learning on entities and relations for remote sensing cross -modal text -image retrieval,

    S. Zhang, Y. Li, and S. Mei, “Exploring uni -modal feature learning on entities and relations for remote sensing cross -modal text -image retrieval,” IEEE Trans. Geosci. Remote Sens., vol. 61, pp. 1-17, 2023

    work page 2023

  8. [8]

    Global –local information soft-alignment for cross -modal remote -sensing image –text retrieval,

    G. Hu, Z. Wen, Y. Lv, J. Zhang, and Q. Wu, “Global –local information soft-alignment for cross -modal remote -sensing image –text retrieval,” IEEE Trans. Geosci. Remote Sens., vol. 62, pp. 1-15, 2024

    work page 2024

  9. [9]

    Remote sensing cross -modal text-image retrieval based on global and local information,

    Z. Yuan et al., “Remote sensing cross -modal text-image retrieval based on global and local information,” IEEE Trans. Geosci. Remote Sens., vol. 60, pp. 1-16, 2022

    work page 2022

  10. [10]

    Interacting - enhancing feature transformer for cross-modal remote-sensing image and text retrieval,

    X. Tang, Y. Wang, J. Ma, X. Zhang, F. Liu, and L. Jiao, “Interacting - enhancing feature transformer for cross-modal remote-sensing image and text retrieval,” IEEE Trans. Geosci. Remote Sens., vol. 61, pp. 1 -15, 2023

    work page 2023

  11. [11]

    Transcending fusion: A multiscale alignment method for remote sensing image–text retrieval,

    R. Yang et al., “Transcending fusion: A multiscale alignment method for remote sensing image–text retrieval,” IEEE Trans. Geosci. Remote Sens., vol. 62, pp. 1-17, 2024

    work page 2024

  12. [12]

    Spatial–channel attention transformer with pseudo regions for remote sensing image -text retrieval,

    D. Wu et al., “Spatial–channel attention transformer with pseudo regions for remote sensing image -text retrieval,” IEEE Trans. Geosci. Remote Sens., vol. 62, pp. 1-15, 2024

    work page 2024

  13. [13]

    Multilanguage transformer for improved text to remote sensing image retrieval,

    M. M. A. Rahhal, Y. Bazi, N. A. Alsharif, L. Bashmal, N. Alajlan, and F. Melgani, “Multilanguage transformer for improved text to remote sensing image retrieval,” IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens., vol. 15, pp. 9115-9126, 2022

    work page 2022

  14. [14]

    Cross -modal prealigned method with global and local information for remote sensing image and text retrieval,

    Z. Sun, M. Zhao, G. Liu, and A. Kaup, “Cross -modal prealigned method with global and local information for remote sensing image and text retrieval,” IEEE Trans. Geosci. Remote Sens., vol. 62, pp. 1-18, 2024

    work page 2024

  15. [15]

    Visual global -salient-guided network for remote sensing image -text retrieval,

    Y. He, X. Xu, H. Chen, J. Li, and F. Pu, “Visual global -salient-guided network for remote sensing image -text retrieval,” IEEE Trans. Geosci. Remote Sens., vol. 62, pp. 1-14, 2024

    work page 2024

  16. [16]

    Context -aware local –global semantic alignment for remote sensing image–text retrieval,

    X. Chen, X. Zheng, and X. Lu, “Context -aware local –global semantic alignment for remote sensing image–text retrieval,” IEEE Trans. Geosci. Remote Sens., vol. 63, pp. 1-12, 2025

    work page 2025

  17. [17]

    Cross-modal progressive perspective matching network for remote sensing image -text retrieval,

    C. Zheng et al., “Cross-modal progressive perspective matching network for remote sensing image -text retrieval,” IEEE Trans. Multimedia, vol. 27, pp. 1-13, 2025

    work page 2025

  18. [18]

    Scale - semantic joint decoupling network for image -text retrieval in remote sensing,

    C. Zheng, N. Song, R. Zhang, L. Huang, Z. Wei, and J. Nie, “Scale - semantic joint decoupling network for image -text retrieval in remote sensing,” ACM Trans. Multimedia Comput. Commun. Appl., vol. 20, no. 1, p. Article 4, 2023

    work page 2023

  19. [19]

    Frequency - and spatial-domain saliency network for remote sensing cross -modal retrieval,

    C. Zheng, J. Nie, B. Yin, X. Li, Y. Qian, and Z. Wei, “Frequency - and spatial-domain saliency network for remote sensing cross -modal retrieval,” IEEE Trans. Geosci. Remote Sens., vol. 63, pp. 1-13, 2025

    work page 2025

  20. [20]

    Fame: Fusion of alignment and multiview enhancement for remote sensing image –text retrieval,

    Y. Su, D. Zhu, Z. Dong, Q. Lin, L. Liu, and S. Bao, “Fame: Fusion of alignment and multiview enhancement for remote sensing image –text retrieval,” IEEE Trans. Geosci. Remote Sens., vol. 63, pp. 1-15, 2025

    work page 2025

  21. [21]

    Fine-grained visual-language alignment for remote sensing image–text retrieval,

    S. Li et al., “Fine-grained visual-language alignment for remote sensing image–text retrieval,” IEEE Trans. Geosci. Remote Sens., vol. 63, pp. 1- 14, 2025

    work page 2025

  22. [22]

    Knowledge -aided momentum contrastive learning for remote-sensing image text retrieval,

    Z. Ji, C. Meng, Y. Zhang, Y. Pang, and X. Li, “Knowledge -aided momentum contrastive learning for remote-sensing image text retrieval,” IEEE Trans. Geosci. Remote Sens., vol. 61, pp. 1-13, 2023

    work page 2023

  23. [23]

    Masking-based cross-modal remote sensing image –text retrieval via dynamic contrastive learning,

    Z. Zhao et al., “Masking-based cross-modal remote sensing image –text retrieval via dynamic contrastive learning,” IEEE Trans. Geosci. Remote Sens., vol. 62, pp. 1-15, 2024

    work page 2024

  24. [24]

    Hypersphere-based remote sensing cross -modal text– image retrieval via curriculum learning,

    W. Zhang et al., “Hypersphere-based remote sensing cross -modal text– image retrieval via curriculum learning,” IEEE Trans. Geosci. Remote Sens., vol. 61, pp. 1-15, 2023

    work page 2023

  25. [25]

    Remoteclip: A vision language foundation model for remote sensing,

    F. Liu et al. , “Remoteclip: A vision language foundation model for remote sensing,” IEEE Trans. Geosci. Remote Sens., vol. 62, pp. 1 -16, 2024

    work page 2024

  26. [26]

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

    Z. Zhang, T. Zhao, Y. Guo, and J. Yin, “Rs5m and georsclip: A large - scale vision - language dataset and a large vision -language model for remote sensing,” IEEE Trans. Geosci. Remote Sens., vol. 62, pp. 1 -23, 2024

    work page 2024

  27. [27]

    Cross -modal remote sensing image –text retrieval via context and uncertainty -aware prompt,

    Y. Wang, X. Tang, J. Ma, X. Zhang, F. Liu, and L. Jiao, “Cross -modal remote sensing image –text retrieval via context and uncertainty -aware prompt,” IEEE Trans. Neural Networks Learn. Syst., pp. 1-15, 2024

    work page 2024

  28. [28]

    Strong and weak prompt engineering for remote sensing image-text cross -modal retrieval,

    T. Sun et al., “Strong and weak prompt engineering for remote sensing image-text cross -modal retrieval,” IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens., vol. 18, pp. 6968-6980, 2025

    work page 2025

  29. [29]

    Knowledge-aware text– image retrieval for remote sensing images,

    L. Mi, X. Dai, J. Castillo-Navarro, and D. Tuia, “Knowledge-aware text– image retrieval for remote sensing images,” IEEE Trans. Geosci. Remote Sens., vol. 62, pp. 1-13, 2024

    work page 2024

  30. [30]

    Thread the needle: Cues-driven multiassociation for remote sensing cross -modal retrieval,

    Y. Chen, J. Huang, Z. Sun, S. Xiong, and X. Lu, “Thread the needle: Cues-driven multiassociation for remote sensing cross -modal retrieval,” IEEE Trans. Geosci. Remote Sens., vol. 62, pp. 1-13, 2024

    work page 2024

  31. [31]

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

    X. Li, C. Wen, Y. Hu, Z. Yuan, and X. X. Zhu, “Vision-language models in remote sensing: Current progress and future trends,” IEEE Geosci. Remote Sens. Mag., vol. 12, no. 2, pp. 32-66, 2024

    work page 2024

  32. [32]

    Parameter -efficient transfer learning for remote sensing image –text retrieval,

    Y. Yuan, Y. Zhan, and Z. Xiong, “Parameter -efficient transfer learning for remote sensing image –text retrieval,” IEEE Trans. Geosci. Remote Sens., vol. 61, pp. 1-14, 2023

    work page 2023

  33. [33]

    Direction -oriented visual –semantic embedding model for remote sensing image–text retrieval,

    Q. Ma, J. Pan, and C. Bai, “Direction -oriented visual –semantic embedding model for remote sensing image–text retrieval,” IEEE Trans. Geosci. Remote Sens., vol. 62, pp. 1-14, 2024

    work page 2024

  34. [34]

    Integrating multisubspace joint learning with multilevel guidance for cross -modal retrieval of remote sensing images,

    Y. Chen, J. Huang, S. Xiong, and X. Lu, “Integrating multisubspace joint learning with multilevel guidance for cross -modal retrieval of remote sensing images,” IEEE Trans. Geosci. Remote Sens., vol. 62, pp. 1 -17, 2024

    work page 2024

  35. [35]

    Sirs: Multitask joint learning for remote sensing foreground -entity image –text retrieval,

    Z. Zhu, J. Kang, W. Diao, Y. Feng, J. Li, and J. Ni, “Sirs: Multitask joint learning for remote sensing foreground -entity image –text retrieval,” IEEE Trans. Geosci. Remote Sens., vol. 62, pp. 1-15, 2024

    work page 2024

  36. [36]

    Parameter -efficient reparameterization tuning for remote sensing image –text retrieval,

    J. Yang, S. Li, and M. Zhao, “Parameter -efficient reparameterization tuning for remote sensing image –text retrieval,” IEEE Trans. Geosci. Remote Sens., vol. 63, pp. 1-15, 2025

    work page 2025

  37. [37]

    Theorie der horizontalen sichtweite,

    H. Koschmieder, “Theorie der horizontalen sichtweite,” Beitr. Phys. Freien Atm., vol. 12, pp. 171-181, 1924

    work page 1924

  38. [38]

    Single image haze removal using dark channel prior,

    K. He, J. Sun, and X. Tang, “Single image haze removal using dark channel prior,” IEEE Trans. Pattern Anal. Mach. Intell., vol. 33, no. 12, pp. 2341-2353, 2011

    work page 2011

  39. [39]

    Summary of current radiometric calibration coefficients for landsat mss, tm, etm+, and eo -1 ali sensors,

    G. Chander, B. L. Markham, and D. L. Helder, “Summary of current radiometric calibration coefficients for landsat mss, tm, etm+, and eo -1 ali sensors,” Remote Sens. Environ., vol. 113, no. 5, pp. 893-903, 2009

    work page 2009

  40. [40]

    Dgdnet: Deep gradient descent network for remotely sensed image denoising,

    Z. Huang, Z. Zhu, Z. Wang, Y. Shi, H. Fang, and Y. Zhang, “Dgdnet: Deep gradient descent network for remotely sensed image denoising,” IEEE Geosci. Remote Sens. Lett., vol. 20, pp. 1-5, 2023

    work page 2023

  41. [41]

    Toward optimal destriping of modis data using a unidirectional variational model,

    M. Bouali and S. Ladjal, “Toward optimal destriping of modis data using a unidirectional variational model,” IEEE Trans. Geosci. Remote Sens., vol. 49, no. 8, pp. 2924-2935, 2011

    work page 2011

  42. [42]

    Noise in remote -sensing systems: The effect on classification error,

    D. A. Landgrebe and E. Malaret, “Noise in remote -sensing systems: The effect on classification error,” IEEE Trans. Geosci. Remote Sens., vol. GE-24, no. 2, pp. 294-300, 1986

    work page 1986

  43. [43]

    Robust cross -modal remote sensing image retrieval via maximal correlation augmentation,

    Z. Wang, X. Wang, G. Li, and C. Li, “Robust cross -modal remote sensing image retrieval via maximal correlation augmentation,” IEEE Trans. Geosci. Remote Sens., vol. 62, pp. 1-17, 2024

    work page 2024

  44. [44]

    Rsgpt: A remote sensing vision language model and benchmark,

    Y. Hu, J. Yuan, C. Wen, X. Lu, Y. Liu, and X. Li, “Rsgpt: A remote sensing vision language model and benchmark,” ISPRS J. Photogramm. Remote Sens., vol. 224, pp. 272-286, 2025

    work page 2025

  45. [45]

    Benchmarking robustness of multimodal image -text models under distribution shift,

    J. Qiu et al. , “Benchmarking robustness of multimodal image -text models under distribution shift,” Journal of Data -centric Machine Learning Research, 2024

    work page 2024

  46. [46]

    Text -image matching for cross -modal remote sensing image retrieval via graph neural network,

    H. Yu et al. , “Text -image matching for cross -modal remote sensing image retrieval via graph neural network,” IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens., vol. 16, pp. 812-824, 2023

    work page 2023

  47. [47]

    Hypergraph -enhanced textual -visual matching network for cross-modal remote sensing image retrieval via dynamic hypergraph learning,

    F. Yao et al. , “Hypergraph -enhanced textual -visual matching network for cross-modal remote sensing image retrieval via dynamic hypergraph learning,” IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens., vol. 16, pp. 688-701, 2023

    work page 2023

  48. [48]

    Scene graph -aware hierarchical fusion network for remote sensing image retrieval with text feedback,

    F. Wang, X. Zhu, X. Liu, Y. Zhang, and Y. Li, “Scene graph -aware hierarchical fusion network for remote sensing image retrieval with text feedback,” IEEE Trans. Geosci. Remote Sens., vol. 62, pp. 1-16, 2024. 16 > REPLACE THIS LINE WITH YOUR MANUSCRIPT ID NUMBER (DOUBLE -CLICK HERE TO EDIT) <

    work page 2024

  49. [49]

    R. M. Neal, Bayesian learning for neural networks : Springer Science & Business Media, 2012

    work page 2012

  50. [50]

    Dropout as a bayesian approximation: Representing model uncertainty in deep learning,

    Y. Gal and Z. Ghahramani, “Dropout as a bayesian approximation: Representing model uncertainty in deep learning,” in Proc. Int. Conf. Mach. Learn. (ICML), 2016, pp. 1050-1059

    work page 2016

  51. [51]

    Simple and scalable predictive uncertainty estimation using deep ensembles,

    B. Lakshminarayanan, A. Pritzel, and C. Blundell, “Simple and scalable predictive uncertainty estimation using deep ensembles,” in Proc. Adv. Neural Inf. Process. Syst. (NeurIPS), 2017, pp. 6402-6413

    work page 2017

  52. [52]

    R. R. Yager and L. Liu, Classic works of the dempster -shafer theory of belief functions: Springer, 2008

    work page 2008

  53. [53]

    Jø sang, Subjective logic: Springer, 2016

    A. Jø sang, Subjective logic: Springer, 2016

    work page 2016

  54. [54]

    Evidential deep learning to quantify classification uncertainty,

    M. Sensoy, L. Kaplan, and M. Kandemir, “Evidential deep learning to quantify classification uncertainty,” in Proc. Adv. Neural Inf. Process. Syst. (NeurIPS), 2018, pp. 3179-3189

    work page 2018

  55. [55]

    Trusted multi -view classification with dynamic evidential fusion,

    Z. Han, C. Zhang, H. Fu, and J. T. Zhou, “Trusted multi -view classification with dynamic evidential fusion,” IEEE Trans. Pattern Anal. Mach. Intell., vol. 45, no. 2, pp. 2551-2566, 2023

    work page 2023

  56. [56]

    Evidential multi - source-free unsupervised domain adaptation,

    J. Pei, A. Men, Y. Liu, X. Zhuang, and Q. Chen, “Evidential multi - source-free unsupervised domain adaptation,” IEEE Trans. Pattern Anal. Mach. Intell., vol. 46, no. 8, pp. 5288-5305, 2024

    work page 2024

  57. [57]

    Uncertainty-aware dual-evidential learning for weakly -supervised temporal action localization,

    M. Chen, J. Gao, and C. Xu, “Uncertainty-aware dual-evidential learning for weakly -supervised temporal action localization,” IEEE Trans. Pattern Anal. Mach. Intell., vol. 45, no. 12, pp. 15896-15911, 2023

    work page 2023

  58. [58]

    Deep evidential learning with noisy correspondence for cross -modal retrieval,

    Y. Qin, D. Peng, X. Peng, X. Wang, and P. Hu, “Deep evidential learning with noisy correspondence for cross -modal retrieval,” in Proc. ACM Int. Conf. Multimedia (ACM MM), 2022, pp. 4948-4956

    work page 2022

  59. [59]

    Memo: Test time robustness via adaptation and augmentation,

    M. Zhang, S. Levine, and C. Finn, “Memo: Test time robustness via adaptation and augmentation,” in Proc. Adv. Neural Inf. Process. Syst. (NeurIPS), 2022, pp. 38629-38642

    work page 2022

  60. [60]

    Understanding test -time augmentation,

    M. Kimura, “Understanding test -time augmentation,” in Proceedings of the International Conference on Neural Information Processing , 2021, pp. 558-569

    work page 2021

  61. [61]

    Better aggregation in test -time augmentation,

    D. Shanmugam, D. Blalock, G. Balakrishnan, and J. Guttag, “Better aggregation in test -time augmentation,” in Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit. (CVPR), 2021, pp. 1214-1223

    work page 2021

  62. [62]

    Aleatoric uncertainty estimation with test -time augmentation for medical image segmentation with convolutional neural networks,

    G. Wang, W. Li, M. Aertsen, J. Deprest, S. Ourselin, and T. Vercauteren, “Aleatoric uncertainty estimation with test -time augmentation for medical image segmentation with convolutional neural networks,” Neurocomputing, vol. 338, pp. 34-45, 2019

    work page 2019

  63. [63]

    Improved Text Classification via Test-Time Augmentation

    H. Lu, D. Shanmugam, H. Suresh, and J. Guttag, “Improved text classification via test -time augmentation,” arXiv preprint arXiv:2206.13607, 2022

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  64. [64]

    Eda: Easy data augmentation techniques for boosting performance on text classification tasks,

    J. Wei and K. Zou, “Eda: Easy data augmentation techniques for boosting performance on text classification tasks,” in Proc. Conf. Empir. Methods Natural Lang. Process. Int. Joint Conf. Natural Lang. Process., Hong Kong, China, 2019, pp. 6382-6388

    work page 2019

  65. [65]

    Unsupervised contrastive hashing for cross-modal retrieval in remote sensing,

    G. Mikriukov, M. Ravanbakhsh, and B. Demir, “Unsupervised contrastive hashing for cross-modal retrieval in remote sensing,” in Proc. IEEE Int. Conf. Acoust., Speech, Signal Process (ICASSP) , 2022, pp. 4463-4467

    work page 2022

  66. [66]

    Image -text retrieval with cross-modal semantic importance consistency,

    Z. Liu, F. Chen, J. Xu, W. Pei, and G. Lu, “Image -text retrieval with cross-modal semantic importance consistency,” IEEE Trans. Circuits Syst. Video Technol., vol. 33, no. 5, pp. 2465-2476, 2023

    work page 2023

  67. [67]

    Unicom: Universal and compact representation learning for image retrieval,

    X. An et al. , “Unicom: Universal and compact representation learning for image retrieval,” in Proc. Int. Conf. Learn. Rep. (ICLR), 2023, pp. 1- 16

    work page 2023

  68. [68]

    Towards General Text Embeddings with Multi-stage Contrastive Learning

    Z. Li, X. Zhang, Y. Zhang, D. Long, P. Xie, and M. Zhang, “Towards general text embeddings with multi -stage contrastive learning,” arXiv preprint arXiv:2308.03281, 2023

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  69. [69]

    Latent space semantic supervision based on knowledge distillation for cross -modal retrieval,

    L. Zhang and X. Wu, “Latent space semantic supervision based on knowledge distillation for cross -modal retrieval,” IEEE Trans. Image Process., vol. 31, pp. 7154-7164, 2022

    work page 2022

  70. [70]

    Ckdh: Clip - based knowledge distillation hashing for cross -modal retrieval,

    J. Li, W. K. Wong, L. Jiang, X. Fang, S. Xie, and Y. Xu, “Ckdh: Clip - based knowledge distillation hashing for cross -modal retrieval,” IEEE Trans. Circuits Syst. Video Technol., vol. 34, no. 7, pp. 6530-6541, 2024

    work page 2024

  71. [71]

    Land cover mapping at very high resolution with rotation equivariant cnns: Towards small yet accurate models,

    D. Marcos, M. Volpi, B. Kellenberger, and D. Tuia, “Land cover mapping at very high resolution with rotation equivariant cnns: Towards small yet accurate models,” ISPRS J. Photogramm. Remote Sens., vol. 145, pp. 96-107, 2018

    work page 2018

  72. [72]

    Remote sensing image scene classification: Benchmark and state of the art,

    G. Cheng, J. Han, and X. Lu, “Remote sensing image scene classification: Benchmark and state of the art,” Proc. IEEE, vol. 105, no. 10, pp. 1865-1883, 2017

    work page 2017

  73. [73]

    Exploring models and data for remote sensing image caption generation,

    X. Lu, B. Wang, X. Zheng, and X. Li, “Exploring models and data for remote sensing image caption generation,” IEEE Trans. Geosci. Remote Sens., vol. 56, no. 4, pp. 2183-2195, 2018

    work page 2018

  74. [74]

    Learning transferable visual models from natural language supervision,

    A. Radford et al. , “Learning transferable visual models from natural language supervision,” in Proc. Int. Conf. Mach. Learn. (ICML) , 2021, pp. 8748-8763

    work page 2021

  75. [75]

    Decoupled weight decay regularization,

    I. Loshchilov and F. Hutter, “Decoupled weight decay regularization,” in Proc. Int. Conf. Learn. Rep. (ICLR), 2019, pp. 1-18

    work page 2019

  76. [76]

    VSE++: Improving Visual-Semantic Embeddings with Hard Negatives

    F. Faghri, D. J. Fleet, J. R. Kiros, and S. Fidler, “Vse++: Improving visual-semantic embeddings with hard negatives,” arXiv preprint arXiv:1707.05612, 2017

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  77. [77]

    Universal weighting metric learning for cross -modal retrieval,

    J. Wei, Y. Yang, X. Xu, X. Zhu, and H. T. Shen, “Universal weighting metric learning for cross -modal retrieval,” IEEE Trans. Pattern Anal. Mach. Intell., vol. 44, no. 10, pp. 6534-6545, 2022

    work page 2022

  78. [78]

    Learning the best pooling strategy for visual semantic embedding,

    J. Chen, H. Hu, H. Wu, Y. Jiang, and C. Wang, “Learning the best pooling strategy for visual semantic embedding,” in Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit. (CVPR), 2021, pp. 15784-15793

    work page 2021

  79. [79]

    Representation Learning with Contrastive Predictive Coding

    A. v. d. Oord, Y. Li, and O. Vinyals, “Representation learning with contrastive predictive coding,” arXiv preprint arXiv:1807.03748, 2018

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  80. [80]

    Prototype -based aleatoric uncertainty quantification for cross -modal retrieval,

    H. Li, J. Song, L. Gao, X. Zhu, and H. Shen, “Prototype -based aleatoric uncertainty quantification for cross -modal retrieval,” in Proc. Adv. Neural Inf. Process. Syst. (NeurIPS), 2024, pp. 24564-24585

    work page 2024

Showing first 80 references.