Pith. sign in

REVIEW 3 minor 122 references

Reviewed by Pith at T0; open to challenge.

T0 means a machine referee read the full paper against a public rubric. The mark states how deep the mechanical check went, never who wrote it. the ladder, T0–T4 →

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

T0 review · grok-4.3

Training-free composition methods deliver strong scalable baselines for remote sensing composed image retrieval while change-centric queries require explicit scene identity preservation.

2026-06-30 13:33 UTC pith:HMCDZL3M

load-bearing objection A useful applied benchmark with a new change-centric EO dataset and released code; no load-bearing flaws apparent.

arxiv 2605.24442 v1 pith:HMCDZL3M submitted 2026-05-23 cs.CV

Benchmarking Composed Image Retrieval for Applied Earth Observation

Bill Psomas , Dionysis Christopoulos , Thanasis Petropoulos , Nikos Efthymiadis , Ioannis Kakogeorgiou , Ond\v{r}ej Chum , Yannis Avrithis , Giorgos Tolias
show 1 more author
Konstantinos Karantzalos
This is my paper
classification cs.CV
keywords composed image retrievalremote sensingearth observationbenchmarkchange detectionvision-language modelsdisaster monitoringsatellite imagery
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.

The paper evaluates how well existing composed image retrieval techniques transfer to satellite imagery by testing representative methods across multiple vision-language backbones on the PatternCom dataset under a fixed protocol. It then releases xView2-CIR, a new dataset that frames retrieval around post-disaster state changes while keeping the underlying scene fixed. The central finding is that simple training-free composition approaches already work reliably for Earth observation search tasks, yet the change-focused setting introduces distinct difficulties centered on maintaining scene identity rather than just altering attributes. A sympathetic reader would care because large satellite archives currently lack flexible query interfaces that combine an example image with a textual change description, and this work supplies both the data and the baseline numbers needed to build such interfaces.

Core claim

Training-free composition methods provide strong and scalable baselines for EO retrieval, while change-centric retrieval on xView2-CIR presents different challenges from attribute-based retrieval, particularly due to the need to preserve scene identity.

What carries the argument

Standardized protocol that adapts composed image retrieval methods to remote sensing data using six vision-language backbones on PatternCom for attribute queries and on the new xView2-CIR dataset for change queries.

Load-bearing premise

The chosen vision-language backbones, composition strategies, and the two datasets are representative of how composed retrieval would behave in real Earth observation workflows.

What would settle it

A follow-up experiment on a third, larger, or differently distributed satellite archive in which trained composition methods consistently outperform all training-free baselines by a large margin while change-centric queries show no extra penalty for scene identity.

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

If this is right

  • Training-free methods can serve as immediate practical baselines for archive search and exploration in remote sensing.
  • Change-centric retrieval must treat scene identity preservation as a first-class constraint separate from attribute modification.
  • Composed retrieval can function as a complementary tool alongside existing remote sensing retrieval systems for disaster monitoring and damage assessment.
  • Standardized evaluation protocols enable direct comparison of future methods on both attribute and change query types.

Where Pith is reading between the lines

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

  • Operational systems may need hybrid pipelines that first enforce scene identity before applying textual modifiers.
  • The benchmark could be extended by adding temporal consistency checks across multiple acquisition dates of the same location.
  • Results imply that identity-preserving mechanisms developed for change detection might transfer more readily than generic composition techniques.

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

0 major / 3 minor

Summary. The manuscript benchmarks composed image retrieval (CIR) methods adapted to remote sensing (RSCIR). It evaluates representative training-free composition strategies across six vision-language backbones on the PatternCom dataset under a standardized protocol, analyzing performance by backbone, strategy, and query type. It introduces the xView2-CIR dataset for change-centric retrieval in disaster monitoring, where queries combine a reference image with a textual modifier conditioned on preserving scene identity and reaching a target post-event state. The central empirical claim is that training-free methods yield strong, scalable baselines for EO retrieval while change-centric tasks differ from attribute-based retrieval due to the scene-identity constraint.

Significance. If the reported comparisons hold under the chosen protocol, the work supplies a reproducible benchmark and a new change-centric dataset that positions composed retrieval as a practical complement to existing remote-sensing retrieval tools for archive search and change analysis. Explicit release of the xView2-CIR dataset and evaluation code is a clear strength that supports follow-on research.

minor comments (3)
  1. [Abstract] Abstract: the phrase 'six vision-language backbones' is not expanded; listing the specific models (or citing the section that does) would allow immediate assessment of scope.
  2. [§3] §3 (or wherever the protocol is defined): the description of how negative pairs and scene-identity preservation are operationalized in xView2-CIR could be expanded with one concrete query example to clarify the distinction from PatternCom.
  3. [Tables] Table captions and axis labels: ensure all tables reporting recall or mAP explicitly state the number of queries and the retrieval pool size so that numbers are directly interpretable.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for the positive summary of our work, the recognition of the xView2-CIR dataset release and code as strengths, and the recommendation for minor revision. No major comments were provided in the report, so we have no specific points to address point-by-point. We will proceed with the minor revision to incorporate any editorial suggestions while maintaining the reproducibility focus.

Circularity Check

0 steps flagged

No significant circularity

full rationale

This is an empirical benchmarking study that adapts and evaluates existing composed image retrieval methods across six vision-language backbones on PatternCom and the newly introduced xView2-CIR dataset under a standardized protocol. All central claims (training-free methods as strong baselines, differences between change-centric and attribute-based retrieval) are supported by direct performance measurements on held-out test data. No derivations, fitted parameters presented as predictions, self-citation load-bearing arguments, or equations appear in the provided text. The work is self-contained against external benchmarks and reports no internal reductions of results to inputs by construction.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

This is an empirical benchmarking study; no mathematical derivations, fitted parameters, or postulated entities are introduced.

pith-pipeline@v0.9.1-grok · 5793 in / 1093 out tokens · 42038 ms · 2026-06-30T13:33:39.729682+00:00 · methodology

0 comments
read the original abstract

Remote sensing composed image retrieval (RSCIR) enables search in large satellite image archives using composed queries that combine a reference image with a textual modifier. Although RSCIR offers a flexible interface for expressing targeted retrieval intent, the transferability of modern composition methods to Earth observation (EO) imagery and their relevance to operational EO workflows remain underexplored. We address this gap through a unified benchmark and an application-oriented study. First, we systematically adapt and evaluate representative composed image retrieval methods with six vision-language backbones on PatternCom under a standardized protocol, analyzing their behavior across backbones, composition strategies, and query types. Second, we introduce xView2-CIR, a change-centric dataset for disaster and damage monitoring, where retrieval is conditioned on scene identity and a target post-event state. Our results show that training-free composition methods provide strong and scalable baselines for EO retrieval, while change-centric retrieval presents different challenges from attribute-based retrieval, particularly due to the need to preserve scene identity. Overall, this study establishes a practical benchmark for RSCIR and positions composed retrieval as a complementary tool for remote sensing image retrieval, archive exploration, and change analysis. The dataset and code are available at https://github.com/billpsomas/rscir.

Figures

Figures reproduced from arXiv: 2605.24442 by Bill Psomas, Dionysis Christopoulos, Giorgos Tolias, Ioannis Kakogeorgiou, Konstantinos Karantzalos, Nikos Efthymiadis, Ond\v{r}ej Chum, Thanasis Petropoulos, Yannis Avrithis.

Figure 1
Figure 1. Figure 1: Composed Image Retrieval for applied Earth Observation. We illustrate how composed queries (query image + query text) enable controllable retrieval by specifying a targeted change. (a) Activity monitoring: a parking-lot query image composed with being empty retrieves visually similar parking areas with low vehicle occupancy. (b) Capacity monitoring: A facility query image composed with having four retrieve… view at source ↗
Figure 2
Figure 2. Figure 2: Remote sensing composed image retrieval for disaster monitoring. We restructure xView2 [41] into a composed retrieval setting where a query consists of a pre-event reference image of a specific location and a textual modifier describing the desired post-event state (e.g., post-hurricane). For each disaster type (a–f), we illustrate the query and the corresponding relevance criterion: positives are images o… view at source ↗
Figure 3
Figure 3. Figure 3: Qualitative composed retrieval results on PatternCom with OpenAI CLIP. Comparison between unimodal and multimodal methods. Each query is shown as a reference image combined with a boxed textual modifier. Columns report the top-2 retrieved results. Retrieval is evaluated under the same class + target attribute value relevance criterion. all shown cases, producing results that simultaneously re￾spect the que… view at source ↗
Figure 4
Figure 4. Figure 4: Qualitative composed retrieval results on xView2-CIR with OpenAI CLIP and WeiCom. Each query combines a pre-disaster reference image with a boxed textual modifier (post-*) indicating the target post-event state. We show the top retrieved post-disaster results per disaster type. Retrieval is evaluated under the same scene/location + target state relevance criterion, so visually plausible post-event matches … view at source ↗
Figure 5
Figure 5. Figure 5: Qualitative composed retrieval on LEVIR-CC with OpenAI CLIP and WeiCom. Each query is shown as a reference image combined with a textual change modifier. We report the top retrieved candidates under the same location + target state relevance criterion. 4.4. Sensitivity and Ablation Analysis Impact of 𝜆 in WeiCom [PITH_FULL_IMAGE:figures/full_fig_p009_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Impact of modality control 𝜆 in WeiCom with RemoteCLIP on PatternCom. Curves report attribute-wise mAP (%) as 𝜆 varies. Psomas et al. Page 9 of 19 [PITH_FULL_IMAGE:figures/full_fig_p009_6.png] view at source ↗
Figure 9
Figure 9. Figure 9: Impact of visual expansion on FreeDom with CLIP LAION-2B on PatternCom. The number of proxy images 𝑘 is varied while textual expansion is disabled (𝑚=𝑛=1). 1 5 10 20 50 100 200 500 4 6 8 10 Number of proxy images 𝑘 (with 𝑚=𝑛=1) mAP (%) Hurricane Wildfire Flood Total [PITH_FULL_IMAGE:figures/full_fig_p010_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Impact of visual expansion on FreeDom with OpenAI CLIP on xView2-CIR. The number of proxy images 𝑘 is varied while textual expansion is disabled (𝑚=𝑛=1). Ablation of BASIC components. We further analyze the contribution of BASIC components, with full results reported in A.5. On PatternCom, the largest drops occur when removing centering or semantic projection, indicating that modality calibration and sema… view at source ↗
Figure 8
Figure 8. Figure 8: Impact of textual expansion on FreeDom with CLIP LAION-2B on PatternCom. Visual expansion is disabled (𝑘=1), and the number of aggregated textual concepts is varied as 𝑚=𝑛. Impact of visual expansion on FreeDom [PITH_FULL_IMAGE:figures/full_fig_p010_8.png] view at source ↗
Figure 12
Figure 12. Figure 12: CompoDiff hyperparameter sweep with SkyCLIP￾50 on PatternCom. We report average mAP as a function of source weight for different numbers of diffusion timesteps. SEARLE hyperparameter sensitivity [PITH_FULL_IMAGE:figures/full_fig_p018_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: SEARLE hyperparameter sweep with CLIP LAION￾2B on PatternCom. We report average mAP as a function of learning rate, number of inversion steps, and GPT mixing coefficient 𝜆GPT. Ablation of BASIC components [PITH_FULL_IMAGE:figures/full_fig_p018_13.png] view at source ↗

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

122 extracted references · 122 canonical work pages · 2 internal anchors

  1. [1]

    An environment for content-basedimageretrievalfromlargespatialdatabases

    Agouris, P., Carswell, J., Stefanidis, A., 1999. An environment for content-basedimageretrievalfromlargespatialdatabases. ISPRSJ. Photogramm. Remote Sens

    work page 1999

  2. [2]

    Zero-shotcomposedimageretrievalwithtextualinversion,in:Proc

    Baldrati, A., Agnolucci, L., Bertini, M., Del Bimbo, A., 2023. Zero-shotcomposedimageretrievalwithtextualinversion,in:Proc. IEEE/CVF Int. Conf. Comput. Vis., pp. 15338–15347

    work page 2023

  3. [3]

    Effec- tiveconditionedandcomposedimageretrievalcombiningclip-based features, in: Proc

    Baldrati, A., Bertini, M., Uricchio, T., Del Bimbo, A., 2022. Effec- tiveconditionedandcomposedimageretrievalcombiningclip-based features, in: Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit

    work page 2022

  4. [4]

    Automaticattributediscovery and characterization from noisy web data, in: Proc

    Berg,T.L.,Berg,A.C.,Shih,J.,2010. Automaticattributediscovery and characterization from noisy web data, in: Proc. Eur. Conf. Comput. Vis., Springer

    work page 2010

  5. [5]

    Modeling and detection of geospatialobjectsusingtexturemotifs

    Bhagavathy, S., Manjunath, B.S., 2006. Modeling and detection of geospatialobjectsusingtexturemotifs. IEEETrans.Geosci.Remote Sens. 44, 3706–3715

    work page 2006

  6. [6]

    Enhancedinteractiveremotesensing image retrieval with scene classification convolutional neural net- works model, in: Proc

    Boualleg,Y.,Farah,M.,2018. Enhancedinteractiveremotesensing image retrieval with scene classification convolutional neural net- works model, in: Proc. IEEE Int. Geosci. Remote Sens. Symp., pp. 4748–4751

    work page 2018

  7. [7]

    Brown, T., Mann, B., Ryder, N., Subbiah, M., Kaplan, J.D., Dhari- wal, P., Neelakantan, A., Shyam, P., Sastry, G., Askell, A., et al.,

    work page

  8. [8]

    Language models are few-shot learners. Adv. Neural Inf. Process. Syst. 33, 1877–1901

    work page 1901

  9. [9]

    Enhancingremotesensingimageretrievalusingatripletdeepmetric learning network

    Cao, R., Zhang, Q., Zhu, J., Li, Q., Li, Q., Liu, B., Qiu, G., 2020. Enhancingremotesensingimageretrievalusingatripletdeepmetric learning network. Int. J. Remote Sens. 41, 740–751

    work page 2020

  10. [10]

    Mul- tilabelremotesensingimageretrievalusingasemisupervisedgraph- theoretic method

    Chaudhuri, B., Demir, B., Chaudhuri, S., Bruzzone, L., 2017. Mul- tilabelremotesensingimageretrievalusingasemisupervisedgraph- theoretic method. IEEE Trans. Geosci. Remote Sens. 56, 1144– 1158

    work page 2017

  11. [11]

    Siamese graphconvolutionalnetworkforcontentbasedremotesensingimage retrieval

    Chaudhuri, U., Banerjee, B., Bhattacharya, A., 2019. Siamese graphconvolutionalnetworkforcontentbasedremotesensingimage retrieval. Comput. Vis. Image Underst

    work page 2019

  12. [12]

    Attention-driven graph convolution network for remote sensing im- age retrieval

    Chaudhuri, U., Banerjee, B., Bhattacharya, A., Datcu, M., 2021. Attention-driven graph convolution network for remote sensing im- age retrieval. IEEE Geosci. Remote Sens. Lett. 19, 1–5

    work page 2021

  13. [13]

    Learning joint visual semantic match- ing embeddings for language-guided retrieval, in: Proc

    Chen, Y., Bazzani, L., 2020. Learning joint visual semantic match- ing embeddings for language-guided retrieval, in: Proc. Eur. Conf. Comput. Vis

    work page 2020

  14. [14]

    Image search with text feedback by visiolinguistic attention learning, in: Proc

    Chen, Y., Gong, S., Bazzani, L., 2020. Image search with text feedback by visiolinguistic attention learning, in: Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit

    work page 2020

  15. [15]

    A novel ensemble architecture of residual attention-based deep metric learningforremotesensingimageretrieval

    Cheng, Q., Gan, D., Fu, P., Huang, H., Zhou, Y., 2021a. A novel ensemble architecture of residual attention-based deep metric learningforremotesensingimageretrieval. RemoteSens.13,3445

    work page

  16. [16]

    A semantic-preserving deep hashing model for multi-label remote sensing image retrieval

    Cheng, Q., Huang, H., Ye, L., Fu, P., Gan, D., Zhou, Y., 2021b. A semantic-preserving deep hashing model for multi-label remote sensing image retrieval. Remote Sens. 13, 4965

    work page

  17. [17]

    Openclip: An open source implementation of clip

    Cherti, M., Beaumont, R., Wightman, R., Zhai, X., Beyer, L., Kolesnikov, A., Dosovitskiy, A., Houlsby, N., Minderer, M., 2022. Openclip: An open source implementation of clip

    work page 2022

  18. [18]

    IEEE Int

    Dai,O.E.,Demir,B.,Sankur,B.,Bruzzone,L.,2017.Anovelsystem for content based retrieval of multi-label remote sensing images, in: Proc. IEEE Int. Geosci. Remote Sens. Symp., pp. 1744–1747

    work page 2017

  19. [19]

    Delmas,G.,deRezende,R.S.,Csurka,G.,Larlus,D.,2022.Artemis: Attention-based retrieval with text-explicit matching and implicit similarity

    work page 2022

  20. [20]

    Composed image retrieval for Psomas et al

    Efthymiadis, N., Psomas, B., Laskar, Z., Karantzalos, K., Avrithis, Y., Chum, O., Tolias, G., 2025. Composed image retrieval for Psomas et al. Page 12 of 19 Benchmarking Composed Image Retrieval for Applied Earth Observation training-free domain conversion, in: Proc. IEEE/CVF Winter Conf. Appl. Comput. Vis

    work page 2025

  21. [21]

    Global optimization: Combining locallosswithresultrankinglossinremotesensingimageretrieval

    Fan, L., Zhao, H., Zhao, H., 2020. Global optimization: Combining locallosswithresultrankinglossinremotesensingimageretrieval. IEEE Trans. Geosci. Remote Sens. 59, 7011–7026

    work page 2020

  22. [22]

    Devise: A deep visual-semantic embedding model

    Frome, A., Corrado, G.S., Shlens, J., Bengio, S., Dean, J., Ranzato, M., Mikolov, T., 2013. Devise: A deep visual-semantic embedding model. Adv. Neural Inf. Process. Syst. 26

    work page 2013

  23. [23]

    Exploiting representations from pre-trained convolutional neural networks for high-resolution remote sensing image retrieval

    Ge, Y., Jiang, S., Xu, Q., Jiang, C., Ye, F., 2018. Exploiting representations from pre-trained convolutional neural networks for high-resolution remote sensing image retrieval. Multimed. Tools Appl. 77, 17489–17515

    work page 2018

  24. [24]

    Deep image retrieval:Learningglobalrepresentationsforimagesearch,in:Proc

    Gordo, A., Almazan, J., Revaud, J., Larlus, D., 2016. Deep image retrieval:Learningglobalrepresentationsforimagesearch,in:Proc. Eur. Conf. Comput. Vis

    work page 2016

  25. [25]

    Google earth engine: Planetary-scale geospatial analysis for everyone

    Gorelick, N., Hancher, M., Dixon, M., Ilyushchenko, S., Thau, D., Moore, R., 2017. Google earth engine: Planetary-scale geospatial analysis for everyone. Remote Sens. Environ. 202, 18–27

    work page 2017

  26. [26]

    Compodiff: Versatile composed image retrieval with latent diffusion

    Gu,G.,Chun,S.,Kim,W.,Jun,H.,Kang,Y.,Yun,S.,a. Compodiff: Versatile composed image retrieval with latent diffusion. Transac- tions on Machine Learning Research

    work page

  27. [27]

    Open-vocabulary object detection via vision and language knowledge distillation

    Gu, X., Lin, T.Y., Kuo, W., Cui, Y., b. Open-vocabulary object detection via vision and language knowledge distillation

    work page

  28. [28]

    Openstreetmap: User-generated street maps

    Haklay, M., Weber, P., 2008. Openstreetmap: User-generated street maps. IEEE Pervasive Comput. 7, 12–18

    work page 2008

  29. [29]

    Automatic spatially-aware fashion concept discovery, in: Proc

    Han, X., Wu, Z., Huang, P.X., Zhang, X., Zhu, M., Li, Y., Zhao, Y., Davis, L.S., 2017. Automatic spatially-aware fashion concept discovery, in: Proc. IEEE/CVF Int. Conf. Comput. Vis

    work page 2017

  30. [30]

    Composedqueryimageretrieval using locally bounded features, in: Proc

    Hosseinzadeh,M.,Wang,Y.,2020. Composedqueryimageretrieval using locally bounded features, in: Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit

    work page 2020

  31. [31]

    Exploiting low dimensional features from the mobilenets for remote sensing image retrieval

    Hou, D., Miao, Z., Xing, H., Wu, H., 2020. Exploiting low dimensional features from the mobilenets for remote sensing image retrieval. Earth Sci. Inform. 13, 1437–1443

    work page 2020

  32. [32]

    Delving into deep representations for remote sensing image retrieval, in: Proc

    Hu, F., Tong, X., Xia, G.S., Zhang, L., 2016. Delving into deep representations for remote sensing image retrieval, in: Proc. IEEE Int. Conf. Signal Process., pp. 198–203

    work page 2016

  33. [33]

    Cvm-net: Cross-viewmatchingnetworkforimage-basedground-to-aerialgeo- localization, in: Proc

    Hu, S., Feng, M., Nguyen, R.M., Lee, G.H., 2018. Cvm-net: Cross-viewmatchingnetworkforimage-basedground-to-aerialgeo- localization, in: Proc. IEEE Conf. Comput. Vis. Pattern Recognit., pp. 7258–7267

    work page 2018

  34. [34]

    Aggregated deep local features for remote sensing image retrieval

    Imbriaco, R., Sebastian, C., Bondarev, E., de With, P.H., 2019. Aggregated deep local features for remote sensing image retrieval. Remote Sens. 11, 493

    work page 2019

  35. [35]

    Toward multilabel image retrieval for remote sensing

    Imbriaco, R., Sebastian, C., Bondarev, E., de With, P.H., 2021. Toward multilabel image retrieval for remote sensing. IEEE Trans. Geosci. Remote Sens. 60, 1–14

    work page 2021

  36. [36]

    Discovering states and transformations in image collections, in: Proc

    Isola, P., Lim, J.J., Adelson, E.H., 2015. Discovering states and transformations in image collections, in: Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit., pp. 1383–1391

    work page 2015

  37. [37]

    Scaling up visual and vision-languagerepresentationlearningwithnoisytextsupervision, in: Proc

    Jia, C., Yang, Y., Xia, Y., Chen, Y.T., Parekh, Z., Pham, H., Le, Q., Sung, Y.H., Li, Z., Duerig, T., 2021. Scaling up visual and vision-languagerepresentationlearningwithnoisytextsupervision, in: Proc. Int. Conf. Mach. Learn

    work page 2021

  38. [38]

    Kang, J., Fernandez-Beltran, R., Hong, D., Chanussot, J., Plaza, A.,

    work page

  39. [39]

    IEEE Trans

    Graph relation network: Modeling relations between scenes for multilabel remote-sensing image classification and retrieval. IEEE Trans. Geosci. Remote Sens. 59, 4355–4369

    work page

  40. [40]

    Vision-by- language for training-free compositional image retrieval, in: Int

    Karthik, S., Roth, K., Mancini, M., Akata, Z., 2024. Vision-by- language for training-free compositional image retrieval, in: Int. Conf. Learn. Represent., pp. 16926–16941

    work page 2024

  41. [41]

    Cross-viewim- ageretrieval-groundtoaerialimageretrievalthroughdeeplearning, in: Proc

    Khurshid,N.,Hanif,T.,Tharani,M.,Taj,M.,2019. Cross-viewim- ageretrieval-groundtoaerialimageretrievalthroughdeeplearning, in: Proc. Int. Conf. Neural Inf. Process., Springer. pp. 210–221

    work page 2019

  42. [42]

    Kuznetsova, A., Rom, H., Alldrin, N., Uijlings, J., Krasin, I., Pont- Tuset, J., Kamali, S., Popov, S., Malloci, M., Kolesnikov, A., et al.,

    work page

  43. [43]

    The open images dataset v4: Unified image classification, object detection, and visual relationship detection at scale. Int. J. Comput. Vis

    work page

  44. [44]

    xView: Objects in Context in Overhead Imagery

    Lam, D., Kuzma, R., McGee, K., Dooley, S., Laielli, M., Klaric, M., Bulatov, Y., McCord, B., 2018. xview: Objects in context in overhead imagery. arXiv preprint arXiv:1802.07856

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  45. [45]

    Cosmo: Content-style modulation for image retrieval with text feedback, in: Proc

    Lee, S., Kim, D., Han, B., 2021. Cosmo: Content-style modulation for image retrieval with text feedback, in: Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit

    work page 2021

  46. [46]

    Blip-2: Bootstrapping language-image pre-training with frozen image encoders and large language models, in: Proc

    Li, J., Li, D., Savarese, S., Hoi, S., 2023. Blip-2: Bootstrapping language-image pre-training with frozen image encoders and large language models, in: Proc. Int. Conf. Mach. Learn., pp. 19730– 19742

    work page 2023

  47. [47]

    Li,J.,Li,D.,Xiong,C.,Hoi,S.,2022.Blip:Bootstrappinglanguage- image pre-training for unified vision-language understanding and generation, in: Proc. Int. Conf. Mach. Learn

    work page 2022

  48. [48]

    Integrated spectral and spatial informationmininginremotesensingimagery

    Li, J., Narayanan, R.M., 2004. Integrated spectral and spatial informationmininginremotesensingimagery. IEEETrans.Geosci. Remote Sens. 42, 673–685

    work page 2004

  49. [49]

    IEEE Trans

    Li,Y.,Zhang,Y.,Huang,X.,Ma,J.,2018.Learningsource-invariant deephashingconvolutionalneuralnetworksforcross-sourceremote sensing image retrieval. IEEE Trans. Geosci. Remote Sens. 56, 6521–6536

    work page 2018

  50. [50]

    Content-based high- resolution remote sensing image retrieval via unsupervised feature learning and collaborative affinity metric fusion

    Li, Y., Zhang, Y., Tao, C., Zhu, H., 2016. Content-based high- resolution remote sensing image retrieval via unsupervised feature learning and collaborative affinity metric fusion. Remote Sens. 8, 709

    work page 2016

  51. [51]

    Learning deep representations for ground-to-aerial geolocalization, in: Proc

    Lin, T.Y., Cui, Y., Belongie, S., Hays, J., 2015. Learning deep representations for ground-to-aerial geolocalization, in: Proc. IEEE Conf. Comput. Vis. Pattern Recognit., pp. 5007–5015

    work page 2015

  52. [52]

    Microsoftcoco:Commonobjects in context, in: Proc

    Lin, T.Y., Maire, M., Belongie, S., Hays, J., Perona, P., Ramanan, D.,Dollár,P.,Zitnick,C.L.,2014. Microsoftcoco:Commonobjects in context, in: Proc. Eur. Conf. Comput. Vis

    work page 2014

  53. [53]

    Remote sensing image change captioning with dual-branch transformers: A new methodandalargescaledataset

    Liu, C., Zhao, R., Chen, H., Zou, Z., Shi, Z., 2022. Remote sensing image change captioning with dual-branch transformers: A new methodandalargescaledataset. IEEETrans.Geosci.RemoteSens. 60, 1–20

    work page 2022

  54. [54]

    Liu,F.,Chen,D.,Guan,Z.,Zhou,X.,Zhu,J.,Ye,Q.,Fu,L.,Zhou,J.,

    work page

  55. [55]

    IEEE Trans

    Remoteclip: A vision language foundation model for remote sensing. IEEE Trans. Geosci. Remote Sens. 62, 1–16

    work page

  56. [56]

    Similarity-based unsu- pervised deep transfer learning for remote sensing image retrieval

    Liu, Y., Ding, L., Chen, C., Liu, Y., 2020a. Similarity-based unsu- pervised deep transfer learning for remote sensing image retrieval. IEEE Trans. Geosci. Remote Sens. 58, 7872–7889

    work page

  57. [57]

    Eagle-eyed multitask cnns for aerial image retrieval and scene classification

    Liu, Y., Han, Z., Chen, C., Ding, L., Liu, Y., 2020b. Eagle-eyed multitask cnns for aerial image retrieval and scene classification. IEEE Trans. Geosci. Remote Sens. 58, 6699–6721

    work page

  58. [58]

    Remote-sensing image retrieval with tree-triplet-classification networks

    Liu, Y., Liu, Y., Chen, C., Ding, L., 2020c. Remote-sensing image retrieval with tree-triplet-classification networks. Neurocomputing 405, 48–61

    work page

  59. [59]

    Fusion-based correla- tionlearningmodelforcross-modalremotesensingimageretrieval

    Lv, Y., Xiong, W., Zhang, X., Cui, Y., 2021. Fusion-based correla- tionlearningmodelforcross-modalremotesensingimageretrieval. IEEE Geosci. Remote Sens. Lett. 19, 1–5

    work page 2021

  60. [60]

    An improved svm modelforrelevancefeedbackinremotesensingimageretrieval

    Ma, C., Dai, Q., Liu, J., Liu, S., Yang, J., 2014. An improved svm modelforrelevancefeedbackinremotesensingimageretrieval. Int. J. Digit. Earth 7, 725–745

    work page 2014

  61. [61]

    Cross- source image retrieval based on ensemble learning and knowledge distillation for remote sensing images, in: Proc

    Ma, J., Shi, D., Tang, X., Zhang, X., Han, X., Jiao, L., 2021. Cross- source image retrieval based on ensemble learning and knowledge distillation for remote sensing images, in: Proc. IEEE Int. Geosci. Remote Sens. Symp., IEEE. pp. 2803–2806

    work page 2021

  62. [62]

    Content based image retrieval of satellite imageries using soft query based color composite tech- niques

    Mamatha, Y., Ananth, A., 2010. Content based image retrieval of satellite imageries using soft query based color composite tech- niques. Int. J. Comput. Appl. 7, 0975–8887

    work page 2010

  63. [63]

    Visual descriptors for content-based retrieval of remote-sensing images

    Napoletano, P., 2018. Visual descriptors for content-based retrieval of remote-sensing images. Int. J. Remote Sens. 39, 1343–1376

    work page 2018

  64. [64]

    Probabilistic compositional embeddings for multimodal image retrieval, in: Proc

    Neculai, A., Chen, Y., Akata, Z., 2022. Probabilistic compositional embeddings for multimodal image retrieval, in: Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit

    work page 2022

  65. [65]

    Large- scale image retrieval with attentive deep local features, in: Proc

    Noh, H., Araujo, A., Sim, J., Weyand, T., Han, B., 2017. Large- scale image retrieval with attentive deep local features, in: Proc. IEEE/CVF Int. Conf. Comput. Vis. Psomas et al. Page 13 of 19 Benchmarking Composed Image Retrieval for Applied Earth Observation

    work page 2017

  66. [66]

    Gpt-4o system card

    OpenAI, 2024. Gpt-4o system card

    work page 2024

  67. [67]

    Piedra-Fernandez, J.A., Ortega, G., Wang, J.Z., Canton-Garbin, M.,

    work page

  68. [68]

    IEEE Trans

    Fuzzy content-based image retrieval for oceanic remote sensing. IEEE Trans. Geosci. Remote Sens. 52, 5422–5431

    work page

  69. [69]

    Composed image retrieval forremotesensing,in:Proc.IEEEInt.Geosci.RemoteSens.Symp

    Psomas, B., Kakogeorgiou, I., Efthymiadis, N., Tolias, G., Chum, O., Avrithis, Y., Karantzalos, K., 2024. Composed image retrieval forremotesensing,in:Proc.IEEEInt.Geosci.RemoteSens.Symp

    work page 2024

  70. [70]

    Instance-level composed image retrieval, in: Adv

    Psomas, B., Retsinas, G., Efthymiadis, N., Filntisis, P., Avrithis, Y., Maragos, P., Chum, O., Tolias, G., 2025. Instance-level composed image retrieval, in: Adv. Neural Inf. Process. Syst

    work page 2025

  71. [71]

    Fine-tuning cnn image retrieval with no human annotation

    Radenović, F., Tolias, G., Chum, O., 2019. Fine-tuning cnn image retrieval with no human annotation. IEEE Trans. Pattern Anal. Mach. Intell

    work page 2019

  72. [72]

    Radford,A.,Kim,J.W.,Hallacy,C.,Ramesh,A.,Goh,G.,Agarwal, S.,Sastry,G.,Askell,A.,Mishkin,P.,Clark,J.,etal.,2021.Learning transferable visual models from natural language supervision, in: Proc. Int. Conf. Mach. Learn

    work page 2021

  73. [73]

    Faster r-cnn: Towards real-time object detection with region proposal networks

    Ren, S., He, K., Girshick, R., Sun, J., 2015. Faster r-cnn: Towards real-time object detection with region proposal networks. Adv. Neural Inf. Process. Syst

    work page 2015

  74. [74]

    High-resolution image synthesis with latent diffusion models, in: Proc

    Rombach,R.,Blattmann,A.,Lorenz,D.,Esser,P.,Ommer,B.,2022. High-resolution image synthesis with latent diffusion models, in: Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit

    work page 2022

  75. [75]

    Scalable database indexing and fast image retrieval based on deep learning and hierarchically nested structure applied to remote sens- ing and plant biology

    Sadeghi-Tehran,P.,Angelov,P.,Virlet,N.,Hawkesford,M.J.,2019. Scalable database indexing and fast image retrieval based on deep learning and hierarchically nested structure applied to remote sens- ing and plant biology. J. Imaging 5, 33

    work page 2019

  76. [76]

    Pic2word:Mappingpicturestowordsforzero-shot composed image retrieval, in: Proc

    Saito, K., Sohn, K., Zhang, X., Li, C.L., Lee, C.Y., Saenko, K., Pfister,T.,2023. Pic2word:Mappingpicturestowordsforzero-shot composed image retrieval, in: Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit

    work page 2023

  77. [77]

    Adversarialrepresen- tation learning for text-to-image matching, in: Proc

    Sarafianos,N.,Xu,X.,Kakadiaris,I.A.,2019. Adversarialrepresen- tation learning for text-to-image matching, in: Proc. IEEE/CVF Int. Conf. Comput. Vis., pp. 5814–5824

    work page 2019

  78. [78]

    Laion-5b: An open large-scale dataset for training next generation image-text models

    Schuhmann, C., Beaumont, R., Vencu, R., Gordon, C., Wightman, R., Cherti, M., Coombes, T., Katta, A., Mullis, C., Wortsman, M., et al., 2022. Laion-5b: An open large-scale dataset for training next generation image-text models. Adv. Neural Inf. Process. Syst

    work page 2022

  79. [79]

    Shao,Z.,Zhou,W.,Cheng,Q.,2014.Remotesensingimageretrieval with combined features of salient region. Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci. 40, 83–88

    work page 2014

  80. [80]

    Mul- tilabel remote sensing image retrieval based on fully convolutional network

    Shao, Z., Zhou, W., Deng, X., Zhang, M., Cheng, Q., 2020. Mul- tilabel remote sensing image retrieval based on fully convolutional network. IEEEJ.Sel.Top.Appl.EarthObs.RemoteSens.13,318– 328

    work page 2020

Showing first 80 references.