Pith. sign in

REVIEW 4 major objections 4 minor 1 cited by

DynamicEarth: How Far are We from Open-Vocabulary Change Detection?

T0 review · 4 major / 4 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read Two training-free frameworks, M-C-I and I-M-C, compose off-the-shelf foundation models to detect and name changes in any category a user specifies, and the paper reports they beat supervised and unsupervised baselines on cross-dataset…

desk verdict Useful task framing and a clean codebase, but the headline superiority claim rests on test-set tuning and best-per-dataset selection. read the letter →

arxiv 2501.12931 v1 pith:T4ONG5C4 submitted 2025-01-22 cs.CV

Kaiyu Li , Xiangyong Cao , Yupeng Deng , Chao Pang , Zepeng Xin , Deyu Meng , Zhi Wang This is my paper · ORCID
classification cs.CV
keywords open-vocabularychangedetectionremotesensingfoundationmodelstraining-freeM-C-IframeworkI-M-CSAMCLIP
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

Open-vocabulary change detection (OVCD) is introduced as a task: given two images of the same scene taken at different times, locate and label changes in any category the user names in text. The paper argues that this task is already reachable without any training by composing off-the-shelf foundation models, and proposes two frameworks to do it. M-C-I proposes all possible masks, compares each mask region across dates, then classifies the changed regions with a vision-language model; I-M-C first identifies all objects of interest by name, converts them to masks, then checks whether each object changed. On five building and land-cover change datasets, the resulting methods beat existing unsupervised and supervised baselines in cross-dataset generalization, and the paper releases a codebase for further work. If the claim holds, change detection becomes an open-ended query system rather than a fixed-class classifier.

What carries the argument

The load-bearing machinery is the M-C-I / I-M-C pipeline decomposition, in which Mask Proposal, Comparator, and Identifier (or Identifier, Mask Proposal, Comparator) are chained. The comparator carries the core decision, using negative cosine similarity of DINO/DINOv2 masked features plus an IoU check in I-M-C; the identifier is a CLIP-based open-vocabulary classifier or a grounding detector; the mask proposal uses SAM/SAM2. These components are deliberately training-free, so the operative assumption is that strong off-the-shelf generalization in each component transfers to change detection without fine-tuning.

What would settle it

Take a bi-temporal image pair with no actual change but with strong seasonal color shifts, such as a field turning from green to brown. If non-maximum suppression (NMS) removes masks whose IoU across dates is low and the comparator's cosine similarity then marks those unchanged regions as changed, the framework will report many false positives; conversely, if a genuinely changed object appears very differently so its two masks barely overlap, NMS will delete it and the change will be missed. Measuring recall on a dataset of such seasonal pairs would settle whether the cross-date mask-alignment premise holds.

Watch

Extended reading notes

Core claim

The central discovery is a composition result: the components needed for OVCD already exist, and two orderings of those components cover the task. In M-C-I, a class-agnostic mask proposer (SAM or SAM2) generates candidate regions, a DINO/DINOv2 feature comparator scores each region by negative cosine similarity between the two dates, and an open-vocabulary segmenter (SegEarth-OV on CLIP features) assigns the changed regions to any requested category. In I-M-C, an open-vocabulary identifier (Grounding DINO, APE, or Molmo) first finds the named targets, a mask proposer (SAM/SAM2) refines them, and an IoU-based comparator with a latent-matching check decides change. The paper's experiments show that the M-C-I variants are more stable on complex scenes while I-M-C variants are more precise on simple scenes, and that the best OVCD method outperforms the best cross-dataset supervised transfer on all four building-change datasets.

Load-bearing premise

In the M-C-I framework, the same object is assumed to produce essentially the same proposed mask in both images of a pair, so that duplicate masks can be safely removed before comparison; if seasonal, lighting, or viewpoint changes make the two masks differ a lot, the change itself may be deleted.

Editorial extensions

If this is right

  • Users can query change detection for arbitrary categories, such as building, water, or playground, without retraining on those categories.
  • OVCD methods generalize across sensors and datasets far better than supervised models trained on a single dataset.
  • The M-C-I framework is more robust in complex scenes, while I-M-C yields high precision but depends heavily on the identifier's recall.
  • A universal codebase lets new identifiers, comparators, and mask proposers be swapped in, lowering the barrier for further OVCD research.
  • Fine-grained category division can degrade vision-language-model performance on remote sensing images, so prompt design is part of the method's practical cost.

Reading between the lines

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

  • The NMS deduplication step in M-C-I implicitly assumes that unchanged objects produce similar masks in both dates; under strong seasonal or lighting shifts, comparing date-specific masks separately or using bidirectional matching would be a natural stress test.
  • The two frameworks could be combined so that M-C-I proposes candidate changes and I-M-C verifies them by name, potentially reducing both false positives and missed changes.
  • If CLIP-style classifiers are replaced with stronger open-vocabulary segmenters, I-M-C might close the remaining gap to supervised methods on hard categories such as low vegetation.
  • A useful benchmark protocol for OVCD would report per-category IoU with user-specified text prompts rather than only a binary change mask, which would expose where language grounding fails.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 4 minor

Summary. The paper introduces open-vocabulary change detection (OVCD) as a task in which bi-temporal remote sensing images are compared under arbitrary user-specified textual categories. It proposes two training-free frameworks, M-C-I (mask proposal, comparator, identifier) and I-M-C (identifier, mask proposal, comparator), assembled from off-the-shelf foundation models such as SAM/SAM2, DINO/DINOv2, CLIP/SegEarth-OV, Grounding DINO, APE, and Molmo. The authors instantiate several concrete methods from these frameworks and report experiments on LEVIR-CD, WHU-CD, S2Looking, BANDON, and SECOND, comparing against unsupervised baselines and, in a cross-dataset evaluation, against supervised Changer models. They also release the DynamicEarth codebase. The central quantitative claim is that these OVCD methods significantly surpass existing supervised and unsupervised methods in generalization.

Significance. If the empirical claims were established, the paper would be a useful demonstration that arbitrary-category change detection is feasible today without training, and the released codebase would provide a valuable modular baseline for a new task. The paper has clear strengths: the decomposition into identifier, comparator, and mask proposal is intuitive; the frameworks are genuinely training-free; and the evaluation uses external benchmarks and off-the-shelf components, so there is no circular derivation. The main weakness is that the evaluation protocol does not support the headline superiority claim as written: the comparator threshold is tuned on the test sets, Table 3 selects a different best instantiation per dataset, and no uncertainty or significance information is reported. As a task proposal and codebase contribution the paper is promising, but as a state-of-the-art empirical claim it is not yet convincing.

major comments (4)
  1. [§5.3, Table 3] The 'OVCD' row of Table 3 is not a single reproducible method: for LEVIR-CD it corresponds to APE-/DINO, for WHU-CD to APE-/DINOv2, for S2Looking to SAM-DINOv2-SegEarth-OV, and for BANDON to SAM-DINOv2-SegEarth-OV, i.e., the best instantiation per dataset from Table 1 is selected. The reported Δ values therefore compare a per-dataset best-of-four against each supervised model, and they do not establish that any single proposed OVCD system has superior cross-dataset generalization. This is a selection artifact and should be removed or replaced with a fixed model-selection rule applied uniformly across datasets.
  2. [§5.2, Eq. (1); §5.3, 'Design of prompts'] The comparator threshold β is described as fluctuating across methods or categories, and the authors state that they endeavor to find its optimal value; no held-out validation split, fixed default, or sensitivity analysis is reported. Because change masks are obtained by thresholding the scores defined in Eq. (1), all numbers in Tables 1-3 depend on test-set-tuned hyperparameters. The same concern applies to the prompt templates and synonym lists, which were manually adjusted per category in Section 5.3. Without a validation-based tuning protocol, the abstract's claim of superior performance is not supportable.
  3. [Abstract; §5.3; §6] The abstract states that OVCD methods 'significantly surpass' supervised methods, while the conclusion concedes that OVCD still falls short of purely supervised methods. The Table 3 comparison is against supervised models evaluated on target datasets they were not trained on, with the same-dataset gray cells excluded from the comparison. The text should explicitly distinguish 'supervised models without target-domain retraining' from 'purely supervised methods'; as written, the abstract is misleading. In addition, the word 'significantly' is used without statistical tests, error bars, or multiple runs.
  4. [§4.1, Mask Proposal] The M-C-I pipeline concatenates masks from both dates and then applies non-maximum suppression to remove duplicate masks before comparison. This assumes that unchanged objects produce near-identical masks in the two temporal images. If the same ground object segments differently under seasonal, lighting, or viewpoint changes, NMS can suppress the mask that corresponds to the actual changed region, and the downstream comparator will never see it. This cross-date mask-alignment premise is load-bearing for M-C-I, yet the paper provides no ablation, no analysis of NMS failure cases, and no study of the overlap-threshold sensitivity.
minor comments (4)
  1. [Table 1] The row 'Grounding DINO 1.5 - SAM2 - DINOv2' contains only '-' entries; the text should state explicitly whether this instantiation was not evaluated or failed to produce usable results.
  2. [Table 2] Several I-M-C entries are listed as '-' for categories such as Low vegetation and N.v.g surface; if this means the score is close to zero, the caption should say so explicitly rather than relying on the general table footnote.
  3. [§5.2] The sentence 'Since both proposed frameworks are training-free, we mainly focus on their test/validation sets' blurs the distinction between test and validation. The paper should specify exactly which split is used for hyperparameter tuning and which for final evaluation.
  4. [Figure 3] The text refers to 'compact targets' in the visualization discussion, but this term is not defined; it should be clarified or replaced with a more precise description such as small or low-contrast objects.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: the OVCD frameworks are external compositions of off-the-shelf models evaluated on independent benchmarks; self-citations to SegEarth-OV and SemiCD-VL supply components, not the final claim. Test-set beta tuning and per-dataset best-variant selection are validity concerns, not circular derivation.

full rationale

The paper's claimed contribution is an empirical composition of off-the-shelf foundation models, not a derivation that reduces to its own inputs. The comparator in Eq. (1) uses negative cosine similarity between DINO features in a mask region, and the I-M-C comparator additionally uses IoU and latent matching; neither defines the reported IoU/F1 in terms of the input labels or the method's own predictions. The identifiers (CLIP, SegEarth-OV, Grounding DINO, APE) and mask proposers (SAM, SAM2) are external models with their own training data and evaluations; citing the authors' prior SegEarth-OV [30] and SemiCD-VL [27] supplies component implementations, not a uniqueness theorem or the paper's central superiority claim. The final claim is checked against external benchmark datasets, so the results are falsifiable outside the present paper's fitted values. The main self-referential concerns are protocol-level rather than circular: Section 5.2 states that 'the setting of the hyperparameter beta fluctuates across methods or categories, and we endeavor to find its optimal value' on the test/validation sets, and Table 3 reports one 'OVCD' row that appears to take the best per-dataset variant from Table 1 rather than a single fixed method. These practices can inflate reported numbers, but they do not make the output equal to the input by construction or rename a fitted parameter as a prediction. The acknowledgment that I-M-C is inspired by post-classification comparison [20] and M-C-I is an upgrade of AnyChange [80] also precludes any concealed renaming of existing methods. Under the stated circularity criteria, no load-bearing step reduces to a self-citation or to a definition, so the paper is not circular.

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

The frameworks introduce no new physical or model entities; they recombine published foundation models. The free parameters are thresholds and prompt templates, mostly tuned on test data.

free parameters (4)
  • Comparator threshold beta = not reported; tuned per method and category
    Eq. (1) change score is thresholded by beta; Section 5.2 says beta 'fluctuates across methods or categories' and optimal values are sought, with no validation split given.
  • IoU threshold in I-M-C comparator = not reported
    Section 4.2 classifies an instance as unchanged when the sum of its IoUs with same-category masks exceeds 'a predefined threshold'; the value is not specified.
  • NMS overlap threshold for duplicate masks = not reported
    Section 4.1 removes duplicate masks with non-maximum suppression based on outer bounding boxes and IoU predictions; the overlap threshold is not specified.
  • Text prompt templates and synonyms = not reported
    Section 5.2 notes prompts vary by VLM and that synonyms are added to foreground/background prompts, but exact prompt strings are not given.
assumptions (4)
  • domain assumption Off-the-shelf natural-image foundation models transfer to remote sensing images
    Section 1 and Section 4 assert this based on cited works such as SegEarth-OV and AnyChange and rely on it for both frameworks.
  • domain assumption DINO/DINOv2 feature similarity is a reliable change signal across categories
    The comparator in Eq. (1) uses negative cosine similarity of DINO features as the change score; no category-specific calibration is provided.
  • domain assumption SAM mask proposals cover all changed objects and align across the two dates
    M-C-I concatenates t1 and t2 mask sets and NMS deduplicates (Section 4.1), assuming unchanged objects have near-identical masks so repeated masks are duplicates.
  • domain assumption Instance-level comparison avoids pixel-level pseudo changes
    Section 1 motivates the frameworks by citing works [27, 80] for instance-level superiority; this is adopted rather than experimentally established here.

how reviews work

0 comments
Cite this review

Pith. "Pith review of DynamicEarth: How Far are We from Open-Vocabulary Change Detection?." pith.science (2026). https://pith.science/paper/T4ONG5C4

@misc{pith2026250112931,
  author       = {Pith},
  title        = {Pith review of: DynamicEarth: How Far are We from Open-Vocabulary Change Detection?},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/T4ONG5C4}},
  note         = {Machine review of arXiv:2501.12931}
}
read the original abstract

Monitoring Earth's evolving land covers requires methods capable of detecting changes across a wide range of categories and contexts. Existing change detection methods are hindered by their dependency on predefined classes, reducing their effectiveness in open-world applications. To address this issue, we introduce open-vocabulary change detection (OVCD), a novel task that bridges vision and language to detect changes across any category. Considering the lack of high-quality data and annotation, we propose two training-free frameworks, M-C-I and I-M-C, which leverage and integrate off-the-shelf foundation models for the OVCD task. The insight behind the M-C-I framework is to discover all potential changes and then classify these changes, while the insight of I-M-C framework is to identify all targets of interest and then determine whether their states have changed. Based on these two frameworks, we instantiate to obtain several methods, e.g., SAM-DINOv2-SegEarth-OV, Grounding-DINO-SAM2-DINO, etc. Extensive evaluations on 5 benchmark datasets demonstrate the superior generalization and robustness of our OVCD methods over existing supervised and unsupervised methods. To support continued exploration, we release DynamicEarth, a dedicated codebase designed to advance research and application of OVCD. https://likyoo.github.io/DynamicEarth

Figures

Figures reproduced from arXiv: 2501.12931 by the authors.

Figure 1
Figure 1. The two OVCD frameworks proposed in this paper. (a) [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Different change detection tasks: (a) Binary change detection aims at discovering all (interested) changes and [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Open-vocabulary change detection examples. In each group: [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Annotation-Free Open-Vocabulary Segmentation for Remote-Sensing Images

    cs.CV 2025-08 conditional novelty 6.0 of 10

    SegEarth-OV performs annotation-free open-vocabulary segmentation of remote-sensing images by upsampling CLIP features, removing global bias, and distilling optical knowledge into a SAR encoder.

Reference graph

Works this paper leans on

82 extracted references · 52 canonical work pages · cited by 1 Pith paper

  1. [1]

    Slic superpix- els compared to state-of-the-art superpixel methods

    Radhakrishna Achanta, Appu Shaji, Kevin Smith, Aurelien Lucchi, Pascal Fua, and Sabine S ¨usstrunk. Slic superpix- els compared to state-of-the-art superpixel methods. IEEE transactions on pattern analysis and machine intelligence , 34(11):2274–2282, 2012

    2012

  2. [2]

    A theoreti- cal framework for unsupervised change detection based on change vector analysis in the polar domain

    Francesca Bovolo and Lorenzo Bruzzone. A theoreti- cal framework for unsupervised change detection based on change vector analysis in the polar domain. IEEE Trans- actions on Geoscience and Remote Sensing, 45(1):218–236, 2006

    2006

  3. [3]

    Signature verification using a” siamese” time delay neural network

    Jane Bromley, Isabelle Guyon, Yann LeCun, Eduard S¨ackinger, and Roopak Shah. Signature verification using a” siamese” time delay neural network. Advances in neural information processing systems, 6, 1993

    work page 1993

  4. [4]

    Open-vocabulary remote sensing image semantic segmenta- tion

    Qinglong Cao, Yuntian Chen, Chao Ma, and Xiaokang Yang. Open-vocabulary remote sensing image semantic segmenta- tion. arXiv preprint arXiv:2409.07683, 2024

    arXiv 2024

  5. [5]

    Unsupervised change detection in satellite images using principal component analysis and k-means clustering

    Turgay Celik. Unsupervised change detection in satellite images using principal component analysis and k-means clustering. IEEE geoscience and remote sensing letters , 6(4):772–776, 2009

    work page 2009

  6. [6]

    Hongruixuan Chen, Cuiling Lan, Jian Song, Clifford Broni- Bediako, Junshi Xia, and Naoto Yokoya. Objformer: Learn- ing land-cover changes from paired osm data and optical high-resolution imagery via object-guided transformer.IEEE Transactions on Geoscience and Remote Sensing , 62:1–22, 2024

    work page 2024

  7. [7]

    A spatial-temporal attention- based method and a new dataset for remote sensing image change detection

    Hao Chen and Zhenwei Shi. A spatial-temporal attention- based method and a new dataset for remote sensing image change detection. Remote Sensing, 12(10):1662, 2020

    work page 2020

  8. [8]

    Rsmamba: Remote sens- ing image classification with state space model

    Keyan Chen, Bowen Chen, Chenyang Liu, Wenyuan Li, Zhengxia Zou, and Zhenwei Shi. Rsmamba: Remote sens- ing image classification with state space model. IEEE Geo- science and Remote Sensing Letters, 2024

    work page 2024

Show all 82 references
  1. [9]

    Rsprompter: Learning to prompt for remote sensing instance segmenta- tion based on visual foundation model

    Keyan Chen, Chenyang Liu, Hao Chen, Haotian Zhang, Wenyuan Li, Zhengxia Zou, and Zhenwei Shi. Rsprompter: Learning to prompt for remote sensing instance segmenta- tion based on visual foundation model. IEEE Transactions on Geoscience and Remote Sensing, 2024

    2024

  2. [10]

    Time travelling pixels: Bitemporal features integration with foundation model for remote sensing image change detec- tion

    Keyan Chen, Chengyang Liu, Wenyuan Li, Zili Liu, Hao Chen, Haotian Zhang, Zhengxia Zou, and Zhenwei Shi. Time travelling pixels: Bitemporal features integration with foundation model for remote sensing image change detec- tion. In IGARSS 2024-2024 IEEE International Geoscience...

    2024

  3. [11]

    Masked-attention mask transformer for universal image segmentation

    Bowen Cheng, Ishan Misra, Alexander G Schwing, Alexan- der Kirillov, and Rohit Girdhar. Masked-attention mask transformer for universal image segmentation. In Proceed- ings of the IEEE/CVF conference on computer vision and pattern recognition, pages 1290–1299, 2022

    2022

  4. [12]

    Cat- seg: Cost aggregation for open-vocabulary semantic seg- mentation

    Seokju Cho, Heeseong Shin, Sunghwan Hong, Anurag Arnab, Paul Hongsuck Seo, and Seungryong Kim. Cat- seg: Cost aggregation for open-vocabulary semantic seg- mentation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 4113– 4123, 2024

    2024

  5. [13]

    Molmo and pixmo: Open weights and open data for state-of-the-art multimodal models

    Matt Deitke, Christopher Clark, Sangho Lee, Rohun Tri- pathi, Yue Yang, Jae Sung Park, Mohammadreza Salehi, Niklas Muennighoff, Kyle Lo, Luca Soldaini, et al. Molmo and pixmo: Open weights and open data for state-of-the-art multimodal models. arXiv preprint arXiv:2409.17146, 2024

    2024 arXiv

  6. [14]

    Feature-guided multi- task change detection network

    Yupeng Deng, Jiansheng Chen, Shiming Yi, Anzhi Yue, Yu Meng, Jingbo Chen, and Yi Zhang. Feature-guided multi- task change detection network. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing , 15:9667–9679, 2022

    2022

  7. [15]

    Tchange: a hybrid transformer-cnn change detection network

    Yupeng Deng, Yu Meng, Jingbo Chen, Anzhi Yue, Diyou Liu, and Jing Chen. Tchange: a hybrid transformer-cnn change detection network. Remote Sensing , 15(5):1219, 2023

    2023

  8. [16]

    Un- supervised deep slow feature analysis for change detection in multi-temporal remote sensing images

    Bo Du, Lixiang Ru, Chen Wu, and Liangpei Zhang. Un- supervised deep slow feature analysis for change detection in multi-temporal remote sensing images. IEEE Trans- actions on Geoscience and Remote Sensing , 57(12):9976– 9992, 2019

    2019

  9. [17]

    Convolutional neural network features based change detection in satellite images

    Arabi Mohammed El Amin, Qingjie Liu, and Yunhong Wang. Convolutional neural network features based change detection in satellite images. In First International Work- shop on Pattern Recognition, volume 10011, pages 181–186. SPIE, 2016

    2016

  10. [18]

    Changer: Feature inter- action is what you need for change detection

    Sheng Fang, Kaiyu Li, and Zhe Li. Changer: Feature inter- action is what you need for change detection. IEEE Trans- actions on Geoscience and Remote Sensing, 61:1–11, 2023

    2023

  11. [19]

    Snunet-cd: A densely connected siamese network for change detection of vhr images

    Sheng Fang, Kaiyu Li, Jinyuan Shao, and Zhe Li. Snunet-cd: A densely connected siamese network for change detection of vhr images. IEEE Geoscience and Remote Sensing Let- ters, 19:1–5, 2021

    2021

  12. [20]

    Procedures for change detection using landsat digital data

    Philip J Howarth and Gregory M Wickware. Procedures for change detection using landsat digital data. International Journal of Remote Sensing, 2(3):277–291, 1981

    1981

  13. [21]

    Fully convolutional networks for multisource building extraction from an open aerial and satellite imagery data set

    Shunping Ji, Shiqing Wei, and Meng Lu. Fully convolutional networks for multisource building extraction from an open aerial and satellite imagery data set. IEEE Transactions on geoscience and remote sensing, 57(1):574–586, 2018

    2018

  14. [22]

    Segment anything in high qual- ity

    Lei Ke, Mingqiao Ye, Martin Danelljan, Yu-Wing Tai, Chi- Keung Tang, Fisher Yu, et al. Segment anything in high qual- ity. Advances in Neural Information Processing Systems, 36, 2024

    2024

  15. [23]

    Segment any- thing

    Alexander Kirillov, Eric Mintun, Nikhila Ravi, Hanzi Mao, Chloe Rolland, Laura Gustafson, Tete Xiao, Spencer White- head, Alexander C Berg, Wan-Yen Lo, et al. Segment any- thing. In Proceedings of the IEEE/CVF International Con- ference on Computer Vision, pages 4015–4026, 2023

    2023

  16. [24]

    Clearclip: Decom- posing clip representations for dense vision-language infer- ence

    Mengcheng Lan, Chaofeng Chen, Yiping Ke, Xinjiang Wang, Litong Feng, and Wayne Zhang. Clearclip: Decom- posing clip representations for dense vision-language infer- ence. In European Conference on Computer Vision , pages 143–160. Springer, 2025

    2025

  17. [25]

    Proxyclip: Proxy attention improves clip for open-vocabulary segmentation

    Mengcheng Lan, Chaofeng Chen, Yiping Ke, Xinjiang Wang, Litong Feng, and Wayne Zhang. Proxyclip: Proxy attention improves clip for open-vocabulary segmentation. In European Conference on Computer Vision, pages 70–88. Springer, 2025

    2025

  18. [26]

    Language-driven semantic seg- mentation

    Boyi Li, Kilian Q Weinberger, Serge Belongie, Vladlen Koltun, and Ren ´e Ranftl. Language-driven semantic seg- mentation. arXiv preprint arXiv:2201.03546, 2022

    2022 arXiv

  19. [27]

    Semicd-vl: Visual- language model guidance makes better semi-supervised change detector

    Kaiyu Li, Xiangyong Cao, Yupeng Deng, Jiayi Song, Jun- min Liu, Deyu Meng, and Zhi Wang. Semicd-vl: Visual- language model guidance makes better semi-supervised change detector. IEEE Transactions on Geoscience and Re- mote Sensing, 2024

    2024

  20. [28]

    A new learn- ing paradigm for foundation model-based remote-sensing change detection

    Kaiyu Li, Xiangyong Cao, and Deyu Meng. A new learn- ing paradigm for foundation model-based remote-sensing change detection. IEEE Transactions on Geoscience and Re- mote Sensing, 62:1–12, 2024

    2024

  21. [29]

    Open-cd: A comprehensive tool- box for change detection

    Kaiyu Li, Jiawei Jiang, Andrea Codegoni, Chengxi Han, Yu- peng Deng, Keyan Chen, Zhuo Zheng, Hao Chen, Zhengxia Zou, Zhenwei Shi, et al. Open-cd: A comprehensive tool- box for change detection. arXiv preprint arXiv:2407.15317, 2024

    2024 arXiv

  22. [30]

    Segearth-ov: Towards training-free open-vocabulary segmentation for remote sens- ing images

    Kaiyu Li, Ruixun Liu, Xiangyong Cao, Xueru Bai, Feng Zhou, Deyu Meng, and Zhi Wang. Segearth-ov: Towards training-free open-vocabulary segmentation for remote sens- ing images. arXiv preprint arXiv:2410.01768, 2024

    2024 arXiv

  23. [31]

    Open-vocabulary semantic segmentation with mask-adapted clip

    Feng Liang, Bichen Wu, Xiaoliang Dai, Kunpeng Li, Yinan Zhao, Hang Zhang, Peizhao Zhang, Peter Vajda, and Diana Marculescu. Open-vocabulary semantic segmentation with mask-adapted clip. In Proceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition, pag...

    2023

  24. [32]

    Improved baselines with visual instruction tuning

    Haotian Liu, Chunyuan Li, Yuheng Li, and Yong Jae Lee. Improved baselines with visual instruction tuning. In Pro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 26296–26306, 2024

    2024

  25. [33]

    Visual instruction tuning

    Haotian Liu, Chunyuan Li, Qingyang Wu, and Yong Jae Lee. Visual instruction tuning. Advances in neural information processing systems, 36, 2024

    2024

  26. [34]

    A cnn-transformer network with multiscale context aggrega- tion for fine-grained cropland change detection

    Mengxi Liu, Zhuoqun Chai, Haojun Deng, and Rong Liu. A cnn-transformer network with multiscale context aggrega- tion for fine-grained cropland change detection. IEEE Jour- nal of Selected Topics in Applied Earth Observations and Remote Sensing, 15:4297–4306, 2022

    2022

  27. [35]

    Grounding dino: Marrying dino with grounded pre-training for open-set object detection

    Shilong Liu, Zhaoyang Zeng, Tianhe Ren, Feng Li, Hao Zhang, Jie Yang, Qing Jiang, Chunyuan Li, Jianwei Yang, Hang Su, et al. Grounding dino: Marrying dino with grounded pre-training for open-set object detection. In European Conference on Computer Vision , pages 38–55. Springer, 2025

    2025

  28. [36]

    Change detection techniques

    Dengsheng Lu, Paul Mausel, Eduardo Brondizio, and Emilio Moran. Change detection techniques. International journal of remote sensing, 25(12):2365–2401, 2004

    2004

  29. [37]

    Digital change detection using remotely sensed data for monitoring green space destruction in tabriz

    H Mahmoudzadeh. Digital change detection using remotely sensed data for monitoring green space destruction in tabriz. 2007

    2007

  30. [38]

    Change vector analysis: An approach for detecting forest changes with landsat

    William A Malila. Change vector analysis: An approach for detecting forest changes with landsat. In LARS symposia , page 385, 1980

    1980

  31. [39]

    Dinov2: Learning robust visual features without supervision

    Maxime Oquab, Timoth ´ee Darcet, Th ´eo Moutakanni, Huy V o, Marc Szafraniec, Vasil Khalidov, Pierre Fernandez, Daniel Haziza, Francisco Massa, Alaaeldin El-Nouby, et al. Dinov2: Learning robust visual features without supervision. arXiv preprint arXiv:2304.07193, 2023

    2023 arXiv

  32. [40]

    Detecting building changes with off-nadir aerial images

    Chao Pang, Jiang Wu, Jian Ding, Can Song, and Gui-Song Xia. Detecting building changes with off-nadir aerial images. Science China Information Sciences, 66(4):140306, 2023

    2023

  33. [41]

    Deep learning change detection techniques for optical remote sensing imagery: Status, perspectives and challenges

    Daifeng Peng, Xuelian Liu, Yongjun Zhang, Haiyan Guan, Yansheng Li, and Lorenzo Bruzzone. Deep learning change detection techniques for optical remote sensing imagery: Status, perspectives and challenges. International Jour- nal of Applied Earth Observation and Geoinformation ...

    2025

  34. [42]

    Learning transferable visual models from natural language supervi- sion

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. Learning transferable visual models from natural language supervi- sion. In International conference on machine learning, ...

    2021

  35. [43]

    Sam 2: Segment anything in images and videos

    Nikhila Ravi, Valentin Gabeur, Yuan-Ting Hu, Ronghang Hu, Chaitanya Ryali, Tengyu Ma, Haitham Khedr, Roman R¨adle, Chloe Rolland, Laura Gustafson, et al. Sam 2: Segment anything in images and videos. arXiv preprint arXiv:2408.00714, 2024

    2024 arXiv

  36. [44]

    Learning a classification model for segmen- tation

    Ren and Malik. Learning a classification model for segmen- tation. In Proceedings ninth IEEE international conference on computer vision, pages 10–17. IEEE, 2003

    2003

  37. [45]

    Segment anything, from space? In Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, pages 8355–8365, 2024

    Simiao Ren, Francesco Luzi, Saad Lahrichi, Kaleb Kas- saw, Leslie M Collins, Kyle Bradbury, and Jordan M Malof. Segment anything, from space? In Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, pages 8355–8365, 2024

    2024

  38. [46]

    Dino-x: A unified vision model for open- world object detection and understanding

    Tianhe Ren, Yihao Chen, Qing Jiang, Zhaoyang Zeng, Yuda Xiong, Wenlong Liu, Zhengyu Ma, Junyi Shen, Yuan Gao, Xiaoke Jiang, et al. Dino-x: A unified vision model for open- world object detection and understanding. arXiv preprint arXiv:2411.14347, 2024

    2024 arXiv

  39. [47]

    Grounding dino 1.5: Advance the” edge” of open-set object detection

    Tianhe Ren, Qing Jiang, Shilong Liu, Zhaoyang Zeng, Wen- long Liu, Han Gao, Hongjie Huang, Zhengyu Ma, Xiaoke Jiang, Yihao Chen, et al. Grounding dino 1.5: Advance the” edge” of open-set object detection. arXiv preprint arXiv:2405.10300, 2024

    2024 arXiv

  40. [48]

    Unsupervised deep change vector analysis for multiple- change detection in vhr images

    Sudipan Saha, Francesca Bovolo, and Lorenzo Bruzzone. Unsupervised deep change vector analysis for multiple- change detection in vhr images. IEEE Transactions on Geo- science and Remote Sensing, 57(6):3677–3693, 2019

    2019

  41. [49]

    S2looking: A satellite side-looking dataset for building change detection

    Li Shen, Yao Lu, Hao Chen, Hao Wei, Donghai Xie, Jiabao Yue, Rui Chen, Shouye Lv, and Bitao Jiang. S2looking: A satellite side-looking dataset for building change detection. Remote Sensing, 13(24):5094, 2021

    2021

  42. [50]

    Aligning and prompting everything all at once for univer- sal visual perception

    Yunhang Shen, Chaoyou Fu, Peixian Chen, Mengdan Zhang, Ke Li, Xing Sun, Yunsheng Wu, Shaohui Lin, and Rongrong Ji. Aligning and prompting everything all at once for univer- sal visual perception. In Proceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recogn...

    2024

  43. [51]

    Normalized cuts and image segmentation

    Jianbo Shi and Jitendra Malik. Normalized cuts and image segmentation. IEEE Transactions on pattern analysis and machine intelligence, 22(8):888–905, 2000

    2000

  44. [52]

    Tinysam: Pushing the envelope for efficient segment any- thing model

    Han Shu, Wenshuo Li, Yehui Tang, Yiman Zhang, Yi- hao Chen, Houqiang Li, Yunhe Wang, and Xinghao Chen. Tinysam: Pushing the envelope for efficient segment any- thing model. arXiv preprint arXiv:2312.13789, 2023

    2023 arXiv

  45. [53]

    Zhendong Sun, Yanfei Zhong, Xinyu Wang, and Liangpei Zhang. Identifying cropland non-agriculturalization with high representational consistency from bi-temporal high- resolution remote sensing images: From benchmark datasets to real-world application. ISPRS Journal of Photogra...

    2024

  46. [54]

    Segment change model (scm) for un- supervised change detection in vhr remote sensing images: a case study of buildings

    Xiaoliang Tan, Guanzhou Chen, Tong Wang, Jiaqi Wang, and Xiaodong Zhang. Segment change model (scm) for un- supervised change detection in vhr remote sensing images: a case study of buildings. arXiv preprint arXiv:2312.16410, 2023

    2023 arXiv

  47. [55]

    Xu Tang, Huayu Zhang, Lichao Mou, Fang Liu, Xiangrong Zhang, Xiao Xiang Zhu, and Licheng Jiao. An unsupervised remote sensing change detection method based on multi- scale graph convolutional network and metric learning.IEEE Transactions on Geoscience and Remote Sensing , 60:1...

    2021

  48. [56]

    Gemma: Open models based on gemini research and tech- nology

    Gemma Team, Thomas Mesnard, Cassidy Hardin, Robert Dadashi, Surya Bhupatiraju, Shreya Pathak, Laurent Sifre, Morgane Rivi `ere, Mihir Sanjay Kale, Juliette Love, et al. Gemma: Open models based on gemini research and tech- nology. arXiv preprint arXiv:2403.08295, 2024

    2024 arXiv

  49. [57]

    Hi-ucd: A large-scale dataset for urban semantic change detection in remote sensing imagery

    Shiqi Tian, Ailong Ma, Zhuo Zheng, and Yanfei Zhong. Hi-ucd: A large-scale dataset for urban semantic change detection in remote sensing imagery. arXiv preprint arXiv:2011.03247, 2020

    2011 arXiv

  50. [58]

    Large-scale deep learning based binary and semantic change detection in ultra high resolution remote sensing imagery: From benchmark datasets to urban application

    Shiqi Tian, Yanfei Zhong, Zhuo Zheng, Ailong Ma, Xicheng Tan, and Liangpei Zhang. Large-scale deep learning based binary and semantic change detection in ultra high resolution remote sensing imagery: From benchmark datasets to urban application. ISPRS Journal of Photogrammetry...

    2022

  51. [59]

    Llama 2: Open foundation and fine-tuned chat models.arXiv preprint arXiv:2307.09288, 2023

    Hugo Touvron, Louis Martin, Kevin Stone, Peter Albert, Amjad Almahairi, Yasmine Babaei, Nikolay Bashlykov, Soumya Batra, Prajjwal Bhargava, Shruti Bhosale, et al. Llama 2: Open foundation and fine-tuned chat models.arXiv preprint arXiv:2307.09288, 2023

    2023 arXiv

  52. [60]

    Attention is all you need

    A Vaswani. Attention is all you need. Advances in Neural Information Processing Systems, 2017

    2017

  53. [61]

    Sclip: Rethink- ing self-attention for dense vision-language inference

    Feng Wang, Jieru Mei, and Alan Yuille. Sclip: Rethink- ing self-attention for dense vision-language inference. In European Conference on Computer Vision, pages 315–332. Springer, 2025

    2025

  54. [62]

    A study of test-time contrastive concepts for open-world, open-vocabulary semantic segmentation

    Monika Wysocza ´nska, Antonin V obecky, Amaia Cardiel, Tomasz Trzci´nski, Renaud Marlet, Andrei Bursuc, and Ori- ane Sim ´eoni. A study of test-time contrastive concepts for open-world, open-vocabulary semantic segmentation. arXiv preprint arXiv:2407.05061, 2024

    2024 arXiv

  55. [63]

    Florence-2: Advancing a unified representation for a variety of vision tasks

    Bin Xiao, Haiping Wu, Weijian Xu, Xiyang Dai, Houdong Hu, Yumao Lu, Michael Zeng, Ce Liu, and Lu Yuan. Florence-2: Advancing a unified representation for a variety of vision tasks. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 481...

    2024

  56. [64]

    Groupvit: Semantic segmentation emerges from text supervision

    Jiarui Xu, Shalini De Mello, Sifei Liu, Wonmin Byeon, Thomas Breuel, Jan Kautz, and Xiaolong Wang. Groupvit: Semantic segmentation emerges from text supervision. In Proceedings of the IEEE/CVF Conference on Computer Vi- sion and Pattern Recognition, pages 18134–18144, 2022

    2022

  57. [65]

    Side adapter network for open-vocabulary semantic segmentation

    Mengde Xu, Zheng Zhang, Fangyun Wei, Han Hu, and Xi- ang Bai. Side adapter network for open-vocabulary semantic segmentation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 2945– 2954, 2023

    2023

  58. [66]

    xgen-mm (blip-3): A family of open large multimodal models

    Le Xue, Manli Shu, Anas Awadalla, Jun Wang, An Yan, Senthil Purushwalkam, Honglu Zhou, Viraj Prabhu, Yu- tong Dai, Michael S Ryoo, et al. xgen-mm (blip-3): A family of open large multimodal models. arXiv preprint arXiv:2408.08872, 2024

    2024

  59. [67]

    Semantic change detection with asymmetric siamese networks

    Kunping Yang, Gui-Song Xia, Zicheng Liu, Bo Du, Wen Yang, Marcello Pelillo, and Liangpei Zhang. Semantic change detection with asymmetric siamese networks. arXiv preprint arXiv:2010.05687, 2020

    2010 arXiv

  60. [68]

    Asymmetric siamese networks for semantic change detection in aerial im- ages

    Kunping Yang, Gui-Song Xia, Zicheng Liu, Bo Du, Wen Yang, Marcello Pelillo, and Liangpei Zhang. Asymmetric siamese networks for semantic change detection in aerial im- ages. IEEE Transactions on Geoscience and Remote Sens- ing, 60:1–18, 2021

    2021

  61. [69]

    Unimatch v2: Pushing the limit of semi-supervised semantic segmentation

    Lihe Yang, Zhen Zhao, and Hengshuang Zhao. Unimatch v2: Pushing the limit of semi-supervised semantic segmentation. arXiv preprint arXiv:2410.10777, 2024

    2024 arXiv

  62. [70]

    To- wards open-vocabulary remote sensing image semantic seg- mentation

    Chengyang Ye, Yunzhi Zhuge, and Pingping Zhang. To- wards open-vocabulary remote sensing image semantic seg- mentation. In Proceedings of the AAAI Conference on Arti- ficial Intelligence, 2025

    2025

  63. [71]

    Sigmoid loss for language image pre-training

    Xiaohua Zhai, Basil Mustafa, Alexander Kolesnikov, and Lucas Beyer. Sigmoid loss for language image pre-training. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 11975–11986, 2023

    2023

  64. [72]

    Faster segment anything: Towards lightweight sam for mo- bile applications

    Chaoning Zhang, Dongshen Han, Yu Qiao, Jung Uk Kim, Sung-Ho Bae, Seungkyu Lee, and Choong Seon Hong. Faster segment anything: Towards lightweight sam for mo- bile applications. arXiv preprint arXiv:2306.14289, 2023

    2023 arXiv

  65. [73]

    Corrclip: Recon- structing correlations in clip with off-the-shelf foundation models for open-vocabulary semantic segmentation

    Dengke Zhang, Fagui Liu, and Quan Tang. Corrclip: Recon- structing correlations in clip with off-the-shelf foundation models for open-vocabulary semantic segmentation. arXiv preprint arXiv:2411.10086, 2024

    2024 arXiv

  66. [74]

    Dino: Detr with improved denoising anchor boxes for end-to-end object detection

    Hao Zhang, Feng Li, Shilong Liu, Lei Zhang, Hang Su, Jun Zhu, Lionel M Ni, and Heung-Yeung Shum. Dino: Detr with improved denoising anchor boxes for end-to-end object detection. arXiv preprint arXiv:2203.03605, 2022

    2022 arXiv

  67. [75]

    Cross-domain landslide mapping from large-scale re- mote sensing images using prototype-guided domain-aware progressive representation learning

    Xiaokang Zhang, Weikang Yu, Man-On Pun, and Wenzhong Shi. Cross-domain landslide mapping from large-scale re- mote sensing images using prototype-guided domain-aware progressive representation learning. ISPRS Journal of Pho- togrammetry and Remote Sensing, 197:1–17, 2023

    2023

  68. [76]

    Efficientvit- sam: Accelerated segment anything model without perfor- mance loss

    Zhuoyang Zhang, Han Cai, and Song Han. Efficientvit- sam: Accelerated segment anything model without perfor- mance loss. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 7859– 7863, 2024

    2024

  69. [77]

    An open and com- prehensive pipeline for unified object grounding and detec- tion

    Xiangyu Zhao, Yicheng Chen, Shilin Xu, Xiangtai Li, Xin- jiang Wang, Yining Li, and Haian Huang. An open and com- prehensive pipeline for unified object grounding and detec- tion. arXiv preprint arXiv:2401.02361, 2024

    2024 arXiv

  70. [78]

    Fast segment any- thing

    Xu Zhao, Wenchao Ding, Yongqi An, Yinglong Du, Tao Yu, Min Li, Ming Tang, and Jinqiao Wang. Fast segment any- thing. arXiv preprint arXiv:2306.12156, 2023

    2023 arXiv

  71. [79]

    Changemask: Deep multi-task encoder- transformer-decoder architecture for semantic change detec- tion

    Zhuo Zheng, Yanfei Zhong, Shiqi Tian, Ailong Ma, and Liangpei Zhang. Changemask: Deep multi-task encoder- transformer-decoder architecture for semantic change detec- tion. ISPRS Journal of Photogrammetry and Remote Sens- ing, 183:228–239, 2022

    2022

  72. [80]

    Segment any change

    Zhuo Zheng, Yanfei Zhong, Liangpei Zhang, and Stefano Ermon. Segment any change. In Advances in Neural Infor- mation Processing Systems, 2024

    2024

  73. [81]

    A survey on open- vocabulary detection and segmentation: Past, present, and future

    Chaoyang Zhu and Long Chen. A survey on open- vocabulary detection and segmentation: Past, present, and future. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2024

    2024

  74. [82]

    Semantic-cd: Remote sensing im- age semantic change detection towards open-vocabulary set- ting

    Yongshuo Zhu, Lu Li, Keyan Chen, Chenyang Liu, Fugen Zhou, and Zhenwei Shi. Semantic-cd: Remote sensing im- age semantic change detection towards open-vocabulary set- ting. arXiv preprint arXiv:2501.06808, 2025

    2025 arXiv

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.