Pith. sign in

REVIEW 3 major objections 5 minor 58 references

Zero-OVCD: Bridging Training-Free Foundation Models and Pseudo-Label Learning for Open-Vocabulary Change Detection

T0 review · 3 major / 5 minor · reviewed 2026-08-16 · deepseek-v4-flash

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

Pith's one-line read The paper claims that a text-prompt-only, two-stage pipeline — frozen foundation models producing pseudo-labels followed by a noise-aware-trained detector — outperforms prior open-vocabulary change detection on four benchmarks without any…

desk verdict The two-stage pseudo-label learning idea is real, but the headline F1 numbers are likely tuned on the test splits, so treat the gains as unverified until the authors rerun their sensitivity analysis on validation data. read the letter →

arxiv 2608.11663 v1 pith:M5QNY3KV submitted 2026-08-12 cs.CV

Daifeng Peng , Yuanke Peng , Haiyan Guan This is my paper · ORCID
classification cs.CV
keywords open-vocabularychangedetectionpseudo-labellearningtraining-freeinferencevisionfoundationmodelsremotesensingnoise-awaretrainingmaskrefinement
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

The paper claims that open-vocabulary change detection — finding user-specified land-cover changes between two images of the same area — can be solved without any pixel-level labels from the target domain, using only frozen vision foundation models and a text prompt. Stage I turns the frozen models' raw outputs into high-quality change pseudo-labels through three refinement steps: merging automatic and text-guided candidate masks, fusing multiscale semantic similarity scores with margin-based filtering, and correcting or completing masks using cross-temporal response maps. Stage II trains a lightweight change detector on those pseudo-labels while checkpoint voting and high-agreement sample selection suppress the residual label noise. On three building-change benchmarks the pipeline reports F1 scores of 86.25%, 85.82%, and 50.48% at Stage I and 88.65%, 88.85%, and 57.96% at Stage II, plus a macro-average F1 of 50.92% across six categories on the SECOND dataset. If these results hold, open-vocabulary change detection becomes a prompt-driven service: a user types a category and a detector for that category is trained without manual annotation.

What carries the argument

The central mechanism is the two-stage pseudo-label loop. Stage I chains three corrective modules over frozen foundation models. MRM (Mask Refinement Module) fuses class-agnostic masks from SAM3 with text-guided semantic masks, keeping automatic masks whose non-overlap ratio $r_i \geq \tau_1$ and dropping those whose containment count $n_i \geq \tau_2$; SMFM (Similarity-based Multiscale Fusion Module) averages SegEarth-OV3 category-similarity maps over scales $\{0.8, 1.0, 1.2\}$, computes mask-level foreground–background margins $\delta_i^t$, reassigns low-margin predictions to background, and keeps only automatic-source proposals whose calibrated categories differ across the two time points; MCCM (Mask Correction and Completion Module) filters proposals by target-response coverage under area-adaptive thresholds ($\tau_l/\eta_l$ for large masks, $\tau_s/\eta_s$ for small ones, split at $\gamma=500$ pixels) and adds connected components of the cross-temporal high/low-response discrepancy region. Stage II trains ChangerEx, a lightweight bitemporal exchange network, with a composite loss (symmetric cross-entropy plus Lovász-Softmax plus Dice), retains the last three record-improving checkpoints, fuses their predictions with the original pseudo-labels by equal-weight pixel voting (a $2{:}2$ tie counts as changed), and selects high-agreement samples for a reinitialized training run. The load-bearing identity is that pseudo-labels refined by complementary evidence from several frozen models are a sufficient supervision signal for a task-specific detector.

What would settle it

Re-run the sensitivity analysis of Sec. IV-E-2 on the LEVIR-CD and WHU-CD validation splits only, freeze the resulting default thresholds, and then measure test-set F1; if the score drops substantially below the reported 86.25% and 85.82%, the headline numbers depended on test-set information. Alternatively, apply the published thresholds unchanged to a new open-vocabulary change-detection benchmark with different sensors or cities and check whether Stage I still beats the prior best training-free baseline there.

Watch

Extended reading notes

Core claim

On its own terms, the paper establishes that training-free foundation-model inference and pseudo-label-supervised detector learning can be bridged into a working open-vocabulary change detection system with no target-domain pixel annotations. Its central discovery is that a progressive refinement chain — MRM merging SAM3 automatic masks with text-guided semantic masks, SMFM aggregating multiscale category-similarity maps under a foreground–background margin filter, and MCCM exploiting bitemporal target-response discrepancies to delete false candidates and recover missed regions — converts noisy foundation-model outputs into pseudo-labels accurate enough to supervise a task-specific change detector. The noise-aware training strategy (record-checkpoint voting, tie-as-changed fusion, and F1-agreement sample selection) then lets the detector exceed the quality of the labels themselves. The paper reports that this two-stage design outperforms existing open-vocabulary change detection methods on LEVIR-CD, WHU-CD, S2Looking, and SECOND, with all foundation models kept frozen and the only user input being the target-category text prompt. The conclusion also flags, as open limitations, that fixed thresholds may not generalize across imaging conditions, that per-category Stage II adaptation raises cost, and that agreement-based sample selection may preserve shared prediction errors.

Load-bearing premise

The performance claims assume the pseudo-label thresholds in Stage I (for example $\tau_1=0.6$, $\tau_2=4$, $\tau_\delta=0.1$, $\gamma=500$) were chosen without peeking at test labels: the sensitivity analyses in Sec. IV-E-2 report scores on LEVIR-CD and WHU-CD but never state which split they evaluate on, so if the test labels shaped the defaults, the reported F1 figures overstate what an annotation-free user would achieve on unseen data.

Editorial extensions

If this is right

  • Open-vocabulary change detection becomes a prompt-driven procedure: with a category typed in text, a user obtains pseudo-labels and a trained detector for that category, with no pixel-level annotation effort in the target domain.
  • Stage II consistently improves over Stage I (for example 86.25 to 88.65 F1 on LEVIR-CD), implying that noise-aware training — voting across checkpoints and keeping high-agreement samples — lets a detector extract signal from imperfect pseudo-labels rather than merely memorizing their errors.
  • The refinement chain improved all four foundation-model combinations tested, suggesting that the MRM–SMFM–MCCM design is a portable layer that can be re-applied as newer frozen segmenters, encoders, and similarity models appear.
  • The SECOND results show the method is not limited to building change: six land-cover categories are handled one-vs-rest with the same pipeline, and the macro-average F1 rises from 47.91% at Stage I to 50.92% at Stage II.
  • Because the deployed Stage II detector is lightweight (11.39M parameters and roughly 12 ms per image pair on WHU-CD), the expensive multi-foundation-model Stage I can be amortized into a small detector for repeated use.

Reading between the lines

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

  • A direct portability test would freeze the published thresholds and run Stage I on an unseen pair of datasets without any re-tuning; the paper's own cross-benchmark spread (88.65% F1 on LEVIR-CD versus 57.96% on S2Looking at Stage II) hints at how much of the gain may depend on dataset-specific threshold settings.
  • The tie-as-changed voting rule (a 2:2 split among four voters counts as changed) embeds an implicit recall bias; an inexpensive variant would make the tie-break tunable or weight each checkpoint's vote by its proxy-validation F1 instead of voting equally.
  • Since the text prompt and its background contrast are the only semantic inputs, the method's accuracy likely depends on prompt wording; a prompt-sensitivity study with non-expert phrasings would show whether the reported numbers survive casual usage.
  • The per-category gains on SECOND — largest for building and low vegetation, smallest for playground and water — suggest that the frozen models' zero-shot semantic alignment is category-dependent; measuring per-category calibration and choosing prompts or fusion weights per category could reveal where the pipeline's ceiling sits.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The paper proposes Zero-OVCD, a two-stage framework for open-vocabulary change detection (OVCD) that requires no target-domain pixel-level annotations. Stage I generates change pseudo-labels by combining three training-free modules: mask refinement (MRM), multiscale margin-based semantic verification (SMFM), and response-guided correction/completion (MCCM), built on SAM3, DINOv3, and SegEarth-OV3. Stage II trains a ChangerEx detector on these pseudo-labels with checkpoint voting and high-agreement sample selection to mitigate residual noise. The authors report Stage I F1 scores of 86.25%, 85.82%, and 50.48% on LEVIR-CD, WHU-CD, and S2Looking, and Stage II scores of 88.65%, 88.85%, and 57.96% respectively; on SECOND, macro-average F1 rises from 47.91% to 50.92%. The paper claims these results demonstrate an effective annotation-free OVCD solution.

Significance. If validated, the work would be a meaningful step for OVCD: it shows that complementary frozen foundation models can generate pseudo-labels strong enough to train a lightweight task-specific detector, and the two-phase noise-aware training idea is reasonable and clearly ablated. The paper's strengths include a detailed algorithmic description with equations, careful ablations of each Stage I module, sensitivity tables for several thresholds, a comparison across foundation-model combinations, an accuracy-efficiency analysis, and a plan to release code. However, the central quantitative claim currently rests on threshold-selection analyses whose evaluation split is not stated, which is a load-bearing omission; the headline numbers may partly reflect test-set hyperparameter tuning rather than a purely annotation-free protocol. The reported gains over prior training-free methods are large, so the issue must be resolved before the results can be taken at face value.

major comments (3)
  1. [Sec. IV-E-2, Tables IV-VI and Fig. 10] The sensitivity analyses that set the default thresholds report F1/IoU values on LEVIR-CD and WHU-CD identical to the Stage I test results in Table I (e.g., τ1=0.6 gives 86.25/75.82 on LEVIR-CD and 85.82/75.17 on WHU-CD; the same values reappear for τ2=4 and τδ=0.1). The paper does not state whether these analyses used the validation splits or the test splits. If they used the official test splits, then τ1, τ2, τδ, and the coupled MCCM thresholds were selected by maximizing metrics computed from ground-truth labels of exactly the images on which the headline F1 scores are reported. This would make the reported numbers a fitted evaluation rather than a prediction under the announced annotation-free protocol. The paper must either explicitly state that the sensitivity analyses were restricted to validation splits or re-run the threshold selection (and the resulting final evaluations) on a proper validation split, and report the magnitude of any performance change.
  2. [Sec. III-A-2 and Sec. IV-B-1] The angular threshold θ in Eq. (7) is a core parameter of the cross-temporal change proposal construction, but its value is never reported in the implementation details or in any sensitivity analysis. Since θ controls which candidate masks are retained as change proposals, it is a free tuning degree of freedom whose omitted value prevents reproducibility and leaves open the possibility of another test-set-tuned choice. The authors should report the value of θ and explain how it was set, preferably with a sensitivity analysis on a validation split.
  3. [Sec. IV-E-2, Fig. 10] The coupled-threshold sensitivity analysis reports only the mean changed-class IoU over LEVIR-CD and WHU-CD, computed from what appear to be test splits. The text states that τs/ηs is set to (0.9,0.6) because it is the highest point of a monotone increasing trend and τh/τo to (0.75,0.25) after a marked IoU decrease on either side; these choices are not inert and their effect on the reported gains is not quantified. Even if the individual ranges for τ1, τ2, and τδ are narrow, the MCCM thresholds are coupled and may materially affect the final pseudo-labels. The authors should provide validation-split sensitivity curves and report the F1/IoU differences between the chosen defaults and nearby alternatives.
minor comments (5)
  1. [Sec. IV-B-1] The default parameters list mentions τc=0.95, scales {0.8,1.0,1.2}, and γ=500, but not θ; the value of θ should be added here for completeness.
  2. [Sec. III-B, Eq. (21)] The voting rule assigns a 2:2 tie to the changed state for K=3, which is a deliberate asymmetry in 'equal-weight' voting; this should be stated explicitly as a design choice and its effect analyzed or at least commented on.
  3. [Sec. IV-E-2] The sensitivity tables report 'Average' over LEVIR-CD and WHU-CD only, while Fig. 10 uses mean IoU over the same two datasets; this choice should be clarified, especially because S2Looking has a substantially different change distribution and is excluded from the threshold-setting analyses.
  4. [Fig. 10] The caption says asterisks indicate default settings, but the markers are not visible in the black-and-white rendering of the figure; please ensure the markers and their meaning are legible in the camera-ready version.
  5. [Sec. IV-E-1, Table III] The ablation baseline is described as 'without MRM, SMFM, or MCCM,' but the composition of that baseline (e.g., which candidates are retained and how semantic verification is performed) is not fully specified; a sentence giving the exact baseline protocol would improve reproducibility.

Circularity Check

2 steps flagged · score 6.0 of 10

Default thresholds are selected on test-set F1/IoU, so the reported annotation-free gains are partially fitted to the evaluation labels.

  1. fitted input called prediction [Sec. IV-E-2-a/b, Tables IV-VI]
    "0.6* 86.25 75.82 85.82 75.17 86.04 75.50; The best result in each column is shown in bold. * denotes the default parameter setting used in our experiments."

    Tables IV-VI select the default MRM and SMFM thresholds (tau_1=0.6, tau_2=4, tau_delta=0.1) as the values with the highest average F1/IoU across LEVIR-CD and WHU-CD. The rows marked as defaults report exactly the Stage I test-set numbers from Table I (LEVIR-CD 86.25/75.82 and WHU-CD 85.82/75.17), so the only consistent reading is that the sensitivity analyses were run on the same test splits whose ground-truth labels later appear in the headline Stage I results. The paper never states that these analyses used a validation split. The thresholds are therefore fitted to the test labels, and the reported Stage I 'pseudo-label' F1s are not generated under a fixed, annotation-free protocol; they are partly post-hoc selections from the evaluation set.

  2. fitted input called prediction [Sec. IV-E-2-c, Fig. 10]
    "Performance is measured using the mean changed-class IoU across LEVIR-CD and WHU-CD... Among the evaluated settings, large-area masks achieve the highest mean IoU at (tau_l,eta_l) = (0.4,0.6)... reaching its highest evaluated value at (tau_s,eta_s) = (0.9,0.6)... best performance is obtained at (tau_h,tau_o) = (0.75,0.25)... Accordingly, these three threshold pairs are adopted as the default settings in all experiments."

    The MCCM defaults are chosen by maximizing mean IoU over LEVIR-CD and WHU-CD, and the text does not restrict this analysis to a validation split. Since the Stage I test metrics for these datasets (86.25/75.82 and 85.82/75.17) are exactly the values used elsewhere in the paper, the coupled threshold pairs are effectively tuned against the ground-truth labels of the same test images on which the final annotation-free performance is reported. The choice of (tau_s,eta_s)=(0.9,0.6) at the highest evaluated point of a monotone trend and of (tau_h,tau_o)=(0.75,0.25) at a point with marked IoU decrease on both sides shows that the fitted values are not inert.

full rationale

Most of the pipeline is not circular: Stage I is a training-free composition of frozen SAM3/DINOv3/SegEarth-OV3 masks and similarity maps; Stage II supervises ChangerEx with pseudo-labels and selects checkpoints by proxy F1 against pseudo-labels, not ground truth; no equation in the derivation is defined in terms of its own output. The circularity is concentrated in the selection of the default thresholds. Tables IV-VI and Fig. 10 fix tau_1=0.6, tau_2=4, tau_delta=0.1, (tau_l,eta_l)=(0.4,0.6), (tau_s,eta_s)=(0.9,0.6), and (tau_h,tau_o)=(0.75,0.25) by maximizing F1/IoU on LEVIR-CD and WHU-CD. The values at the chosen defaults are identical to the Stage I test results in Table I (86.25/75.82 and 85.82/75.17), so the sensitivity analyses must have been run on the same test splits used for the headline numbers. Because the paper never states that a validation split was used, and because the exact numerical identity rules out coincidental agreement, the reported annotation-free results are partially fitted to the test ground truth. The MCCM pairs in particular are selected at the end of a monotone trend or at an interior maximum, so the tuning effect is not negligible. No load-bearing self-citation was found: references to the authors' prior work ([14], [29]) are contextual and do not carry the derivation. The score of 6 reflects this partial, evaluation-level circularity rather than an equation-level self-derivation.

Assumptions & free parameters 12 free parameters · 4 assumptions · 0 invented entities

The method introduces no new physical entities or mathematical objects, but it depends on roughly a dozen empirically chosen thresholds, including one (θ) that is not reported. The axioms are standard domain assumptions about foundation-model transferability and the definition of semantic change in remote sensing.

free parameters (12)
  • τ1 (non-overlap ratio threshold) = 0.6
    Retains automatic masks with at least this fraction of area outside the union of semantic masks; chosen by sensitivity analysis on LEVIR-CD and WHU-CD (Table IV).
  • τ2 (containment count threshold) = 4
    Removes automatic masks covering at least this many semantic instances; chosen via Table V.
  • τc (containment ratio threshold) = 0.95
    Fraction of a semantic mask that must be covered by an automatic mask to count as contained; fixed in implementation.
  • τδ (semantic margin threshold) = 0.1
    Foreground-background margin below which a mask is reassigned to background; chosen via Table VI.
  • γ (area threshold) = 500 pixels
    Separates small from large masks in MCCM and filters tiny completion components; fixed in implementation.
  • SMFM scales = {0.8, 1.0, 1.2}
    Multiscale resizing factors for SegEarth-OV3 similarity maps; fixed in implementation.
  • τl, ηl (large-mask response thresholds) = 0.4, 0.6
    Response-intensity and coverage thresholds for large masks in MCCM; selected via Fig. 10.
  • τs, ηs (small-mask response thresholds) = 0.9, 0.6
    Response-intensity and coverage thresholds for small masks; selected via Fig. 10.
  • τh, τo (completion thresholds) = 0.75, 0.25
    High and low response thresholds for cross-temporal response-discrepancy completion; selected via Fig. 10.
  • θ (bitemporal feature similarity threshold) = not reported
    Used in Eq. (7) to form the preliminary change proposal set; the paper never reports its value, which is a reproducibility gap.
  • K (number of voting checkpoints) = 3
    Last K record-improving checkpoints used for pseudo-label voting; fixed by design.
  • High-agreement retention ratios = top 60% samples, top 10% validation
    Fraction of samples retained per checkpoint and fraction used as validation in Stage II; fixed by design.
assumptions (4)
  • domain assumption Frozen foundation models (SAM3, DINOv3, SegEarth-OV3) provide sufficient candidate masks, features, and semantic similarity for remote sensing change detection without fine-tuning.
    Invoked throughout Stage I (Algorithm 1, Eqs. 6-8). If these features are inadequate for a target domain, candidate proposal and semantic verification fail.
  • domain assumption A semantic label change between the two timestamps is a valid proxy for a land-cover change of the target category.
    Used in Eq. (11), where proposals are kept only if the calibrated category label differs across time points. This equates semantic transition with change and ignores appearance-only changes.
  • domain assumption Cosine similarity of DINOv3 mask features below cos(θ) indicates a change.
    Used in Eq. (7) to form the preliminary change proposal set; the threshold θ is not reported or analyzed in the paper.
  • domain assumption Baselines marked with † were evaluated under protocols equivalent to the authors' test splits.
    The paper reuses results from prior papers (Tables I and II); if their evaluation settings differ, the headline F1 gains may be inflated.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Zero-OVCD: Bridging Training-Free Foundation Models and Pseudo-Label Learning for Open-Vocabulary Change Detection." pith.science (2026). https://pith.science/paper/M5QNY3KV

@misc{pith2026260811663,
  author       = {Pith},
  title        = {Pith review of: Zero-OVCD: Bridging Training-Free Foundation Models and Pseudo-Label Learning for Open-Vocabulary Change Detection},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/M5QNY3KV}},
  note         = {Machine review of arXiv:2608.11663}
}
read the original abstract

Open-vocabulary change detection (OVCD) enables the identification of user-specified land-cover changes in bitemporal remote sensing images, but existing training-free pipelines remain vulnerable to inaccurate candidate masks, ambiguous semantic assignments, and accumulated inference errors. To address these issues, we propose Zero-OVCD, a two-stage framework that requires no pixel-level annotations from the target domain. In the first stage, high-quality change pseudo-labels are generated through complementary candidate-mask refinement, multiscale semantic similarity fusion with margin-based reliability filtering, and response-guided mask correction and completion. These components jointly suppress noisy candidates, enhance mask-level semantic discrimination, and recover missed change regions. In the second stage, a change detector is trained using the generated pseudo-labels, while checkpoint voting and high-agreement sample selection are introduced to mitigate residual pseudo-label noise. On LEVIR-CD, WHU-CD, and S2Looking, Stage I achieves F1 scores of 86.25%, 85.82%, and 50.48%, while Stage II further improves them to 88.65%, 88.85%, and 57.96%, respectively. On SECOND, the macro-average F1 across six category-wise one-vs-rest tasks increases from 47.91% to 50.92%. These results demonstrate that bridging training-free foundation-model inference with noise-aware pseudo-label learning provides an effective solution for open-vocabulary change detection without target-domain pixel-level annotations. Code will be available at https://github.com/1321663019/Zero-OVCD.

Figures

Figures reproduced from arXiv: 2608.11663 by the authors.

Figure 1
Figure 1. Schematic comparison between existing staged OVCD and progressive pseudo-label refinement in Zero-OVCD. (a) Candidate-mask defects and [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Overview of Zero-OVCD. Stage I generates refined change pseudo-labels using frozen vision foundation models, whereas Stage II updates the [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Illustration of multi-source candidate mask refinement (MRM). Automatic masks are filtered according to their complementarity to and containment [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (7 more)
Figure 4
Figure 4. Figure 4: Illustration of similarity-based multiscale fusion (SMFM). Multiscale category-similarity maps are fused for mask-level classification, followed by [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 5
Figure 5. Figure 5: Illustration of response-guided mask correction and completion (MCCM). Semantic-support filtering removes unreliable proposals, whereas cross [PITH_FULL_IMAGE:figures/full_fig_p008_5.png]
Figure 6
Figure 6. Figure 6: Overview of noise-aware pseudo-label learning. Warm-up checkpoints are used for voting-based pseudo-label updating and high-agreement sample [PITH_FULL_IMAGE:figures/full_fig_p009_6.png]
Figure 7
Figure 7. Figure 7: Qualitative comparison on LEVIR-CD, WHU-CD, and S2Looking. SOV denotes SegEarth-OV. [PITH_FULL_IMAGE:figures/full_fig_p011_7.png]
Figure 8
Figure 8. Figure 8: Category-wise qualitative comparison on SECOND. Colored pixels indicate changes in the queried category, whereas black denotes background. [PITH_FULL_IMAGE:figures/full_fig_p012_8.png]
Figure 9
Figure 9. Figure 9: Qualitative ablation of the Stage I modules on binary CD benchmarks. [PITH_FULL_IMAGE:figures/full_fig_p013_9.png]
Figure 10
Figure 10. Figure 10: Sensitivity to the coupled MCCM thresholds. Mean changed-class [PITH_FULL_IMAGE:figures/full_fig_p015_10.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

58 extracted references · 38 canonical work pages

  1. [1]

    Land use land cover change detection and monitoring of urban growth using remote sensing and GIS techniques: A micro-level study,

    S. Das and D. P. Angadi, “Land use land cover change detection and monitoring of urban growth using remote sensing and GIS techniques: A micro-level study,”GeoJournal, vol. 87, no. 3, pp. 2101–2123, 2022

    work page 2022

  2. [2]

    Remote sensing in urban planning: Contri- butions towards ecologically sound policies?

    T. Wellmann, A. Lausch, E. Andersson, S. Knapp, C. Cortinovis, J. Jache, S. Scheuer, P. Kremer, A. Mascarenhas, R. Kraemer, A. Haase, F. Schug, and D. Haase, “Remote sensing in urban planning: Contri- butions towards ecologically sound policies?”Landscape and Urban Planning, vol. 204, p. 103921, 2020

    work page 2020

  3. [3]

    Destroyed-buildings detection from VHR SAR images using deep features,

    S. Saha, F. Bovolo, and L. Bruzzone, “Destroyed-buildings detection from VHR SAR images using deep features,” inImage and Signal Processing for Remote Sensing XXIV, L. Bruzzone and F. Bovolo, Eds., vol. 10789, International Society for Optics and Photonics. SPIE, 2018, p. 107890Z

    work page 2018

  4. [4]

    Fully convolutional siamese networks for change detection,

    R. Caye Daudt, B. Le Saux, and A. Boulch, “Fully convolutional siamese networks for change detection,” in2018 25th IEEE International Conference on Image Processing (ICIP), 2018, pp. 4063–4067

    work page 2018

  5. [5]

    Remote sensing image change detection with transformers,

    H. Chen, Z. Qi, and Z. Shi, “Remote sensing image change detection with transformers,”IEEE Transactions on Geoscience and Remote Sensing, vol. 60, pp. 1–14, 2022

    2022

  6. [6]

    Mamba: Linear-time sequence modeling with selective state spaces,

    A. Gu and T. Dao, “Mamba: Linear-time sequence modeling with selective state spaces,” inFirst Conference on Language Modeling, 2024

    2024

  7. [7]

    Segment anything,

    A. Kirillov, E. Mintun, N. Ravi, H. Mao, C. Rolland, L. Gustafson, T. Xiao, S. Whitehead, A. C. Berg, W.-Y . Lo, P. Doll´ar, and R. Girshick, “Segment anything,” in2023 IEEE/CVF International Conference on Computer Vision (ICCV), 2023, pp. 3992–4003

    work page 2023

  8. [8]

    Emerging properties in self-supervised vision transform- ers,

    M. Caron, H. Touvron, I. Misra, H. Jegou, J. Mairal, P. Bojanowski, and A. Joulin, “Emerging properties in self-supervised vision transform- ers,” in2021 IEEE/CVF International Conference on Computer Vision (ICCV), 2021, pp. 9630–9640

    2021

Show all 58 references
  1. [9]

    Learning transferable visual models from natural language supervi- sion,

    A. Radford, J. W. Kim, C. Hallacy, A. Ramesh, G. Goh, S. Agarwal, G. Sastry, A. Askell, P. Mishkin, J. Clark, G. Krueger, and I. Sutskever, “Learning transferable visual models from natural language supervi- sion,” inProceedings of the 38th International Conference on Machine ...

    2021

  2. [10]

    Segment any change,

    Z. Zheng, Y . Zhong, L. Zhang, and S. Ermon, “Segment any change,” inAdvances in Neural Information Processing Systems, A. Globerson, L. Mackey, D. Belgrave, A. Fan, U. Paquet, J. Tomczak, and C. Zhang, Eds., vol. 37. Curran Associates, Inc., 2024, pp. 81 204–81 224

    2024

  3. [11]

    Segment change model (SCM) for unsupervised change detection in VHR remote sensing images: A case study of buildings,

    X. Tan, G. Chen, T. Wang, J. Wang, and X. Zhang, “Segment change model (SCM) for unsupervised change detection in VHR remote sensing images: A case study of buildings,” inIGARSS 2024 - 2024 IEEE International Geoscience and Remote Sensing Symposium, 2024, pp. 8577–8580

    2024

  4. [12]

    Fast segment anything,

    X. Zhao, W. Ding, Y . An, Y . Du, T. Yu, M. Li, M. Tang, and J. Wang, “Fast segment anything,” 2023, arXiv:2306.12156

    2023 arXiv

  5. [13]

    DynamicEarth: How far are we from open-vocabulary change detection?

    K. Li, X. Cao, Y . Deng, C. Pang, Z. Xin, H. Qiao, T. Gong, D. Meng, and Z. Wang, “DynamicEarth: How far are we from open-vocabulary change detection?”Proceedings of the AAAI Conference on Artificial Intelligence, vol. 40, no. 8, pp. 6279–6287, 2026

    2026

  6. [14]

    End-to-end change detection for high resolution satellite images using improved UNet++,

    D. Peng, Y . Zhang, and H. Guan, “End-to-end change detection for high resolution satellite images using improved UNet++,”Remote Sensing, vol. 11, no. 11, p. 1382, 2019

    2019

  7. [15]

    A transformer-based siamese net- work for change detection,

    W. G. C. Bandara and V . M. Patel, “A transformer-based siamese net- work for change detection,” inIGARSS 2022 - 2022 IEEE International Geoscience and Remote Sensing Symposium, 2022, pp. 207–210

    2022

  8. [16]

    Hybrid attention-aware transformer network collaborative multiscale feature alignment for building change detection,

    C. Xu, Z. Ye, L. Mei, H. Yu, J. Liu, Y . Yalikun, S. Jin, S. Liu, W. Yang, and C. Lei, “Hybrid attention-aware transformer network collaborative multiscale feature alignment for building change detection,”IEEE Trans- actions on Instrumentation and Measurement, vol. 73, pp. 1–14, 2024

    2024

  9. [17]

    ChangeMamba: Re- mote sensing change detection with spatiotemporal state space model,

    H. Chen, J. Song, C. Han, J. Xia, and N. Yokoya, “ChangeMamba: Re- mote sensing change detection with spatiotemporal state space model,” IEEE Transactions on Geoscience and Remote Sensing, vol. 62, pp. 1– 20, 2024

    2024

  10. [18]

    CDMamba: Incorporating local clues into Mamba for remote sensing image binary change detection,

    H. Zhang, K. Chen, C. Liu, H. Chen, Z. Zou, and Z. Shi, “CDMamba: Incorporating local clues into Mamba for remote sensing image binary change detection,”IEEE Transactions on Geoscience and Remote Sens- ing, vol. 63, pp. 1–16, 2025

    2025

  11. [19]

    S2C: A noise-resistant difference learning framework for unsupervised change detection in VHR remote sensing images,

    L. Ding, X. Zuo, H. Guo, J. Lu, Z. Gong, X. Liu, and J. Lu, “S2C: A noise-resistant difference learning framework for unsupervised change detection in VHR remote sensing images,”Proceedings of the AAAI Conference on Artificial Intelligence, vol. 40, no. 5, pp. 3596–3604, 2026

    2026

  12. [20]

    Prototype-oriented unsupervised change detection for disaster management,

    Y . Oh, M. Seo, D. Kim, and J. Seo, “Prototype-oriented unsupervised change detection for disaster management,” inNeurIPS 2023 Workshop on Tackling Climate Change with Machine Learning, 2023

    2023

  13. [21]

    Unsupervised change detection based on image reconstruction loss,

    H. Noh, J. Ju, M. Seo, J. Park, and D.-G. Choi, “Unsupervised change detection based on image reconstruction loss,” in2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW), 2022, pp. 1351–1360

    2022

  14. [22]

    Unsupervised change detection based on image reconstruction loss with segment anything,

    H. Noh, J. Ju, Y . Kim, M. Kim, and D.-G. Choi, “Unsupervised change detection based on image reconstruction loss with segment anything,” Remote Sensing Letters, vol. 15, no. 9, pp. 919–929, 2024

    2024

  15. [23]

    Exchange means change: An unsupervised single-temporal change detection framework based on intra- and inter-image patch exchange,

    H. Chen, J. Song, C. Wu, B. Du, and N. Yokoya, “Exchange means change: An unsupervised single-temporal change detection framework based on intra- and inter-image patch exchange,”ISPRS Journal of Photogrammetry and Remote Sensing, vol. 206, pp. 87–105, 2023

    2023

  16. [24]

    Integrating segment anything model with instance-level change generation for single-temporal unsupervised change detection,

    X. Zuo, J. Rui, L. Ding, F. Jin, Y . Lin, S. Wang, X. Liu, and J. Lei, “Integrating segment anything model with instance-level change generation for single-temporal unsupervised change detection,”IEEE Transactions on Geoscience and Remote Sensing, vol. 63, pp. 1–17, 2025

    2025

  17. [25]

    Adapting Segment Anything Model for change detection in VHR remote sensing images,

    L. Ding, K. Zhu, D. Peng, H. Tang, K. Yang, and L. Bruzzone, “Adapting Segment Anything Model for change detection in VHR remote sensing images,”IEEE Transactions on Geoscience and Remote Sensing, vol. 62, pp. 1–11, 2024

    2024

  18. [26]

    SCD-SAM: Adapting Segment Anything Model for semantic change 17 detection in remote sensing imagery,

    L. Mei, Z. Ye, C. Xu, H. Wang, Y . Wang, C. Lei, W. Yang, and Y . Li, “SCD-SAM: Adapting Segment Anything Model for semantic change 17 detection in remote sensing imagery,”IEEE Transactions on Geoscience and Remote Sensing, vol. 62, pp. 1–13, 2024

    2024

  19. [27]

    ESAM-CD: Fine-tuned EfficientSAM network with LoRA for weakly supervised remote sensing image change detection,

    M. Wang, L. Zhou, K. Zhang, X. Li, M. Hao, and Y . Ye, “ESAM-CD: Fine-tuned EfficientSAM network with LoRA for weakly supervised remote sensing image change detection,”IEEE Transactions on Geo- science and Remote Sensing, vol. 62, pp. 1–16, 2024

    2024

  20. [28]

    CS-WSCDNet: Class activation mapping and Segment Anything Model-based framework for weakly supervised change detection,

    L. Wang, M. Zhang, and W. Shi, “CS-WSCDNet: Class activation mapping and Segment Anything Model-based framework for weakly supervised change detection,”IEEE Transactions on Geoscience and Remote Sensing, vol. 61, pp. 1–12, 2023

    2023

  21. [29]

    SemSAM-CD: A novel weakly supervised change detection method based on semantic guidance and Segment Anything Model refinement,

    M. Liu, D. Peng, Y . Zhang, and H. Guan, “SemSAM-CD: A novel weakly supervised change detection method based on semantic guidance and Segment Anything Model refinement,”IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, vol. 19, pp. 3352–3370, 2026

    2026

  22. [30]

    SAM 2: Segment anything in images and videos,

    N. Ravi, V . Gabeur, Y .-T. Hu, R. Hu, C. Ryali, T. Ma, H. Khedr, R. R ¨adle, C. Rolland, L. Gustafson, E. Mintun, J. Pan, K. V . Alwala, N. Carion, C.-Y . Wu, R. Girshick, P. Dollar, and C. Feichtenhofer, “SAM 2: Segment anything in images and videos,” inInternational Confere...

    2025

  23. [31]

    SAM 3: Segment anything with concepts,

    N. Carion, L. Gustafson, Y .-T. Hu, S. Debnath, R. Hu, D. S. Coll-Vinent, C. Ryali, K. V . Alwala, H. Khedr, A. Huang, J. Lei, T. Ma, B. Guo, A. Kalla, M. Marks, J. Greer, M. Wang, P. Sun, R. R ¨adle, T. Afouras, E. Mavroudi, K. Xu, T.-H. Wu, Y . Zhou, L. Momeni, R. HAZRA, S. ...

    2026

  24. [32]

    DINOv2: Learning robust visual features without supervision,

    M. Oquab, T. Darcet, T. Moutakanni, H. V . V o, M. Szafraniec, V . Khali- dov, P. Fernandez, D. HAZIZA, F. Massa, A. El-Nouby, M. Assran, N. Ballas, W. Galuba, R. Howes, P.-Y . Huang, S.-W. Li, I. Misra, M. Rabbat, V . Sharma, G. Synnaeve, H. Xu, H. Jegou, J. Mairal, P. Labatu...

    2024

  25. [33]

    Sim ´eoni, H

    O. Sim ´eoni, H. V . V o, M. Seitzer, F. Baldassarre, M. Oquab, C. Jose, V . Khalidov, M. Szafraniec, S. E. Yi, M. Ramamonjisoa, F. Massa, D. HAZIZA, L. Wehrstedt, J. Wang, T. Darcet, T. Moutakanni, L. Sen- tana, C. Roberts, A. Vedaldi, J. Tolan, J. Brandt, C. Couprie, J. Mair...

    2026

  26. [34]

    Aligning and prompting everything all at once for universal visual perception,

    Y . Shen, C. Fu, P. Chen, M. Zhang, K. Li, X. Sun, Y . Wu, S. Lin, and R. Ji, “Aligning and prompting everything all at once for universal visual perception,” in2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2024, pp. 13 193–13 203

    2024

  27. [35]

    SegEarth-OV: Towards training-free open-vocabulary segmentation for remote sensing images,

    K. Li, R. Liu, X. Cao, X. Bai, F. Zhou, D. Meng, and Z. Wang, “SegEarth-OV: Towards training-free open-vocabulary segmentation for remote sensing images,” in2025 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2025, pp. 10 545–10 556

    2025

  28. [36]

    SegEarth-OV3: Exploring SAM 3 for open-vocabulary semantic seg- mentation in remote sensing images,

    K. Li, S. Zhang, Y . Wang, Y . Deng, Z. Wang, D. Meng, and X. Cao, “SegEarth-OV3: Exploring SAM 3 for open-vocabulary semantic seg- mentation in remote sensing images,” 2026, arXiv:2512.08730

    2026 arXiv

  29. [37]

    Extract free dense labels from CLIP,

    C. Zhou, C. C. Loy, and B. Dai, “Extract free dense labels from CLIP,” in Computer Vision – ECCV 2022, S. Avidan, G. Brostow, M. Ciss´e, G. M. Farinella, and T. Hassner, Eds. Cham: Springer Nature Switzerland, 2022, pp. 696–712

    2022

  30. [38]

    SCLIP: Rethinking self-attention for dense vision-language inference,

    F. Wang, J. Mei, and A. Yuille, “SCLIP: Rethinking self-attention for dense vision-language inference,” inComputer Vision – ECCV 2024, A. Leonardis, E. Ricci, S. Roth, O. Russakovsky, T. Sattler, and G. Varol, Eds. Cham: Springer Nature Switzerland, 2025, pp. 315–332

    2024

  31. [39]

    ProxyCLIP: Proxy attention improves CLIP for open-vocabulary segmentation,

    M. Lan, C. Chen, Y . Ke, X. Wang, L. Feng, and W. Zhang, “ProxyCLIP: Proxy attention improves CLIP for open-vocabulary segmentation,” in Computer Vision – ECCV 2024, A. Leonardis, E. Ricci, S. Roth, O. Russakovsky, T. Sattler, and G. Varol, Eds. Cham: Springer Nature Switzerla...

    2024

  32. [40]

    Semantic-CD: Re- mote sensing image semantic change detection towards open-vocabulary setting,

    Y . Zhu, L. Li, K. Chen, C. Liu, F. Zhou, and Z. Shi, “Semantic-CD: Re- mote sensing image semantic change detection towards open-vocabulary setting,” inIGARSS 2025 - 2025 IEEE International Geoscience and Remote Sensing Symposium, 2025, pp. 6388–6392

    2025

  33. [41]

    UniVCD: A new method for unsupervised change detection in the open-vocabulary era,

    Z. Zhu and B. Yang, “UniVCD: A new method for unsupervised change detection in the open-vocabulary era,” 2025, arXiv:2512.13089

    2025

  34. [42]

    OpenDPR: Open-vocabulary change detection via vision-centric diffusion-guided prototype retrieval for remote sensing imagery,

    Q. Guo, J. Wang, Y . Liu, and Y . Zhong, “OpenDPR: Open-vocabulary change detection via vision-centric diffusion-guided prototype retrieval for remote sensing imagery,” inProceedings of the IEEE/CVF Confer- ence on Computer Vision and Pattern Recognition (CVPR), June 2026, pp....

    2026

  35. [43]

    Seg2Change: Adapting open- vocabulary semantic segmentation model for remote sensing change detection,

    Y . Su, Y . Song, J. Chen, and Z. Wen, “Seg2Change: Adapting open- vocabulary semantic segmentation model for remote sensing change detection,” 2026, arXiv:2604.11231

    2026 arXiv

  36. [44]

    AdaptOVCD: Training-free open-vocabulary remote sensing change detection via adaptive information fusion,

    M. Dou, S. Qiu, M. Hu, Y . Chen, H. Ye, X. Liao, and Z. Sun, “AdaptOVCD: Training-free open-vocabulary remote sensing change detection via adaptive information fusion,” 2026, arXiv:2602.06529

    2026

  37. [45]

    Om- niOVCD: Streamlining open-vocabulary change detection with SAM 3,

    X. Zhang, D. Li, Y . Xia, X. Dong, H. Yu, J. Wang, and Q. Li, “Om- niOVCD: Streamlining open-vocabulary change detection with SAM 3,” 2026, arXiv:2601.13895

    2026 arXiv

  38. [46]

    CoRegOVCD: Consistency-regularized open-vocabulary change detection,

    W. Tang, H. Sun, Z. Li, Y . Wang, and F. Zhang, “CoRegOVCD: Consistency-regularized open-vocabulary change detection,” 2026, arXiv:2604.02160

    2026

  39. [47]

    MemOVCD: Training-free open-vocabulary change detection via cross- temporal memory reasoning and global-local adaptive rectification,

    Z. Kuang, H. Chang, B. Liang, H. Wang, L. He, F. Li, and H. Bi, “MemOVCD: Training-free open-vocabulary change detection via cross- temporal memory reasoning and global-local adaptive rectification,” 2026, arXiv:2604.26774

    2026 arXiv

  40. [48]

    ReA-OVCD: Reliability- aware open-vocabulary change detection via semantic and spatial refine- ment,

    H. Zhu, H. Chen, B. Du, S. Liu, and Q. Liu, “ReA-OVCD: Reliability- aware open-vocabulary change detection via semantic and spatial refine- ment,” 2026, arXiv:2606.20032

    2026 arXiv

  41. [49]

    Changer: Feature interaction is what you need for change detection,

    S. Fang, K. Li, and Z. Li, “Changer: Feature interaction is what you need for change detection,”IEEE Transactions on Geoscience and Remote Sensing, vol. 61, pp. 1–11, 2023

    2023

  42. [50]

    Learning from noisy pseudo-labels for all-weather land cover mapping,

    W. Liu, Z. Wang, X. Guo, P. Duan, X. Kang, and S. Li, “Learning from noisy pseudo-labels for all-weather land cover mapping,” inIGARSS 2025 - 2025 IEEE International Geoscience and Remote Sensing Sym- posium, 2025, pp. 219–222

    2025

  43. [51]

    Symmetric cross entropy for robust learning with noisy labels,

    Y . Wang, X. Ma, Z. Chen, Y . Luo, J. Yi, and J. Bailey, “Symmetric cross entropy for robust learning with noisy labels,” in2019 IEEE/CVF International Conference on Computer Vision (ICCV), 2019, pp. 322– 330

    2019

  44. [52]

    The Lovasz-Softmax loss: A tractable surrogate for the optimization of the intersection-over- union measure in neural networks,

    M. Berman, A. R. Triki, and M. B. Blaschko, “The Lovasz-Softmax loss: A tractable surrogate for the optimization of the intersection-over- union measure in neural networks,” in2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2018, pp. 4413–4421

    2018

  45. [53]

    V-Net: Fully convolutional neural networks for volumetric medical image segmentation,

    F. Milletari, N. Navab, and S.-A. Ahmadi, “V-Net: Fully convolutional neural networks for volumetric medical image segmentation,” in2016 Fourth International Conference on 3D Vision (3DV), 2016, pp. 565– 571

    2016

  46. [54]

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

    H. Chen and Z. Shi, “A spatial-temporal attention-based method and a new dataset for remote sensing image change detection,”Remote Sensing, vol. 12, no. 10, p. 1662, 2020

    2020

  47. [55]

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

    S. Ji, S. Wei, and M. Lu, “Fully convolutional networks for multisource building extraction from an open aerial and satellite imagery data set,” IEEE Transactions on Geoscience and Remote Sensing, vol. 57, no. 1, pp. 574–586, 2019

    2019

  48. [56]

    S2Looking: A satellite side-looking dataset for building change detection,

    L. Shen, Y . Lu, H. Chen, H. Wei, D. Xie, J. Yue, R. Chen, S. Lv, and B. Jiang, “S2Looking: A satellite side-looking dataset for building change detection,”Remote Sensing, vol. 13, no. 24, p. 5094, 2021

    2021

  49. [57]

    Asymmetric siamese networks for semantic change detection in aerial images,

    K. Yang, G.-S. Xia, Z. Liu, B. Du, W. Yang, M. Pelillo, and L. Zhang, “Asymmetric siamese networks for semantic change detection in aerial images,”IEEE Transactions on Geoscience and Remote Sensing, vol. 60, pp. 1–18, 2022

    2022

  50. [58]

    BiSAM-CD: Zero-shot remote sensing change detection via bidirectional temporal memory in SAM2,

    Y . Qin, J. Chen, C. Wang, and C. Pan, “BiSAM-CD: Zero-shot remote sensing change detection via bidirectional temporal memory in SAM2,” IEEE Transactions on Geoscience and Remote Sensing, vol. 63, pp. 1– 12, 2025

    2025

Pith tools

Reviewed August 16, 2026 · model on record in the stance chip above.