The reviewed record of science sign in
Pith

arxiv: 2606.03114 · v1 · pith:MVTMQVLS · submitted 2026-06-02 · cs.CV

FAF-CD: Frequency-Aware Fusion for Change Detection under Imperfect Multimodal Remote Sensing

Yufan Wang , Sokratis Makrogiannis , Chandra Kambhamettu This is my paper

Reviewed by Pith T0 review T1 audit T2 compute T3 formal T4 kernel 2026-06-28 10:45 UTCgrok-4.3pith:MVTMQVLSrecord.jsonopen to challenge →

classification cs.CV
keywords change detectionremote sensingmultimodal fusionfrequency analysisdeformable alignmentEO-SARdisaster mapping
0
0 comments X

The pith

Frequency-aware tri-branch fusion improves change detection in mismatched multimodal remote sensing imagery while lowering compute cost.

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

The paper presents FAF-CD to address change detection when pre- and post-event images come from different sensors or acquisition times, where lighting, seasonal, or modality differences can mimic actual structural damage. It pairs a DINOv3-pretrained ConvNeXt encoder with a VMamba decoder and introduces a rectification-aware tri-branch fusion module that aligns images deformably and compares them via Fourier and Haar-wavelet transforms under adaptive gating. This setup aggregates scale-consistent cues to separate nuisance shifts from real changes. The approach reports higher clean and perturbed scores on heterogeneous EO-SAR validation data than prior methods, plus strong results on standard optical benchmarks, all at reduced computational expense. A sympathetic reader would care because reliable automated monitoring of disasters requires working with the imperfect, asynchronous data that real satellite streams provide.

Core claim

FAF-CD is a frequency-aware hybrid framework with a DINOv3-pretrained ConvNeXt encoder and linear-complexity VMamba decoder whose rectification-aware tri-branch fusion module performs deformable spatial alignment together with Fourier and Haar-wavelet domain comparisons through adaptive gating to aggregate complementary cues across scales, yielding improved tc-mIoU/tc-mAP on BRIGHT heterogeneous EO-SAR validation, best average perturbed cIoU/cF1 on LEVIR-CD and WHU-CD under pseudo-change stress tests, and an approximate 24 GFLOPs cost reduction relative to NeXt2Former-CD.

What carries the argument

The rectification-aware tri-branch fusion module, which combines deformable spatial alignment with Fourier and Haar-wavelet comparisons under adaptive gating to isolate structural damage from nuisance variations.

If this is right

  • Heterogeneous EO-SAR adaptation becomes more accurate for disaster mapping without added computational overhead.
  • The same framework generalizes to binary optical change detection and maintains top perturbed scores on LEVIR-CD and WHU-CD.
  • Lower GFLOPs cost enables deployment on resource-constrained monitoring pipelines while preserving or improving accuracy.
  • Adaptive gating across scales supports handling of asynchronous and cross-sensor observations that standard fusion methods struggle with.

Where Pith is reading between the lines

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

  • The frequency comparison strategy could transfer to other registration or alignment problems where modality shifts dominate.
  • Testing the module on additional sensor pairs beyond EO-SAR would clarify whether the gains are modality-specific.
  • Real-time streaming scenarios with continuous modality drift might expose whether the current gating mechanism scales without retraining.
  • Replacing the VMamba decoder with alternative efficient backbones could further trade accuracy against speed in operational settings.

Load-bearing premise

The frequency-aware tri-branch fusion components, rather than the pretrained encoder choice or dataset particulars, are what enable separation of nuisance variations from structural damage.

What would settle it

An ablation that disables the Fourier and Haar-wavelet comparison branches and re-runs the BRIGHT validation suite under the same pseudo-change perturbations, checking whether the perturbed cIoU/cF1 advantage disappears.

Figures

Figures reproduced from arXiv: 2606.03114 by Chandra Kambhamettu, Sokratis Makrogiannis, Yufan Wang.

Figure 1
Figure 1. Figure 1: Overview of the proposed end-to-end change detection framework in the binary optical CD setting. Given a bi-temporal image [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Detailed architecture of the Frequency-Aware Fusion [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Qualitative results on the LEVIR-CD [3] and WHU-CD [15] datasets. White represents true positives, black represents true negatives, green represents false positives and red represents false negatives. ancy modeling, and Haar wavelet-based boundary cues. An adaptive gating mechanism aggregates these complemen￾tary representations across scales, and a VMamba-based decoder enables efficient global reasoning f… view at source ↗
Figure 4
Figure 4. Figure 4: Severity sweep for the pseudo-change perturbations. The same clean WHU-CD [ [PITH_FULL_IMAGE:figures/full_fig_p012_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Qualitative BRIGHT validation comparison [ [PITH_FULL_IMAGE:figures/full_fig_p013_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Internal structures of the frequency-domain fusion branches. [PITH_FULL_IMAGE:figures/full_fig_p015_6.png] view at source ↗
read the original abstract

Remote sensing change detection for real-world monitoring often relies on imperfect heterogeneous observations, where pre- and post-event images may be asynchronous, cross-sensor, or affected by illumination, seasonal, and modality shifts. This setting is especially challenging for EO-SAR disaster mapping, where nuisance variation can resemble structural damage. We propose FAF-CD, a frequency-aware hybrid framework with a DINOv3-pretrained ConvNeXt encoder and a linear-complexity VMamba-based decoder. Its rectification-aware tri-branch fusion module combines deformable spatial alignment with Fourier and Haar-wavelet comparisons, using adaptive gating to aggregate complementary cues across scales. On BRIGHT validation, a matched heterogeneous EO-SAR adaptation improves clean and perturbed tc-mIoU/tc-mAP over NeXt2Former-CD. FAF-CD also generalizes to binary optical CD, achieving 0.924 cF1 on LEVIR-CD and 0.955 cF1 on WHU-CD, and obtains the best average perturbed cIoU/cF1 on both binary datasets among M-CD and NeXt2Former-CD under pseudo-change-aligned stress tests. It further reduces cost by approximately 24 GFLOPs relative to NeXt2Former-CD while maintaining or improving accuracy.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The paper proposes FAF-CD, a hybrid change detection framework for imperfect multimodal remote sensing (especially EO-SAR disaster mapping). It combines a DINOv3-pretrained ConvNeXt encoder, a linear-complexity VMamba decoder, and a rectification-aware tri-branch fusion module that performs deformable spatial alignment together with Fourier and Haar-wavelet comparisons under adaptive gating. The central empirical claims are improved clean and perturbed tc-mIoU/tc-mAP on BRIGHT validation relative to NeXt2Former-CD, state-of-the-art cF1 scores of 0.924 on LEVIR-CD and 0.955 on WHU-CD, best average perturbed cIoU/cF1 under pseudo-change stress tests, and an approximately 24 GFLOPs reduction in cost.

Significance. If the reported gains are shown to arise specifically from the frequency-aware fusion rather than the choice of backbone, the work would supply a concrete, efficiency-aware recipe for distinguishing nuisance variation from structural change in heterogeneous remote-sensing settings. The combination of frequency-domain cues with deformable alignment and the reported computational saving constitute a practical contribution to multimodal CD.

major comments (2)
  1. [Abstract and §4] Abstract and §4 (Experiments): the manuscript attributes the reported gains on BRIGHT (clean/perturbed tc-mIoU/tc-mAP) and the binary datasets (0.924/0.955 cF1, best perturbed averages) to the rectification-aware tri-branch fusion. No ablation is described that holds the DINOv3-ConvNeXt encoder and VMamba decoder fixed while removing or replacing the deformable-alignment + Fourier/Haar-wavelet + gating module; without this isolation the load-bearing claim that the frequency-aware fusion is what distinguishes nuisance variation cannot be evaluated.
  2. [Abstract] Abstract: the performance numbers are stated without error bars, number of runs, or statistical significance tests. Given that the central claim rests on outperformance under both clean and perturbed conditions, the absence of these controls makes it impossible to judge whether the observed margins are reliable.
minor comments (1)
  1. [Abstract] The abstract mentions “pseudo-change-aligned stress tests” but does not define the perturbation protocol or the alignment procedure; a concise description or reference to the exact protocol used would improve reproducibility.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive and detailed feedback. Below we address each major comment point by point, indicating where revisions will be made.

read point-by-point responses

  1. Referee: [Abstract and §4] Abstract and §4 (Experiments): the manuscript attributes the reported gains on BRIGHT (clean/perturbed tc-mIoU/tc-mAP) and the binary datasets (0.924/0.955 cF1, best perturbed averages) to the rectification-aware tri-branch fusion. No ablation is described that holds the DINOv3-ConvNeXt encoder and VMamba decoder fixed while removing or replacing the deformable-alignment + Fourier/Haar-wavelet + gating module; without this isolation the load-bearing claim that the frequency-aware fusion is what distinguishes nuisance variation cannot be evaluated.

    Authors: We agree that the manuscript would be strengthened by an ablation that isolates the tri-branch fusion module while freezing the DINOv3-ConvNeXt encoder and VMamba decoder. The existing comparisons are against NeXt2Former-CD, which differs in multiple components. In the revision we will add a controlled ablation that replaces the deformable-alignment + Fourier/Haar + gating module with a baseline fusion (e.g., simple concatenation followed by a 1×1 convolution) while keeping the encoder and decoder identical, and report the resulting clean and perturbed metrics on BRIGHT. revision: yes

  2. Referee: [Abstract] Abstract: the performance numbers are stated without error bars, number of runs, or statistical significance tests. Given that the central claim rests on outperformance under both clean and perturbed conditions, the absence of these controls makes it impossible to judge whether the observed margins are reliable.

    Authors: We acknowledge that the reported point estimates lack error bars and statistical tests. All experiments used a single fixed random seed for reproducibility; additional runs with varied seeds were not performed owing to the computational cost of training on the full BRIGHT and binary-CD suites. In the revision we will (i) explicitly state the single-seed protocol, (ii) add a limitations paragraph noting the absence of multi-run statistics, and (iii) if compute permits, rerun the final models on LEVIR-CD and WHU-CD with three seeds and report mean ± std for cF1. revision: partial

Circularity Check

0 steps flagged

No circularity; empirical performance claims only

full rationale

The provided abstract and context describe an empirical model proposal (FAF-CD with tri-branch fusion) evaluated via measured metrics on BRIGHT, LEVIR-CD, and WHU-CD datasets against baselines like NeXt2Former-CD. No equations, derivations, or self-citations are shown that reduce any claimed result to its own inputs by construction. Performance figures (e.g., 0.924 cF1) are presented as observed outcomes, not predictions forced by parameter fitting or definitional equivalence. The central claims rest on external dataset benchmarks rather than internal self-reference.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract provides no information on free parameters, axioms, or invented entities.

pith-pipeline@v0.9.1-grok · 5762 in / 1075 out tokens · 32891 ms · 2026-06-28T10:45:48.595762+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

34 extracted references · 9 canonical work pages · 3 internal anchors

  1. [1]

    Wele Gedara Chaminda Bandara and Vishal M. Patel. A Transformer-Based Siamese Network for Change Detection. InIGARSS 2022 - 2022 IEEE International Geoscience and Remote Sensing Symposium, pages 207–210, 2022. ISSN: 2153-7003

    2022

  2. [2]

    Wele Gedara Chaminda Bandara, Nithin Gopalakrishnan Nair, and Vishal M. Patel. DDPM-CD: Denoising Diffusion Probabilistic Models as Feature Extractors for Change De- tection, 2024. arXiv:2206.11892 [cs]

    work page arXiv 2024

  3. [3]

    A Spatial-Temporal Attention- Based Method and a New Dataset for Remote Sensing Image Change Detection.Remote Sensing, 12(10):1662, 2020

    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

    2020

  4. [4]

    Remote Sensing Image Change Detection With Transformers.IEEE Transac- tions on Geoscience and Remote Sensing, 60:1–14, 2022

    Hao Chen, Zipeng Qi, and Zhenwei Shi. Remote Sensing Image Change Detection With Transformers.IEEE Transac- tions on Geoscience and Remote Sensing, 60:1–14, 2022

    2022

  5. [5]

    ChangeMamba: Remote Sensing Change Detection With Spatiotemporal State Space Model.IEEE Transactions on Geoscience and Remote Sensing, 62:1–20, 2024

    Hongruixuan Chen, Jian Song, Chengxi Han, Junshi Xia, and Naoto Yokoya. ChangeMamba: Remote Sensing Change Detection With Spatiotemporal State Space Model.IEEE Transactions on Geoscience and Remote Sensing, 62:1–20, 2024

    2024

  6. [6]

    Hongruixuan Chen, Jian Song, Oliver Dietrich, Clifford Broni-Bediako, Weikang Xuan, Junjue Wang, Xuanlong Shao, Yinhe Wei, Junshi Xia, Cuiling Lan, Konrad Schindler, and Naoto Yokoya. BRIGHT: A globally distributed multi- modal building damage assessment dataset with very-high- resolution for all-weather disaster response.Earth System Science Data, 17:621...

    2025

  7. [7]

    RSMamba: Remote Sens- ing Image Classification With State Space Model.IEEE Geoscience and Remote Sensing Letters, 21:1–5, 2024

    Keyan Chen, Bowen Chen, Chenyang Liu, Wenyuan Li, Zhengxia Zou, and Zhenwei Shi. RSMamba: Remote Sens- ing Image Classification With State Space Model.IEEE Geoscience and Remote Sensing Letters, 21:1–5, 2024

    2024

  8. [8]

    ChessMamba: Structure-Aware Interleav- ing of State Spaces for Change Detection in Remote Sensing Images, 2025

    Lei Ding, Tong Liu, Xuanguang Liu, Xiangyun Liu, Haitao Guo, and Jun Lu. ChessMamba: Structure-Aware Interleav- ing of State Spaces for Change Detection in Remote Sensing Images, 2025. arXiv:2511.19882 [cs]

    work page arXiv 2025

  9. [9]

    An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale

    Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Syl- vain Gelly, Jakob Uszkoreit, and Neil Houlsby. An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale, 2021. arXiv:2010.11929 [cs]

    work page internal anchor Pith review Pith/arXiv arXiv 2021

  10. [10]

    SNUNet- CD: A Densely Connected Siamese Network for Change De- tection of VHR Images.IEEE Geoscience and Remote Sens- ing Letters, 19:1–5, 2022

    Sheng Fang, Kaiyu Li, Jinyuan Shao, and Zhe Li. SNUNet- CD: A Densely Connected Siamese Network for Change De- tection of VHR Images.IEEE Geoscience and Remote Sens- ing Letters, 19:1–5, 2022

    2022

  11. [11]

    Efficient Multi-scale Network with Learnable Discrete Wavelet Transform for Blind Motion Deblurring

    Xin Gao, Tianheng Qiu, Xinyu Zhang, Hanlin Bai, Kang Liu, Xuan Huang, Hu Wei, Guoying Zhang, and Huaping Liu. Efficient Multi-scale Network with Learnable Discrete Wavelet Transform for Blind Motion Deblurring. InCVPR, pages 2733–2742, 2024

    2024

  12. [12]

    Mamba: Linear-Time Sequence Modeling with Selective State Spaces

    Albert Gu and Tri Dao. Mamba: Linear-Time Sequence Modeling with Selective State Spaces. 2024

    2024

  13. [13]

    FD- FENet: Cropland Change Detection in Remote Sensing Im- ages Based on Frequency Domain Feature Exchange and Multiscale Feature Enhancement.Remote Sensing, 18(1): 128, 2026

    Yujiang He, Yurong Qian, Xin Wang, Lu Bai, Yuanxu Wang, Hanming Wei, Xingke Huang, Junyi Lv, Xin Yang, Min Duan, Weijun Gong, and Madina Mansurova. FD- FENet: Cropland Change Detection in Remote Sensing Im- ages Based on Frequency Domain Feature Exchange and Multiscale Feature Enhancement.Remote Sensing, 18(1): 128, 2026

    2026

  14. [14]

    3D- SSM: A Novel 3D Selective Scan Module for Remote Sens- ing Change Detection, 2025

    Rui Huang, Jincheng Zeng, Sen Gao, and Yan Xing. 3D- SSM: A Novel 3D Selective Scan Module for Remote Sens- ing Change Detection, 2025. arXiv:2506.19263 [cs]

    work page arXiv 2025

  15. [15]

    Shunping Ji, Shiqing Wei, and Meng Lu. Fully Convolu- tional Networks for Multisource Building Extraction From an Open Aerial and Satellite Imagery Data Set.IEEE Trans- actions on Geoscience and Remote Sensing, 57(1):574–586, 2019

    2019

  16. [16]

    Yi Liu, Chao Pang, Zongqian Zhan, Xiaomeng Zhang, and Xue Yang. Building Change Detection for Remote Sensing Images Using a Dual-Task Constrained Deep Siamese Con- volutional Network Model.IEEE Geoscience and Remote Sensing Letters, 18(5):811–815, 2021

    2021

  17. [17]

    VMamba: Visual State Space Model

    Yue Liu, Yunjie Tian, Yuzhong Zhao, Hongtian Yu, Lingxi Xie, Yaowei Wang, Qixiang Ye, Jianbin Jiao, and Yunfan Liu. VMamba: Visual State Space Model. 2024

    2024

  18. [18]

    A ConvNet for the 2020s

    Zhuang Liu, Hanzi Mao, Chao-Yuan Wu, Christoph Feicht- enhofer, Trevor Darrell, and Saining Xie. A ConvNet for the 2020s. InCVPR, pages 11976–11986, 2022

    2022

  19. [19]

    Can Lu, Feng Wang, Zhen Wang, Nan Xu, Zhuhong You, and De-Shuang Huang. BWFNet: Bitemporal Wavelet Fre- quency Network for Change Detection in High-Resolution Remote Sensing Images.IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 18:25562– 25582, 2025

    2025

  20. [20]

    Paranjape, Celso de Melo, and Vishal M

    Jay N. Paranjape, Celso de Melo, and Vishal M. Patel. A Mamba-based Siamese Network for Remote Sensing Change Detection, 2024. arXiv:2407.06839 [cs]

    work page arXiv 2024

  21. [21]

    Oriane Sim ´eoni, Huy V . V o, Maximilian Seitzer, Fed- erico Baldassarre, Maxime Oquab, Cijo Jose, Vasil Khali- dov, Marc Szafraniec, Seungeun Yi, Micha ¨el Ramamon- jisoa, Francisco Massa, Daniel Haziza, Luca Wehrstedt, Jianyuan Wang, Timoth´ee Darcet, Th´eo Moutakanni, Leonel Sentana, Claire Roberts, Andrea Vedaldi, Jamie Tolan, John Brandt, Camille Co...

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  22. [22]

    FA-Mamba: a Mamba- based approach for remote sensing change detection

    Qihang Sun, Yingchao Liu, Longfei Li, Huifang Sun, Lingyun Zhao, and Chunpeng Tian. FA-Mamba: a Mamba- based approach for remote sensing change detection. In Third Asia Conference on Computer Vision, Image Process- ing, and Pattern Recognition (CVIPPR 2025), pages 55–61. SPIE, 2025

    2025

  23. [23]

    SAMBA: Synthetic Data-Augmented Mamba-Based Change Detection Algorithm Using KOMPSAT-3A Imagery.Korean Journal of Remote Sensing, 40:1505–1521, 2024

    Rogelio Tobias, Bae Sejeong, Hwanhee Cho, and Jungho Im. SAMBA: Synthetic Data-Augmented Mamba-Based Change Detection Algorithm Using KOMPSAT-3A Imagery.Korean Journal of Remote Sensing, 40:1505–1521, 2024

    2024

  24. [24]

    Change Detection Mamba with Boundary-Specific Supervision

    Guangxing Wang, Gong Cheng, and Junwei Han. Change Detection Mamba with Boundary-Specific Supervision. IEEE Transactions on Circuits and Systems for Video Tech- nology, pages 1–1, 2025

    2025

  25. [25]

    NeXt2Former-CD: Efficient Remote Sensing Change Detection with Modern Vision Architectures, 2026

    Yufan Wang, Sokratis Makrogiannis, and Chandra Kamb- hamettu. NeXt2Former-CD: Efficient Remote Sensing Change Detection with Modern Vision Architectures, 2026. arXiv:2602.18717 [cs] version: 1. 9

    work page arXiv 2026

  26. [26]

    Multiscale Spa- tial Frequency Fusion and Prior Change Guidance Network for Remote Sensing Change Detection.IEEE Transactions on Instrumentation and Measurement, 74:1–13, 2025

    Hongguang Wei, Yuan Liu, Yueran Ma, Dongdong Pang, Yuanxin Ye, Xiubao Sui, and Qian Chen. Multiscale Spa- tial Frequency Fusion and Prior Change Guidance Network for Remote Sensing Change Detection.IEEE Transactions on Instrumentation and Measurement, 74:1–13, 2025

    2025

  27. [27]

    FDA: Fourier Domain Adaptation for Semantic Segmentation

    Yanchao Yang and Stefano Soatto. FDA: Fourier Domain Adaptation for Semantic Segmentation. InCVPR, pages 4085–4095, 2020

    2020

  28. [28]

    Wave-ViT: Unifying Wavelet and Transformers for Vi- sual Representation Learning

    Ting Yao, Yingwei Pan, Yehao Li, Chong-Wah Ngo, and Tao Mei. Wave-ViT: Unifying Wavelet and Transformers for Vi- sual Representation Learning. InComputer Vision – ECCV 2022, pages 328–345, Cham, 2022. Springer Nature Switzer- land

    2022

  29. [29]

    Chenxiao Zhang, Peng Yue, Deodato Tapete, Liangcun Jiang, Boyi Shangguan, Li Huang, and Guangchao Liu. A deeply supervised image fusion network for change de- tection in high resolution bi-temporal remote sensing im- ages.ISPRS Journal of Photogrammetry and Remote Sens- ing, 166:183–200, 2020

    2020

  30. [30]

    CDMamba: Incorporating Local Clues Into Mamba for Remote Sensing Image Binary Change Detection.IEEE Transactions on Geoscience and Remote Sensing, 63:1–16, 2025

    Haotian Zhang, Keyan Chen, Chenyang Liu, Hao Chen, Zhengxia Zou, and Zhenwei Shi. CDMamba: Incorporating Local Clues Into Mamba for Remote Sensing Image Binary Change Detection.IEEE Transactions on Geoscience and Remote Sensing, 63:1–16, 2025

    2025

  31. [31]

    CMSNet: A SAM-Enhanced CNN–Mamba Framework for Damaged Building Change Detection in Re- mote Sensing Imagery.Remote Sensing, 17(23):3913, 2025

    Jianli Zhang, Liwei Tao, Wenbo Wei, Pengfei Ma, and Mengdi Shi. CMSNet: A SAM-Enhanced CNN–Mamba Framework for Damaged Building Change Detection in Re- mote Sensing Imagery.Remote Sensing, 17(23):3913, 2025

    2025

  32. [32]

    Xiaoyang Zhang, Guodong Fan, Guang-Yong Chen, Zhen Hua, Jinjiang Li, Min Gan, and C. L. Philip Chen. Wavelet- Guided Dual-Frequency Encoding for Remote Sensing Change Detection, 2025. arXiv:2508.05271 [cs]

    work page arXiv 2025

  33. [33]

    Deformable DETR: Deformable Transformers for End-to-End Object Detection

    Xizhou Zhu, Weijie Su, Lewei Lu, Bin Li, Xiaogang Wang, and Jifeng Dai. Deformable DETR: Deformable Transformers for End-to-End Object Detection, 2021. arXiv:2010.04159 [cs]. 10 FAF-CD: Frequency-Aware Fusion for Change Detection under Imperfect Multimodal Remote Sensing Supplementary Material

    work page internal anchor Pith review Pith/arXiv arXiv 2021

  34. [34]

    Protocol We test robustness to pseudo-changes by perturbing one im- age in each bi-temporal test pair and leaving the other image unchanged

    Pseudo-change Robustness Study 6.1. Protocol We test robustness to pseudo-changes by perturbing one im- age in each bi-temporal test pair and leaving the other image unchanged. This isolates asymmetric acquisition effects, such as illumination, color balance, haze, and cast shadows, that can resemble structural change. On LEVIR-CD [3] and WHU-CD [15], we ...

    2026