Pith. sign in

REVIEW 3 major objections 5 minor 50 references

A new construction change-detection dataset from synthetic LiDAR views shows that standard 2D methods fail on oblique, viewpoint-diverse scenes.

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

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

T0 review · grok-4.5

2026-07-12 01:39 UTC pith:HGJKL3T4

load-bearing objection Solid dataset paper that fills a real gap; ranking inversions are real, but the real-world claim still rests on unquantified synthetic fidelity. the 3 major comments →

arxiv 2607.03553 v1 pith:HGJKL3T4 submitted 2026-07-03 cs.CV cs.RO

iVISION-2DCD: A Long-Term Change Detection Dataset for Large-Scale Outdoor Construction Monitoring

Dayou Mao , Yuchen Lin , Ashkan Ebadi , John Zelek , Alexander Wong , Yuhao Chen This is my paper
classification cs.CV cs.RO
keywords 2D change detectionconstruction monitoringsynthetic novel-view synthesisLiDAR point cloudsviewpoint-robust CDUAV surveyingbenchmark dataset
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

Construction progress monitoring needs change detection that works from any camera angle, not just straight-down satellite or aerial views. Real drone flights on active sites cannot safely collect enough diverse viewpoints or perfectly aligned image pairs, and no public 2D change-detection benchmark has filled that gap. This paper introduces iVISION-2DCD: photorealistic image pairs and change labels synthesized from dense multi-temporal LiDAR of a large outdoor construction site, using novel-view sampling and semi-automated 3D semantic labeling. Benchmarks of 18 established algorithms show large accuracy drops and ranking reversals relative to classic nadir datasets, establishing that viewpoint-robust construction change detection is a distinct and harder problem.

Core claim

iVISION-2DCD formally defines viewpoint-robust 2D change detection for long-term outdoor construction monitoring and shows, through systematic evaluation of 18 state-of-the-art methods, that the dataset creates novel challenges: every method suffers large mean-IoU drops and dramatic ranking inversions compared with established nadir-view benchmarks.

What carries the argument

Synthetic novel-view generation from dense LiDAR (vertical, mirror, and forward camera sampling plus coverage-optimized rendering) combined with bi-directional 3D semantic comparison that yields symmetric 2D change masks.

Load-bearing premise

That synthetic images rendered from discrete LiDAR points and change labels derived from semi-automated 3D semantics are faithful enough to real RGB captures for the measured performance gaps to transfer to real deployment.

What would settle it

Train and test the same 18 methods on a held-out set of real, non-synthetic RGB image pairs captured at matching construction viewpoints and timestamps; if the large mIoU drops and ranking inversions disappear, the synthetic fidelity claim fails.

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

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The paper introduces iVISION-2DCD, a synthetically generated 2D change-detection (2DCD) benchmark for large-scale outdoor construction monitoring. Dense multi-temporal LiDAR point clouds from UAV missions are rendered into photorealistic RGB images via novel-view synthesis (vertical/mirror/forward camera sampling, 1/8-resolution point projection) and paired with change labels obtained by semi-automated 3D semantic segmentation (DJI Modify + manual refinement) followed by bi-directional 3D comparison and union into 2D masks. The authors argue that existing 2DCD datasets are limited to nadir views and short temporal spans, and that iVISION-2DCD formally defines viewpoint-robust construction 2DCD while preserving real-world difficulties (snow, invisible terrain change, dynamic objects, RGB projection errors). A controlled benchmark of 18 SOTA 2DCD methods on six datasets (Table 1) shows large mIoU drops and ranking inversions on iVISION-2DCD relative to established aerial/satellite benchmarks, which the authors interpret as evidence of novel research challenges.

Significance. If the synthetic images and labels are sufficiently faithful to real RGB captures, the work fills a clear gap: there is no open, multi-view, long-horizon 2DCD benchmark at the intersection of construction automation and computer vision. The generation pipeline is concrete (§3), challenging cases are explicitly retained (§4.1), and the 18-method comparison under a shared training protocol (§5) is a useful baseline resource. The ranking inversions in Table 1 are interesting even as a synthetic stress test. Strengths that should be credited include the systematic camera-synthesis cascade, the bi-directional 3D-to-2D label construction that restores symmetry for standard 2DCD training, and the decision to keep hard real-world artifacts rather than sanitize them. The contribution is therefore potentially high for robotics and construction-monitoring communities, contingent on clearer evidence that the observed gaps transfer beyond the synthetic domain.

major comments (3)
  1. [§3.1.2, §3.2, §5.2, Table 1, Fig. 3, Fig. 7] The central claim (abstract, §5.2, Table 1) that large mIoU drops and ranking inversions prove iVISION-2DCD poses novel challenges for real-world viewpoint-robust construction monitoring rests on an untested fidelity premise. Images are rendered from discrete LiDAR at 1/8 resolution with 1-pixel points (~97% coverage; §3.1.2, Fig. 3), colors inherit systematic projection errors already present in the reconstruction (crane/pole colors onto ground; Fig. 7), and change masks are the bi-directional union of 3D semantic differences after DJI Modify + ignore filtering (§3.2.2). No quantitative real-vs-synthetic metric (PSNR/SSIM/LPIPS on co-registered real RGB, or label agreement on any real bi-temporal pair) is reported. Without such evidence, the ranking inversions may partly reflect rendering sparsity, color bleed, and ignore-class policy rather than the intended construction/viewpoint diff
  2. [§1, abstract, §4.2, §5] The manuscript markets long-term coverage ("over 50 acquisition sessions spanning one full year," seasonal winter conditions; §1 distinguishing dimensions and abstract framing) while the released benchmark and all experiments use only the Nov 27 2024–Mar 12 2025 window (17 scenes, 16 consecutive pairs; §4.2, §5). The authors note that the work is ongoing and that raw data through Aug 2025 is public, but the paper-level claims and dataset statistics still overstate what is actually evaluated. Either restrict the long-term claims to the 4-month subset used for Table 1, or include the fuller temporal span in the reported benchmark so that seasonal and multi-month dynamics are part of the evidence.
  3. [§4.2, §5.1, Table 1] Dataset construction and experimental protocol leave load-bearing details underspecified for reproducibility of Table 1. §4.2 states 4800 synthetic cameras from 16 pairs but does not define train/val/test splits, whether pairs are scene-disjoint, how ignore pixels are handled in mIoU, or whether the same synthetic cameras appear in both temporal directions. §5.1 removes random rotation only for iVISION-2DCD and uses inverse class weighting, yet does not report class frequencies or the fraction of ignore pixels. These choices can materially affect rankings; they should be stated explicitly (and ideally released with the dataset) so that the performance gaps can be attributed to the intended domain rather than protocol differences.
minor comments (5)
  1. [title page / headers] Title and running headers use inconsistent spacing/capitalization ("Long-TermChangeDetection", "LARGE-SCALEOUTDOORCONSTRUCTION"). Normalize throughout.
  2. [abstract, §1] Abstract and §1 contain several grammatical issues ("change detection need to be performed", "UA Vs", "bi-temporally aligned"). A careful copy-edit pass would improve readability.
  3. [Fig. 1] Fig. 1 caption says "not consecutive" scenes but the change maps are described as "compared from immediate previous scene"; clarify temporal adjacency.
  4. [§2.2] Related work on single-temporal CD and 3DCD is useful but could more explicitly position why the authors still evaluate only bi-temporal 2D methods rather than also releasing a 3DCD track from the same point clouds.
  5. [abstract / conclusion] Project page URL is given; ensure the camera-generation code, ignore-class policy, and exact train/test lists will be released with the camera-ready version so others can reproduce Table 1.

Circularity Check

0 steps flagged

No circularity: synthetic dataset construction and empirical SOTA benchmarks are independent; change labels derive from 3D semantics, not from the 2DCD models under test.

full rationale

This is a dataset-and-benchmark paper with no theoretical derivation chain, no fitted parameters renamed as predictions, and no equations that reduce a claimed result to its own inputs. Novel-view synthesis (vertical/mirror/forward camera sampling, 1/8-resolution rendering) and semi-automated 3D semantic labeling (DJI Modify + manual refinement + bi-directional union for symmetric change masks) produce the images and ground-truth independently of any of the 18 evaluated 2DCD networks. Table 1 mIoU drops and ranking inversions are purely empirical outcomes of running those external methods on the new data; they are not forced by construction or by self-citation. Citations are to standard tooling (COLMAP, DJI software) and prior CD literature; none is a load-bearing uniqueness theorem or ansatz from the present authors. The fidelity gap between synthetic renders and real RGB is a correctness/assumption risk, not circularity. Score 0 is therefore the correct finding.

Axiom & Free-Parameter Ledger

4 free parameters · 4 axioms · 2 invented entities

The work is an empirical dataset + benchmark paper. It rests on standard computer-vision and remote-sensing assumptions plus a small set of engineering choices that define the synthetic data. No free physical constants or new theoretical entities are introduced; the free parameters are the discrete design decisions of the rendering and labeling pipeline that determine what the benchmark actually measures.

free parameters (4)
  • render_resolution_scale = 1/8
    Fixed at 1/8 of original camera resolution to raise pixel coverage to ~97%; chosen by inspection of coverage vs. texture trade-off (§3.1.2).
  • point_splat_diameter = 1 pixel (chosen)
    Alternative volumetric model of 3-pixel diameter considered but discarded in favor of 1-pixel at 1/8 scale; hand-selected for coverage (§3.1.2).
  • synthetic_cameras_per_real = 32 then 300/scene
    Cascaded vertical/mirror/forward sampling produces 32 synthetic views per real camera, later down-sampled to 300 per scene; engineering choice (§3.1.1, §4.2).
  • class_weighting_and_ignore_policy = inverse pixel frequency + domain-specific Ignore
    Inverse-frequency class weights and the decision to map airborne snow, exterior fence regions, and certain terrain micro-changes to 'Ignore' directly shape the mIoU numbers (§3.2, §5.1).
axioms (4)
  • domain assumption Dense multi-temporal LiDAR point clouds plus novel-view rendering can produce photorealistic RGB images whose change statistics match real construction photography sufficiently for algorithm ranking.
    Load-bearing premise of the entire synthetic pipeline (§3.1); no quantitative domain-gap measurement is supplied.
  • domain assumption Single-temporal 3D semantic labels compared bi-directionally and unioned yield symmetric 2D change masks that are the correct supervision target for standard 2DCD networks.
    Stated in §3.2.2 as the solution to asymmetric point-cloud change; assumes semantic classes fully capture construction-relevant change.
  • domain assumption Operational safety constraints (no flights inside fence, altitude >65 m) make real multi-view bi-temporal capture infeasible, justifying synthetic camera generation.
    Motivates the whole approach (Introduction, §3.1).
  • standard math Standard supervised 2DCD training protocol (SGD, poly LR, inverse class weights, listed augmentations) is a fair comparison across methods and datasets.
    Follows prior CD literature cited in §5.1; used uniformly for Table 1.
invented entities (2)
  • iVISION-2DCD dataset no independent evidence
    purpose: Provide the first open multi-view, long-term construction 2DCD benchmark with photorealistic renders and change labels.
    The central artifact of the paper; independent evidence will come from external adoption and re-use, not yet demonstrated inside the manuscript.
  • vertical / mirror / forward synthetic camera sampling cascade no independent evidence
    purpose: Generate viewpoint diversity (altitude, interior, close-range) forbidden by real flight safety rules.
    Engineering construct introduced in §3.1.1; no external validation that the resulting distribution matches real deployable camera poses.

pith-pipeline@v1.1.0-grok45 · 16819 in / 3499 out tokens · 26381 ms · 2026-07-12T01:39:15.409514+00:00 · methodology

0 comments
read the original abstract

Automation in construction is essential for reducing costs and human errors in large-scale projects. We approach the construction progress monitoring from the aspect of detecting changes in construction sites. As construction buildings continue to evolve in geometry and appearance over time, change detection need to be performed from arbitrary camera viewpoints. This necessitates developing 2D Change Detection (2DCD) algorithms that operate robustly across diverse camera perspectives at construction sites. While developing and evaluating such systems is data-intensive, no open-source benchmark dataset exists at the intersection of 2D change detection and construction automation research. Data collection using Unmanned Aerial Vehicles (UAVs) is gaining its popularity in outdoor large-scale surveying. However, in active construction sites conducting drone missions equipped with high-end sensors imposes safety concerns. Flight trajectory and collected camera viewpoints can be significantly limited. To address this critical gap, we introduce iVISION-2DCD, a large-scale synthetically generated dataset from dense LiDAR point clouds with photorealistic input images and accurate ground truth annotations. Our dataset formally defines the problem of viewpoint-robust 2DCD at construction sites and captures the inherent complexities of real-world deployment. In this paper, we present our systematic methodology for synthetic data generation, developing novel view synthesis techniques to overcome bi-temporal alignment and viewpoint diversity challenges, and implementing semi-automated semantic segmentation with change label generation while preserving challenging real-world cases. Benchmark evaluations using state-of-the-art 2DCD algorithms demonstrate that iVISION-2DCD poses novel research challenges for the computer vision and robotics communities.

Figures

Figures reproduced from arXiv: 2607.03553 by Alexander Wong, Ashkan Ebadi, Dayou Mao, John Zelek, Yuchen Lin, Yuhao Chen.

Figure 1
Figure 1. Figure 1: Example scenes (not consecutive) rendered from a distant camera: Dec-03-2024, Jan-19-2025, Feb-08-2025, [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: From left to right: Illustration of vertical, mirror, forward sampling, and visualization of complete synthetic [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Left: original capture of the first image in the November 27 LiDAR mission. Right: synthetic rendering at [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Example of airborne snow captured in the point clouds. Left column: time [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Example of ignore ground semantic changes hard to recognize from RGB inputs. From left to right: time [PITH_FULL_IMAGE:figures/full_fig_p007_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Examples of dynamic elements in the construction site. Figures shown by DJI Modify to amplify the [PITH_FULL_IMAGE:figures/full_fig_p007_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Examples of RGB projection error in the reconstructed LiDAR point clouds. Figures shown by DJI Modify to [PITH_FULL_IMAGE:figures/full_fig_p007_7.png] view at source ↗

discussion (0)

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

Reference graph

Works this paper leans on

50 extracted references · 5 linked inside Pith

  1. [1]

    W., & Leite, F

    Czerniawski, T., Ma, J. W., & Leite, F. (2021). Automated building change detection with amodal completion of point clouds. Automation in construction, 124, 103568

    2021

  2. [2]

    Meyer, T., Brunn, A., & Stilla, U. (2022). Change detection for indoor construction progress monitoring based on BIM, point clouds and uncertainties. Automation in Construction, 141, 104442

    2022

  3. [3]

    B., Song, M., & Cho, J

    Han, D., Lee, S. B., Song, M., & Cho, J. S. (2021). Change detection in unmanned aerial vehicle images for progress monitoring of road construction. Buildings, 11(4), 150

    2021

  4. [4]

    W., Zhu, Z., & Zhao, Y

    Suh, J. W., Zhu, Z., & Zhao, Y . (2024). Monitoring construction changes using dense satellite time series and deep learning. Remote Sensing of Environment, 309, 114207

    2024

  5. [5]

    H., Pang, Y ., & Soergel, U

    Yang, C. H., Pang, Y ., & Soergel, U. (2017). Monitoring of building construction by 4D change detection using multi-temporal SAR images. ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences, 4, 35-42

    2017

  6. [6]

    Huang, R., Xu, Y ., Hoegner, L., & Stilla, U. (2022). Semantics-aided 3D change detection on construction sites using UA V-based photogrammetric point clouds. Automation in construction, 134, 104057

    2022

  7. [7]

    J., Valentino, G., & Di Castro, M

    Attard, L., Debono, C. J., Valentino, G., & Di Castro, M. (2018). Vision-based change detection for inspection of tunnel liners. Automation in Construction, 91, 142-154

    2018

  8. [8]

    K., Lin, J

    Ham, Y ., Han, K. K., Lin, J. J., & Golparvar-Fard, M. (2016). Visual monitoring of civil infrastructure systems via camera-equipped Unmanned Aerial Vehicles (UA Vs): a review of related works. Visualization in Engineering, 4(1), 1

    2016

  9. [9]

    Agarwal, R., Chandrasekaran, S., & Sridhar, M. (2016). Imagining construction’s digital future. McKinsey & Company, 24(06), 1-13

    2016

  10. [10]

    Golparvar-Fard, M., Pena-Mora, F., & Savarese, S. (2015). Automated progress monitoring using unordered daily construction photographs and IFC-based building information models. Journal of Computing in Civil En- gineering, 29(1), 04014025

    2015

  11. [11]

    W., Vela, P

    Yang, J., Park, M. W., Vela, P. A., & Golparvar-Fard, M. (2015). Construction performance monitoring via still images, time-lapse photos, and video streams: Now, tomorrow, and the future. Advanced Engineering Informat- ics, 29(2), 211-224

    2015

  12. [12]

    Ji, S., Wei, S., & Lu, M. (2018). 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. 9 iVISION-2DCD: Long-Term Construction Change DetectionACCEPTED ATICRA 2026

    2018

  13. [13]

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

    2020

  14. [14]

    A., Vizilter, Y

    Lebedev, M. A., Vizilter, Y . V ., Vygolov, O. V ., Knyaz, V . A., & Rubis, A. Y . (2018). Change detection in re- mote sensing images using conditional adversarial networks. The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, 42, 565-571

    2018

  15. [15]

    C., Le Saux, B., Boulch, A., & Gousseau, Y

    Daudt, R. C., Le Saux, B., Boulch, A., & Gousseau, Y . (2018, July). Urban change detection for multispectral earth observation using convolutional neural networks. In IGARSS 2018-2018 IEEE International Geoscience and Remote Sensing Symposium (pp. 2115-2118). Ieee

    2018

  16. [16]

    Benedek, C., & Szirányi, T. (2009). Change detection in optical aerial images by a multilayer conditional mixed Markov model. IEEE Transactions on Geoscience and Remote Sensing, 47(10), 3416-3430

    2009

  17. [17]

    & Gaston, M

    Gupta, R., Hosfelt, R., Sajeev, S., Patel, N., Goodman, B., Doshi, J., ... & Gaston, M. (2019). xbd: A dataset for assessing building damage from satellite imagery. arXiv preprint arXiv:1911.09296

    Pith/arXiv arXiv 2019

  18. [18]

    de Gélis, I., Lefèvre, S., & Corpetti, T. (2021). Change detection in urban point clouds: An experimental com- parison with simulated 3d datasets. Remote Sensing, 13(13), 2629

    2021

  19. [19]

    Wang, Z., Zhang, Y ., Luo, L., Yang, K., & Xie, L. (2023). An end-to-end point-based method and a new dataset for street-level point cloud change detection. IEEE Transactions on Geoscience and Remote Sensing, 61, 1-15

    2023

  20. [20]

    C., Le Saux, B., & Boulch, A

    Daudt, R. C., Le Saux, B., & Boulch, A. (2018, October). Fully convolutional siamese networks for change detection. In 2018 25th IEEE international conference on image processing (ICIP) (pp. 4063-4067). IEEE

    2018

  21. [21]

    Fang, S., Li, K., Shao, J., & Li, Z. (2021). SNUNet-CD: A densely connected Siamese network for change detection of VHR images. IEEE Geoscience and Remote Sensing Letters, 19, 1-5

    2021

  22. [22]

    Yan, T., Wan, Z., & Zhang, P. (2022). Fully transformer network for change detection of remote sensing images. In Proceedings of the Asian Conference on Computer Vision (pp. 1691-1708)

    2022

  23. [23]

    Bandara, W. G. C., & Patel, V . M. (2022, July). A transformer-based siamese network for change detection. In IGARSS 2022-2022 IEEE International Geoscience and Remote Sensing Symposium (pp. 207-210). IEEE

    2022

  24. [24]

    Zhang, H., Chen, K., Liu, C., Chen, H., Zou, Z., & Shi, Z. (2024). CDMamba: Incorporating Local Clues into Mamba for Remote Sensing Image Binary Change Detection. arXiv preprint arXiv:2406.04207

    Pith/arXiv arXiv 2024

  25. [25]

    Chen, H., Song, J., Han, C., Xia, J., & Yokoya, N. (2024). ChangeMamba: Remote sensing change detection with spatiotemporal state space model. IEEE Transactions on Geoscience and Remote Sensing, 62, 1-20

    2024

  26. [26]

    Shi, Q., Liu, M., Li, S., Liu, X., Wang, F., & Zhang, L. (2021). A deeply supervised attention metric-based network and an open aerial image dataset for remote sensing change detection. IEEE transactions on geoscience and remote sensing, 60, 1-16

    2021

  27. [27]

    Zhang, C., Yue, P., Tapete, D., Jiang, L., Shangguan, B., Huang, L., & Liu, G. (2020). A deeply supervised image fusion network for change detection in high resolution bi-temporal remote sensing images. ISPRS Journal of Photogrammetry and Remote Sensing, 166, 183-200

    2020

  28. [28]

    Chang, S., Kopp, M., Ghamisi, P., & Du, B. (2024). Dsfer-Net: A deep supervision and feature retrieval network for bitemporal change detection using modern Hopfield networks. IEEE Transactions on Geoscience and Remote Sensing

    2024

  29. [29]

    Han, C., Wu, C., Guo, H., Hu, M., & Chen, H. (2023). HANet: A hierarchical attention network for change de- tection with bitemporal very-high-resolution remote sensing images. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 16, 3867-3878

    2023

  30. [30]

    Codegoni, A., Lombardi, G., & Ferrari, A. (2023). TINYCD: A (not so) deep learning model for change detec- tion. Neural Computing and Applications, 35(11), 8471-8486

    2023

  31. [31]

    Gan, Y ., Xuan, W., Chen, H., Liu, J., & Du, B. (2024). RFL-CDNet: Towards accurate change detection via richer feature learning. Pattern Recognition, 153, 110515

    2024

  32. [32]

    Zhang, H., Chen, H., Zhou, C., Chen, K., Liu, C., Zou, Z., & Shi, Z. (2024). Bifa: Remote sensing image change detection with bitemporal feature alignment. IEEE Transactions on Geoscience and Remote Sensing, 62, 1-17

    2024

  33. [33]

    Wu, Z., Ma, X., Lian, R., Lin, Z., & Zhang, W. (2024). CDXFormer: Boosting Remote Sensing Change Detection with Extended Long Short-Term Memory. arXiv e-prints, arXiv-2411

    2024

  34. [34]

    Fang, S., Li, K., & Li, Z. (2023). Changer: Feature interaction is what you need for change detection. IEEE Transactions on Geoscience and Remote Sensing, 61, 1-11. 10 iVISION-2DCD: Long-Term Construction Change DetectionACCEPTED ATICRA 2026

    2023

  35. [35]

    (2023, July)

    Han, C., Wu, C., & Du, B. (2023, July). HCGMNet: A hierarchical change guiding map network for change detection. In IGARSS 2023-2023 IEEE International Geoscience and Remote Sensing Symposium (pp. 5511- 5514). IEEE

    2023

  36. [36]

    Zheng, Z., Ma, A., Zhang, L., & Zhong, Y . (2021). Change is everywhere: Single-temporal supervised object change detection in remote sensing imagery. In Proceedings of the IEEE/CVF international conference on com- puter vision (pp. 15193-15202)

    2021

  37. [37]

    Zheng, Z., Zhong, Y ., Ma, A., & Zhang, L. (2024). Single-temporal supervised learning for universal remote sensing change detection. International Journal of Computer Vision, 132(12), 5582-5602

    2024

  38. [38]

    Chen, H., Song, J., Wu, C., Du, B., & Yokoya, N. (2023). Exchange means change: An unsupervised single- temporal change detection framework based on intra-and inter-image patch exchange. ISPRS journal of pho- togrammetry and remote sensing, 206, 87-105

    2023

  39. [39]

    Shu, Q., Pan, J., Zhang, Z., & Wang, M. (2022). MTCNet: Multitask consistency network with single temporal supervision for semi-supervised building change detection. International Journal of Applied Earth Observation and Geoinformation, 115, 103110

    2022

  40. [40]

    & Wang, C

    Du, Q., Peng, J., Chen, X., He, Q., He, L., Nie, Q., ... & Wang, C. (2024). Single-temporal supervised remote change detection for domain generalization. arXiv preprint arXiv:2404.11326

    Pith/arXiv arXiv 2024

  41. [41]

    de Gélis, I., Lefèvre, S., & Corpetti, T. (2023). Siamese KPConv: 3D multiple change detection from raw point clouds using deep learning. ISPRS Journal of Photogrammetry and Remote Sensing, 197, 274-291

    2023

  42. [42]

    Tran, T. H. G., Ressl, C., & Pfeifer, N. (2018). Integrated change detection and classification in urban areas based on airborne laser scanning point clouds. Sensors, 18(2), 448

    2018

  43. [43]

    Zhan, W., Cheng, R., & Chen, J. (2024). PGN3DCD: Prior-knowledge-guided network for urban 3D point cloud change detection. IEEE Transactions on Geoscience and Remote Sensing

    2024

  44. [44]

    & Wang, H

    Lu, J., Dai, C., Zhang, Z., Liu, X., Zhou, R., Ji, S., ... & Wang, H. (2025). Ms-DANet: Multi-scale Difference- aware Network for 3D Point Cloud Change Detection. IEEE Transactions on Geoscience and Remote Sensing

    2025

  45. [45]

    DJI. (2025). DJI Terra (Version 5.0.2) [Computer software]. https://enterprise.dji.com/dji-terra

    2025

  46. [46]

    DJI. (2025). DJI Modify (Version 1.4.0) [Computer software]. https://enterprise.dji.com/modify

    2025

  47. [47]

    Y ., Johnson, J., & Gkioxari, G

    Ravi, N., Reizenstein, J., Novotny, D., Gordon, T., Lo, W. Y ., Johnson, J., & Gkioxari, G. (2020). Accelerating 3d deep learning with pytorch3d. arXiv preprint arXiv:2007.08501

    Pith/arXiv arXiv 2020

  48. [48]

    Y ., Park, J., & Koltun, V

    Zhou, Q. Y ., Park, J., & Koltun, V . (2018). Open3D: A modern library for 3D data processing. arXiv preprint arXiv:1801.09847

    Pith/arXiv arXiv 2018

  49. [49]

    Blender Development Team. (2022). Blender (Version 3.1.0) [Computer software]. https://www.blender.org

    2022

  50. [50]

    L., & Frahm, J

    Schonberger, J. L., & Frahm, J. M. (2016). Structure-from-motion revisited. In Proceedings of the IEEE confer- ence on computer vision and pattern recognition (pp. 4104-4113). 11

    2016