Pith. sign in

REVIEW 5 major objections 5 minor 60 references

Flow-CDNet: A Novel Network for Detecting Both Slow and Fast Changes in Bitemporal Images

T0 review · 5 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read A single dual-branch network, Flow-CDNet, claims to detect both slow displacements and sudden changes in bitemporal images, outperforming flow-only and change-only baselines on a synthetic benchmark and generalizing to real dam-bank…

desk verdict A plausible dual-branch joint slow/fast change detection idea with a coherent synthetic ablation, but real-world evidence is too weak to support the flagship claim of detecting gradual deformation. read the letter →

arxiv 2507.02307 v1 pith:4OGV44ZB submitted 2025-07-03 cs.CV

Haoxuan Li , Chenxu Wei , Haodong Wang , Xiaomeng Hu , Boyuan An , Lingyan Ran , Baosen Zhang , Jin Jin
show 4 more authors
Omirzhan Taukebayev Amirkhan Temirbayev Junrui Liu Xiuwei Zhang
This is my paper · ORCID
classification cs.CV
keywords changedetectionopticalflowbitemporalimagesslowfastdual-branchnetworkFEPEmetricsyntheticdataset
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

Change detection in monitoring imagery usually targets fast changes, such as objects appearing or disappearing between two pictures. This paper takes on the harder, practically relevant problem of also catching slow changes, the small displacements of soil, rock, or structure that often precede failures like slope slides or dam damage. The paper's central claim is that one dual-branch network, Flow-CDNet, can detect both kinds at once: one branch estimates dense optical flow, and the other uses that flow to produce a binary change mask. On the paper's synthetic Flow-Change dataset, the joint model reaches an F1 score of 0.892, a mean end-point error of 1.027, and an FEPE score of 0.869, beating flow-only, change-only, and combined baselines, and it transfers to 20 real dam-bank image pairs with an F1 score of 0.8126.

What carries the argument

The machine that carries the argument is the dual-branch coupling between optical flow and change detection. The optical-flow branch uses a RAFT-style architecture with a four-level correlation pyramid and convolutional GRU updates to estimate a dense displacement field; the change-detection branch warps the second image by that flow, takes the absolute difference with the first image, applies an adaptive mask, and pushes the result through a ResNet50 backbone with pyramid pooling to produce a binary change map. This coupling is supervised by a composite loss, L2 flow error on pixels outside fast-change regions plus Tversky loss on the change mask, and is measured by FEPE, a single score defined as F1 divided by the sum of mean end-point error and a small epsilon, so a model must be good at both tasks to rank well.

What would settle it

Take a set of real slope or dam image pairs with independently measured dense displacement fields, for example from survey markers or controlled laboratory deformation, run the trained Flow-CDNet without fine-tuning, and compare its mean end-point error in slowly deforming regions against a flow-only RAFT baseline; if the dual-branch model does not beat or match that baseline there, or if its change masks miss gradually deforming areas, the claimed simultaneous slow and fast change detection is not supported.

Watch

Extended reading notes

Core claim

The paper introduces Flow-CDNet, a unified framework in which a pyramid optical-flow branch and a ResNet-based binary change-detection branch are trained together so that slow changes, where an object is present in both images but shifted or deformed, and fast changes, where an object appears in only one image, are detected simultaneously. The flow branch is built on RAFT-style feature extraction, a multi-scale 4D correlation volume, and iterative update, and it feeds the change branch with a motion-compensated warped image, an absolute difference map, and an adaptive mask; the change branch then outputs a binary map under a combined loss of L2 flow error, masked to exclude fast-change regions, and Tversky segmentation loss. The authors report that Flow-CDNet reaches F1 0.892 and a mean end-point error of 1.027 on the synthetic Flow-Change benchmark, giving FEPE 0.869, and their ablation shows the two branches improve each other: adding the flow branch raises F1 from 0.753 to 0.892, while adding the change branch lowers the flow error from 2.409 to 1.027. They further claim that the pretrained model, without fine-tuning, detects both abrupt collapses and gradual surface deformation in real dam-bank imagery, reaching F1 0.8126 after per-scene thresholds. On its own terms, the contribution is a working demonstration that a single network can produce both a usable dense flow field and a usable change mask for monitoring scenarios.

Load-bearing premise

The paper's central comparison rests on the assumption that its synthetic dataset, made by pasting transformed objects from one image collection onto optical-flow pairs from another, faithfully represents real slow and fast changes; the slow changes it contains are simple object shifts, not gradual small-scale ground movements.

Editorial extensions

If this is right

  • A monitoring system could run one inference per image pair and receive both a dense displacement field and a binary change mask, so gradual precursors and sudden events are flagged together.
  • Because the two branches improve each other, with F1 rising from 0.753 to 0.892 and flow error falling from 2.409 to 1.027, architectures built for either task alone may gain from adding the other task's head and loss.
  • The FEPE metric gives a single ranking for models that output both flow and change maps, making slow-plus-fast performance comparable rather than reporting two separate numbers.
  • The reported transfer to real dam-bank pairs without fine-tuning suggests that a model trained on synthetic flow-plus-change data can generalize to unseen monitoring sites, at least for the change types represented in the test set.

Reading between the lines

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

  • The decisive next test is not a new network but a new dataset: real slope or dam image pairs with independently measured subpixel displacement fields, because the synthetic slow changes are rigid object shifts and do not exercise genuine gradual deformation.
  • A useful ablation the paper does not report is freezing a pretrained flow branch and training only the change head; that would separate the benefit of motion-aligned inputs from the benefit of joint gradient updates, pinning down where the mutual improvement comes from.
  • The FEPE ratio should be treated cautiously as a composite: a model with a slightly worse F1 but a much smaller flow error can outrank a more balanced model, so applications should inspect the two components as well as the combined score.
  • An extension this framework invites is predicting a continuous change-magnitude map instead of a binary mask, which would let slow and fast changes live on the same output and make the method directly usable for early-warning thresholds.
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

5 major / 5 minor

Summary. The paper proposes Flow-CDNet, a dual-branch network for detecting both 'slow' changes (object displacement) and 'fast' changes (appearance/disappearance) in bitemporal images. An optical flow branch (based on RAFT) and a binary change detection branch (based on ResNet/PSP) are trained jointly with an L2 loss on masked flow regions and a Tversky loss on change masks. To support training and evaluation, the authors construct a synthetic dataset Flow-Change from FlyingChairs and PASCAL VOC, introduce a composite loss, and propose a new metric FEPE. Quantitative results on Flow-Change show that the full model outperforms single-task baselines and the authors' own variants, and an ablation indicates mutual improvement between the two branches. A small real-world dataset of 20 dam-image pairs is used to demonstrate generalization, with a reported F1 of 0.8126.

Significance. The idea of jointly estimating optical flow and change detection is relevant for monitoring applications such as dam and slope surveillance, and the internal ablation (F1 rising from 0.753 to 0.892, mEPE falling from 2.409 to 1.027 when both branches are used) is a useful empirical observation. The new dataset and metric, if released and properly validated, could benefit the change-detection community. However, the evaluation is almost entirely in-house: the synthetic benchmark is constructed by the authors, the comparison set consists of single-task baselines and the authors' own variants, and the real-world validation uses a small sample with unified labels that cannot attribute detections to the flow branch. As a result, the central claim of robust simultaneous detection of slow and fast changes in real scenes is not yet substantiated.

major comments (5)
  1. [4.3] The real-world evaluation cannot support the claim that slow changes are detected. The unified ground-truth annotations do not distinguish slow from fast changes, the reported F1=0.8126 is obtained after per-scene threshold selection with no protocol or held-out tuning, and the statement that detections from either branch are considered valid means the metric cannot attribute performance to the optical flow branch. Please provide separate ground-truth maps for slow and fast changes (or at least disjoint subsets), fix the threshold-selection protocol, and report baseline comparisons (e.g., the CD branch alone, the flow branch alone, and a frame-differencing method) on the same real-world data.
  2. [4.2, Table 1] The model naming is inconsistent: the text states that Flow-CDNet utilizes SpyNetC2, while Table 1 lists Flow-CDNet-S (SpyNet+CDNet) separately and later sections identify the proposed Flow-CDNet as RAFT-based. This ambiguity makes it unclear which variant is the proposed method. In addition, Table 1 compares only single-task baselines and the authors' own variants; no existing change detection or optical flow network is compared on the same data, so the claim of outperforming existing methods is not demonstrated. Please clarify the naming and add comparisons to strong off-the-shelf baselines (e.g., DASNet, ChangeViT, GMFlow, FlowFormer).
  3. [3.2, Eq. (4)] The Tversky term in Eq. (4) is the Tversky index, not the Tversky loss; minimizing this quantity would drive the true-positive term masked_gt toward zero, degenerating the change-detection output. The standard Tversky loss is 1 - (masked_gt / (masked_gt + α·wrong_classified + β·unmasked_gt)). Please correct the equation and explicitly state the optimization direction (minimization or maximization) used in training.
  4. [3.3, Eqs. (9)-(11)] The definitions of EPE and FEPE are imprecise: Eq. (9) defines EPE as the square root of the absolute difference of flow vectors rather than the Euclidean norm of their difference, and Eq. (11) leaves the perturbation ε unspecified. The numerical values in Table 1 (F1=0.892, mEPE=1.027, FEPE=0.869) are approximately consistent with ε=0, but this is never stated. Since FEPE is used as the ranking metric, please define it rigorously, report the value of ε and its sensitivity, and use the standard endpoint-error definition.
  5. [4.1] The Flow-Change synthetic dataset models slow changes as rigid displacements of PASCAL VOC objects on FlyingChairs backgrounds, which is not representative of the gradual, non-rigid deformation (e.g., soil movement or crack growth) described in the introduction. The real-world dam images lack flow ground truth, leaving the synthetic-to-real transfer for deformation unquantified. Please either temper the claims about detecting gradual deformation or add non-rigid deformation to the synthetic data and provide quantitative transfer evidence on real images.
minor comments (5)
  1. [4.2] In the ablation paragraph, 'Flow-CDNet achieves a higher mEPE metric' should read 'lower mEPE', since a smaller mEPE indicates better optical flow estimation.
  2. [3.2, 4.2] The loss weighting coefficient is denoted ψ in Eq. (5) but φ in Section 4.2; please unify the notation.
  3. [3.3] Eqs. (9) and (10) should use explicit vector norms; the current notation is ambiguous and does not match standard definitions of endpoint error.
  4. [4.1] Please clarify how the fast-change objects are placed (e.g., appearing only in the second image) and how the binary change labels are computed, and provide a public link to the dataset and code to facilitate reproducibility.
  5. [3.1.1] The 'adaptive mask mechanism' is described only qualitatively; please specify its inputs, form, and role in the network, or omit the term if it is not a separate component.

Circularity Check

1 steps flagged · score 4.0 of 10

Core dual-branch derivation is self-contained; one real-world evaluation step is a post-hoc threshold fit and the real-world GT protocol cannot validate the slow-change claim.

  1. fitted input called prediction [Section 4.3, 'Experiment on Real-World Data' (annotation paragraph and final F1 paragraph)]
    "Ground truth (GT) annotations were manually constructed to identify regions exhibiting noticeable change between the two time points. ... all observable change regions are marked without categorizing them as abrupt or gradual. Consequently, both the detections from the optical flow estimation branch and the change detection branch are considered valid if they correctly localize these annotated regions. ..."

    The real-world F1=0.8126 is obtained after per-scene threshold selection on the same data used for scoring, so the threshold is a fitted parameter and the score is a post-hoc optimized quantity rather than a prediction at a fixed operating point. The paper gives no protocol for choosing these thresholds and no held-out set, so the number cannot be compared with any baseline at an equivalent operating point. In addition, the GT is a union of slow and fast change regions and detections from either branch are counted as valid, so the F1 cannot attribute any part of the measured success to the optical flow branch or to gradual-deformation detection. The real-world generalization claim is therefore supported by a metric that is defined and threshold-tuned after seeing the test data.

full rationale

The central model derivation is not circular: the dual-branch architecture is a straightforward supervised combination of a RAFT-style flow estimator and a PSP-style change classifier, trained with a masked L2 loss on flow plus a Tversky loss on change masks. No load-bearing uniqueness theorem or author self-citation is invoked, and the synthetic Flow-Change experiments and ablations are internally consistent: the joint model is compared against single-branch versions of the same pipeline, so the mutual-improvement claim is at least measurable. The main circularity burden is concentrated in the real-world validation (Section 4.3). The paper itself states the limitation that flow vectors cannot be labeled and that either branch's detection counts as valid; this makes the slow-change half of the central claim unfalsifiable on real data. The separately reported F1=0.8126 is further weakened by per-scene threshold selection, which turns the evaluation score into a fitted value. Because the abstract and conclusion use this real-world evidence to claim robust detection of gradual deformations, the paper has partial evaluation-circularity, but the core network derivation and synthetic ablation remain self-contained, so the score is moderate rather than severe.

Assumptions & free parameters 5 free parameters · 5 assumptions · 3 invented entities

The central claim rests on a domain decomposition (flow equals slow change, binary appearance equals fast change), a synthetic dataset built by the authors as the sole benchmark, and several hand-chosen losses and thresholds. The FEPE metric and Flow-Change dataset are new instruments with no external validation, so the headline comparison is internally defined. No novel physics or entities are postulated; the invented entities are evaluation artifacts whose validity is asserted rather than measured.

free parameters (5)
  • Tversky loss weights alpha and beta = alpha=0.7, beta=0.3
    Hand-chosen penalty weights in Eq (4) that control the Tversky loss; no sensitivity analysis is reported.
  • Multi-task loss weight psi (called phi in Section 4.2) = 10 (fixed ratio)
    Weight of the Tversky loss in Eq (5); it is fixed at a ratio of 10 without an ablation study.
  • FEPE perturbation epsilon = unspecified
    Small constant in Eq (11) that prevents division by zero and sets the tradeoff between F1 and mEPE; its value is never stated, so the reported FEPE scores cannot be reproduced or interpreted.
  • Per-scene change thresholds on real-world data = not reported, selected per scene
    Section 4.3 reports F1=0.8126 after selecting appropriate change thresholds for different scenes; the selection protocol is post hoc and uses the test set itself.
  • Training schedule (learning rates, epochs, batch size) = 1e-5 OF branch, 1e-4 CD branch, 1000 epochs, batch size 4
    Settings in Section 4.2 that the reported results depend on; they are standard choices but are still hand-chosen configuration.
assumptions (5)
  • domain assumption Slow changes are representable as dense 2D optical flow displacement, and fast changes as binary regions of appearance or disappearance
    Section 1 defines the two change types this way; the two-branch design and the FEPE metric inherit this decomposition.
  • domain assumption The synthetic Flow-Change dataset is a valid proxy for real monitoring scenarios
    Section 4.1 constructs the dataset by pasting PASCAL VOC objects onto FlyingChairs pairs; all quantitative claims in Table 1 are measured on it.
  • domain assumption Optical flow supervision restricted to non-fast-change regions is a well-posed training signal
    Eq (3) multiplies the flow L2 loss by (1-label2), so the flow branch is never supervised inside fast-change regions.
  • domain assumption FlyingChairs flow labels remain meaningful after pasting objects and applying brightness and contrast augmentation
    Section 4.1 keeps FlyingChairs flow as label1 while modifying the images; pasted objects can violate the original motion model in their regions.
  • domain assumption The Tversky loss with fixed alpha and beta is an appropriate surrogate for F1 optimization on change maps
    Standard practice in segmentation; the paper asserts rather than justifies the choice for change detection.
invented entities (3)
  • FEPE evaluation metric
    purpose: A single score combining change detection F1 and optical flow mEPE to rank methods on both tasks
    Defined in Eq (11) as F1 divided by mEPE plus epsilon; no external benchmark calibrates the composite and epsilon is unspecified, so the ranking is meaningful only inside this paper.
  • Flow-Change dataset
    purpose: Training and evaluation benchmark with bitemporal image pairs, flow labels, and change labels
    Self-built synthetic dataset from FlyingChairs and PASCAL VOC (Section 4.1); not released, and its fidelity to real slow and fast change is asserted rather than measured.
  • Adaptive mask mechanism in the CD branch
    purpose: Described in Section 3.1.1 as emphasizing fast-change regions using motion uncertainty and intensity
    Section 3.1.3 only mentions a mask mechanism employing optical flow magnitude analysis; the implementation is unspecified, so its contribution to the reported gains cannot be assessed.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Flow-CDNet: A Novel Network for Detecting Both Slow and Fast Changes in Bitemporal Images." pith.science (2026). https://pith.science/paper/4OGV44ZB

@misc{pith2026250702307,
  author       = {Pith},
  title        = {Pith review of: Flow-CDNet: A Novel Network for Detecting Both Slow and Fast Changes in Bitemporal Images},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/4OGV44ZB}},
  note         = {Machine review of arXiv:2507.02307}
}
read the original abstract

Change detection typically involves identifying regions with changes between bitemporal images taken at the same location. Besides significant changes, slow changes in bitemporal images are also important in real-life scenarios. For instance, weak changes often serve as precursors to major hazards in scenarios like slopes, dams, and tailings ponds. Therefore, designing a change detection network that simultaneously detects slow and fast changes presents a novel challenge. In this paper, to address this challenge, we propose a change detection network named Flow-CDNet, consisting of two branches: optical flow branch and binary change detection branch. The first branch utilizes a pyramid structure to extract displacement changes at multiple scales. The second one combines a ResNet-based network with the optical flow branch's output to generate fast change outputs. Subsequently, to supervise and evaluate this new change detection framework, a self-built change detection dataset Flow-Change, a loss function combining binary tversky loss and L2 norm loss, along with a new evaluation metric called FEPE are designed. Quantitative experiments conducted on Flow-Change dataset demonstrated that our approach outperforms the existing methods. Furthermore, ablation experiments verified that the two branches can promote each other to enhance the detection performance.

Figures

Figures reproduced from arXiv: 2507.02307 by the authors.

Figure 1
Figure 1. Overview of the proposed Flow-CDNet framework. [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. OFbranch. 3.1.3 Change Detection Branch. As shown in [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. CDbranch. 3.2 Loss Function To evaluate the results of optical flow detection, it is necessary to remove the binary changed regions (label2) and extract only the regions containing slow change, then calculate the L2 norm loss with label1 and output1, as shown in Equation (3). lossl2 = ||output1 − label1||2 · (1 − label2) (3) To evaluate the results of change detection, we employ the change detection labels (label2) … view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: Dataset Visualization, T 0 0 represents image in the first frame, T 1 0 represents image in the second frame, Optical Flow column shows the ground truth transition between frame one and two, and Fast Change column shows ground truth of the item that only showed up in f…
Figure 5
Figure 5. Figure 5: Model result visualization on the Synthetic Flow-Change Dataset [PITH_FULL_IMAGE:figures/full_fig_p011_5.png]
Figure 6
Figure 6. Figure 6: Visual comparison of model performance on real-world data. From top to bottom, the rows display: the input [PITH_FULL_IMAGE:figures/full_fig_p012_6.png]
Figure 7
Figure 7. Figure 7: Visual comparison of slow change detection performance using different optical flow backbones on real-world [PITH_FULL_IMAGE:figures/full_fig_p014_7.png]
Figure 8
Figure 8. Figure 8: Visual comparison of fast change detection performance with different optical flow backbones on real-world [PITH_FULL_IMAGE:figures/full_fig_p015_8.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

60 extracted references · 57 canonical work pages

  1. [1]

    Optical flow estimation using a spatial pyramid network

    Anurag Ranjan and Michael J Black. Optical flow estimation using a spatial pyramid network. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 4161–4170, 2017

    work page 2017

  2. [2]

    Continual occlusion and optical flow estimation

    Michal Neoral, Jan Šochman, and Jiˇrí Matas. Continual occlusion and optical flow estimation. In Asian conference on computer vision, pages 159–174. Springer, 2018

    work page 2018

  3. [3]

    Maskflownet: Asymmetric feature matching with learnable occlusion mask

    Shengyu Zhao, Yilun Sheng, Yue Dong, Eric I Chang, Yan Xu, et al. Maskflownet: Asymmetric feature matching with learnable occlusion mask. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 6278–6287, 2020

    work page 2020

  4. [4]

    Liteflownet: A lightweight convolutional neural network for optical flow estimation

    Tak-Wai Hui, Xiaoou Tang, and Chen Change Loy. Liteflownet: A lightweight convolutional neural network for optical flow estimation. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 8981–8989, 2018

    work page 2018

  5. [5]

    Raft: Recurrent all-pairs field transforms for optical flow

    Zachary Teed and Jia Deng. Raft: Recurrent all-pairs field transforms for optical flow. In Computer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Proceedings, Part II 16, pages 402–419. Springer, 2020. 15 A PREPRINT - S EPTEMBER 23, 2025

    work page 2020

  6. [6]

    Accflow: Backward accumulation for long-range optical flow

    Guangyang Wu, Xiaohong Liu, Kunming Luo, Xi Liu, Qingqing Zheng, Shuaicheng Liu, Xinyang Jiang, Guangtao Zhai, and Wenyi Wang. Accflow: Backward accumulation for long-range optical flow. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 12119–12128, 2023

    work page 2023

  7. [7]

    Videoflow: Exploiting temporal cues for multi-frame optical flow estimation

    Xiaoyu Shi, Zhaoyang Huang, Weikang Bian, Dasong Li, Manyuan Zhang, Ka Chun Cheung, Simon See, Hongwei Qin, Jifeng Dai, and Hongsheng Li. Videoflow: Exploiting temporal cues for multi-frame optical flow estimation. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 12469–12480, 2023

    work page 2023

  8. [8]

    Pyramid scene parsing network

    Hengshuang Zhao, Jianping Shi, Xiaojuan Qi, Xiaogang Wang, and Jiaya Jia. Pyramid scene parsing network. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 2881–2890, 2017

    2017

Show all 60 references
  1. [9]

    Dasnet: Dual attentive fully convolutional siamese networks for change detection in high-resolution satellite images

    Jie Chen, Ziyang Yuan, Jian Peng, Li Chen, Haozhe Huang, Jiawei Zhu, Yu Liu, and Haifeng Li. Dasnet: Dual attentive fully convolutional siamese networks for change detection in high-resolution satellite images. IEEE Journal of Selected Topics in Applied Earth Observations and ...

    2020

  2. [10]

    Building change detection for remote sensing images using a dual-task constrained deep siamese convolutional network model

    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 convolutional network model. IEEE Geoscience and Remote Sensing Letters, 18(5):811–815, 2020

    2020

  3. [11]

    Epicflow: Edge-preserving interpolation of correspondences for optical flow

    Jerome Revaud, Philippe Weinzaepfel, Zaid Harchaoui, and Cordelia Schmid. Epicflow: Edge-preserving interpolation of correspondences for optical flow. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 1164–1172, 2015

    2015

  4. [12]

    Deepflow: Large displacement optical flow with deep matching

    Philippe Weinzaepfel, Jerome Revaud, Zaid Harchaoui, and Cordelia Schmid. Deepflow: Large displacement optical flow with deep matching. In Proceedings of the IEEE international conference on computer vision, pages 1385–1392, 2013

    2013

  5. [13]

    Mirrorflow: Exploiting symmetries in joint optical flow and occlusion estimation

    Junhwa Hur and Stefan Roth. Mirrorflow: Exploiting symmetries in joint optical flow and occlusion estimation. In Proceedings of the IEEE International Conference on Computer Vision, pages 312–321, 2017

    2017

  6. [14]

    Flownet: Learning optical flow with convolutional networks

    Alexey Dosovitskiy, Philipp Fischer, Eddy Ilg, Philip Hausser, Caner Hazirbas, Vladimir Golkov, Patrick Van Der Smagt, Daniel Cremers, and Thomas Brox. Flownet: Learning optical flow with convolutional networks. In Proceedings of the IEEE international conference on computer v...

    2015

  7. [15]

    Flownet 2.0: Evolution of optical flow estimation with deep networks

    Eddy Ilg, Nikolaus Mayer, Tonmoy Saikia, Margret Keuper, Alexey Dosovitskiy, and Thomas Brox. Flownet 2.0: Evolution of optical flow estimation with deep networks. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 2462–2470, 2017

    2017

  8. [16]

    Liteflownet3: Resolving correspondence ambiguity for more accurate optical flow estimation

    Tak-Wai Hui and Chen Change Loy. Liteflownet3: Resolving correspondence ambiguity for more accurate optical flow estimation. In Computer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Proceedings, Part XX 16, pages 169–184. Springer, 2020

    2020

  9. [17]

    Global matching with overlapping attention for optical flow estimation

    Shiyu Zhao, Long Zhao, Zhixing Zhang, Enyu Zhou, and Dimitris Metaxas. Global matching with overlapping attention for optical flow estimation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 17592–17601, 2022

    2022

  10. [18]

    Gmflow: Learning optical flow via global matching

    Haofei Xu, Jing Zhang, Jianfei Cai, Hamid Rezatofighi, and Dacheng Tao. Gmflow: Learning optical flow via global matching. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 8121–8130, 2022

    2022

  11. [19]

    Learning optical flow from a few matches

    Shihao Jiang, Yao Lu, Hongdong Li, and Richard Hartley. Learning optical flow from a few matches. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 16592–16600, 2021

    2021

  12. [20]

    Flowformer: A transformer architecture for optical flow

    Zhaoyang Huang, Xiaoyu Shi, Chao Zhang, Qiang Wang, Ka Chun Cheung, Hongwei Qin, Jifeng Dai, and Hongsheng Li. Flowformer: A transformer architecture for optical flow. In European conference on computer vision, pages 668–685. Springer, 2022

    2022

  13. [21]

    Samflow: Eliminating any fragmentation in optical flow with segment anything model

    Shili Zhou, Ruian He, Weimin Tan, and Bo Yan. Samflow: Eliminating any fragmentation in optical flow with segment anything model. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 38, pages 7695–7703, 2024

    2024

  14. [22]

    Anyflow: Arbitrary scale optical flow with implicit neural representation

    Hyunyoung Jung, Zhuo Hui, Lei Luo, Haitao Yang, Feng Liu, Sungjoo Yoo, Rakesh Ranjan, and Denis Demandolx. Anyflow: Arbitrary scale optical flow with implicit neural representation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 545...

    2023

  15. [23]

    Distractflow: Improving optical flow estimation via realistic distractions and pseudo-labeling

    Jisoo Jeong, Hong Cai, Risheek Garrepalli, and Fatih Porikli. Distractflow: Improving optical flow estimation via realistic distractions and pseudo-labeling. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 13691–13700, 2023

    2023

  16. [24]

    Rapidflow: Recurrent adaptable pyramids with iterative decoding for efficient optical flow estimation

    Henrique Morimitsu, Xiaobin Zhu, Roberto M Cesar, Xiangyang Ji, and Xu-Cheng Yin. Rapidflow: Recurrent adaptable pyramids with iterative decoding for efficient optical flow estimation. In 2024 IEEE International Conference on Robotics and Automation (ICRA), pages 2946–2952. IE...

    2024

  17. [25]

    Lightweight optical flow estimation using 1d matching.IEEE Access, 2024

    Wonyong Seo, Woonsung Park, and Munchurl Kim. Lightweight optical flow estimation using 1d matching.IEEE Access, 2024

    2024

  18. [26]

    Rethinking optical flow from geometric matching consistent perspective

    Qiaole Dong, Chenjie Cao, and Yanwei Fu. Rethinking optical flow from geometric matching consistent perspective. In Proceedings of the IEEE/CVF Conference on computer vision and pattern recognition , pages 1337–1347, 2023

    2023

  19. [27]

    Craft: Cross-attentional flow transformer for robust optical flow

    Xiuchao Sui, Shaohua Li, Xue Geng, Yan Wu, Xinxing Xu, Yong Liu, Rick Goh, and Hongyuan Zhu. Craft: Cross-attentional flow transformer for robust optical flow. In Proceedings of the IEEE/CVF conference on Computer Vision and Pattern Recognition, pages 17602–17611, 2022

    2022

  20. [28]

    I-raft: Optical flow estimation model based on multi-scale initialization strategy

    Shunpan Liang, Xirui Zhang, and Yulei Hou. I-raft: Optical flow estimation model based on multi-scale initialization strategy. In International Conference on Neural Information Processing, pages 16–29. Springer, 2023

    2023

  21. [29]

    Learning optical flow with kernel patch attention

    Ao Luo, Fan Yang, Xin Li, and Shuaicheng Liu. Learning optical flow with kernel patch attention. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 8906–8915, 2022

    2022

  22. [30]

    Flowdiffuser: Advancing optical flow estimation with diffusion models

    Ao Luo, Xin Li, Fan Yang, Jiangyu Liu, Haoqiang Fan, and Shuaicheng Liu. Flowdiffuser: Advancing optical flow estimation with diffusion models. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 19167–19176, 2024

    2024

  23. [31]

    Deeppynet: A deep feature pyramid network for optical flow estimation

    Afsana Ahsan Jeny, Md Baharul Islam, and Tarkan Aydin. Deeppynet: A deep feature pyramid network for optical flow estimation. In 2021 36th International Conference on Image and Vision Computing New Zealand (IVCNZ), pages 1–6. IEEE, 2021

    2021

  24. [32]

    Patchflow: A two-stage patch-based approach for lightweight optical flow estimation

    Ahmed Hammad Alhawwary, Janne Mustaniemi, and Janne Heikkila. Patchflow: A two-stage patch-based approach for lightweight optical flow estimation. In Proceedings of the Asian Conference on Computer Vision, pages 3740–3756, 2022

    2022

  25. [33]

    Deep equilibrium optical flow estimation

    Shaojie Bai, Zhengyang Geng, Yash Savani, and J Zico Kolter. Deep equilibrium optical flow estimation. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 620–630, 2022

    2022

  26. [34]

    Towards equivariant optical flow estimation with deep learning

    Stefano Savian, Pietro Morerio, Alessio Del Bue, Andrea A Janes, and Tammam Tillo. Towards equivariant optical flow estimation with deep learning. In Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, pages 5088–5097, 2023

    2023

  27. [35]

    Remote sensing image semantic change detection boosted by semi-supervised contrastive learning of semantic segmentation

    Xiuwei Zhang, Yizhe Yang, Lingyan Ran, Liang Chen, Kangwei Wang, Lei Yu, Peng Wang, and Yanning Zhang. Remote sensing image semantic change detection boosted by semi-supervised contrastive learning of semantic segmentation. IEEE Transactions on Geoscience and Remote Sensing, 2024

    2024

  28. [36]

    Difunet++: A satellite images change detection network based on unet++ and differential pyramid

    Xiuwei Zhang, Yuanzeng Yue, Wenxiang Gao, Shuai Yun, Qian Su, Hanlin Yin, and Yanning Zhang. Difunet++: A satellite images change detection network based on unet++ and differential pyramid. IEEE Geoscience and Remote Sensing Letters, 19:1–5, 2021

    2021

  29. [37]

    Adhr-cdnet: Attentive differential high-resolution change detection network for remote sensing images

    Xiuwei Zhang, Mu Tian, Yinghui Xing, Yuanzeng Yue, Yanping Li, Hanlin Yin, Runliang Xia, Jin Jin, and Yanning Zhang. Adhr-cdnet: Attentive differential high-resolution change detection network for remote sensing images. IEEE Transactions on Geoscience and Remote Sensing, 60:1–13, 2022

    2022

  30. [38]

    Deep learning in remote sensing applications: A meta-analysis and review

    Lei Ma, Yu Liu, Xueliang Zhang, Yuanxin Ye, Gaofei Yin, and Brian Alan Johnson. Deep learning in remote sensing applications: A meta-analysis and review. ISPRS journal of photogrammetry and remote sensing , 152:166–177, 2019

    2019

  31. [39]

    Deep learning for fluid velocity field estimation: A review

    Changdong Yu, Xiaojun Bi, and Yiwei Fan. Deep learning for fluid velocity field estimation: A review. Ocean Engineering, 271:113693, 2023

    2023

  32. [40]

    Fully convolutional networks for semantic segmentation

    Jonathan Long, Evan Shelhamer, and Trevor Darrell. Fully convolutional networks for semantic segmentation. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 3431–3440, 2015

    2015

  33. [41]

    Fully convolutional siamese networks for change detection

    Rodrigo Caye Daudt, Bertr Le Saux, and Alexandre Boulch. Fully convolutional siamese networks for change detection. In 2018 25th IEEE international conference on image processing (ICIP), pages 4063–4067. IEEE, 2018

    2018

  34. [42]

    Bsuv-net: A fully-convolutional neural network for background subtraction of unseen videos

    Ozan Tezcan, Prakash Ishwar, and Janusz Konrad. Bsuv-net: A fully-convolutional neural network for background subtraction of unseen videos. In Proceedings of the IEEE/CVF winter conference on applications of computer vision, pages 2774–2783, 2020

    2020

  35. [43]

    Research of moving object detection based on deep frame difference convolution neural network

    XF Ou, Peng-cheng YAN, Han-pu WANG, et al. Research of moving object detection based on deep frame difference convolution neural network. Acta Electron. Sin, 48(12):2384–2393, 2020. 17 A PREPRINT - S EPTEMBER 23, 2025

    2020

  36. [44]

    Multiple time scale motion images for action recognition

    Qin Cheng, Ziliang Ren, Jianming Liu, and Jun Cheng. Multiple time scale motion images for action recognition. In 2020 IEEE International Conference on E-health Networking, Application & Services (HEALTHCOM), pages 1–5. IEEE, 2021

    2020

  37. [45]

    Explicit change-relation learning for change detection in vhr remote sensing images

    Dalong Zheng, Zebin Wu, Jia Liu, Yang Xu, Chih-Cheng Hung, and Zhihui Wei. Explicit change-relation learning for change detection in vhr remote sensing images. IEEE Geoscience and Remote Sensing Letters, 2024

    2024

  38. [46]

    Progressive modality- alignment for unsupervised heterogeneous change detection

    Yinghui Xing, Qi Zhang, Lingyan Ran, Xiuwei Zhang, Hanlin Yin, and Yanning Zhang. Progressive modality- alignment for unsupervised heterogeneous change detection. IEEE Transactions on Geoscience and Remote Sensing, 61:1–12, 2023

    2023

  39. [47]

    Remote sensing image change detection based on deep dictionary learning

    Yuqun Yang, Xu Tang, Fang Liu, Jingjing Ma, and Licheng Jiao. Remote sensing image change detection based on deep dictionary learning. In IGARSS 2022-2022 IEEE International Geoscience and Remote Sensing Symposium, pages 1416–1419. IEEE, 2022

    2022

  40. [48]

    Deep siamese network with contextual transformer for remote sensing images change detection

    Mengxuan Zhang, Zhao Liu, Jie Feng, Licheng Jiao, and Long Liu. Deep siamese network with contextual transformer for remote sensing images change detection. In International Conference on Intelligence Science, pages 193–200. Springer, 2022

    2022

  41. [49]

    A hybrid method for remote sensing change detection

    Daobo Sun, Haohao Yu, Tao Dong, Xiangxu Meng, and Bin Tang. A hybrid method for remote sensing change detection. In 2023 2nd International Conference on Artificial Intelligence, Human-Computer Interaction and Robotics (AIHCIR), pages 215–219. IEEE, 2023

    2023

  42. [50]

    Building change detection using deep learning for remote sensing images

    Chang Wang, Shijing Han, Wen Zhang, and Shufeng Miao. Building change detection using deep learning for remote sensing images. Journal of Information Processing Systems, 18(4):587–598, 2022

    2022

  43. [51]

    Change detection by deep learning models

    Josephina Paul. Change detection by deep learning models. In 2022 IEEE International Women in Engineering (WIE) Conference on Electrical and Computer Engineering (WIECON-ECE), pages 323–326. IEEE, 2022

    2022

  44. [52]

    Semi supervised change detection method of remote sensing image

    Wei Nie, Peng Gou, Yang Liu, Bhaskar Shrestha, Tianyu Zhou, Nuo Xu, Peng Wang, and QiQi Du. Semi supervised change detection method of remote sensing image. In2022 IEEE 6th Advanced Information Technology, Electronic and Automation Control Conference (IAEAC), pages 1013–1019. ...

    2022

  45. [53]

    Changevit: Unleashing plain vision transformers for change detection

    Duowang Zhu, Xiaohu Huang, Haiyan Huang, Zhenfeng Shao, and Qimin Cheng. Changevit: Unleashing plain vision transformers for change detection. arXiv preprint arXiv:2406.12847, 2024

    2024 arXiv

  46. [54]

    Changeclip: Remote sensing change detection with multimodal vision-language representation learning

    Sijun Dong, Libo Wang, Bo Du, and Xiaoliang Meng. Changeclip: Remote sensing change detection with multimodal vision-language representation learning. ISPRS Journal of Photogrammetry and Remote Sensing, 208:53–69, 2024

    2024

  47. [55]

    Vision-language joint learning for box-supervised change detec- tion in remote sensing

    Kanghua Yin, Fang Liu, Jia Liu, and Liang Xiao. Vision-language joint learning for box-supervised change detec- tion in remote sensing. In IGARSS 2024-2024 IEEE International Geoscience and Remote Sensing Symposium, pages 10254–10258. IEEE, 2024

    2024

  48. [56]

    Sganet: A siamese geometry-aware network for remote sensing change detection

    Jiangwei Chen, Sijun Dong, and Xiaoliang Meng. Sganet: A siamese geometry-aware network for remote sensing change detection. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2025

    2025

  49. [57]

    Changead: Enhanced remote sensing change detection via bi-temporal alignment and differential feature integration

    Jiangtao Meng, Xinying Xu, Pengyue Li, and Zhe Zhang. Changead: Enhanced remote sensing change detection via bi-temporal alignment and differential feature integration. In 2024 China Automation Congress (CAC), pages 5391–5396. IEEE, 2024

    2024

  50. [58]

    Improving remote sensing change detection via locality induction on feed-forward vision transformer

    Lhuqita Fazry, Mgs M Luthfi Ramadhan, and Wisnu Jatmiko. Improving remote sensing change detection via locality induction on feed-forward vision transformer. Jurnal Ilmu Komputer dan Informasi (Journal of Computer Science and Information), 16(2), 2023

    2023

  51. [59]

    Tversky loss function for image segmentation using 3d fully convolutional deep networks

    Seyed Sadegh Mohseni Salehi, Deniz Erdogmus, and Ali Gholipour. Tversky loss function for image segmentation using 3d fully convolutional deep networks. In International workshop on machine learning in medical imaging, pages 379–387. Springer, 2017

    2017

  52. [60]

    The pascal visual object classes challenge: A retrospective

    Mark Everingham, SM Ali Eslami, Luc Van Gool, Christopher KI Williams, John Winn, and Andrew Zisserman. The pascal visual object classes challenge: A retrospective. International journal of computer vision, 111:98–136, 2015. 18

    2015

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.