Pith. sign in

REVIEW 5 major objections 5 minor 115 references

Dual form Complementary Masking for Domain-Adaptive Image Segmentation

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 ↗

Pith's one-line read This paper claims that masking each target image into two complementary halves, rather than two random masks, provably extracts domain-invariant features and yields state-of-the-art unsupervised domain adaptation for segmentation.

desk verdict Empirically useful complementary-masking trick for UDA, but the advertised theory self-destructs in its own appendix; reject current version, worth a look after stripping or fixing the proofs. read the letter →

arxiv 2507.12008 v1 pith:XH2ADCK2 submitted 2025-07-16 cs.CV cs.AI

Jiawen Wang , Yinda Chen , Xiaoyu Liu , Che Liu , Dong Liu , Jianqing Gao , Zhiwei Xiong This is my paper
classification cs.CVcs.AI
keywords unsuperviseddomainadaptationsemanticsegmentationcomplementarymaskingmaskedimagemodelingconsistencyregularizationcompressedsensingdomain-invariantfeatureselectronmicroscopy
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

This paper tries to put a theoretical foundation under a simple consistency trick for unsupervised domain adaptation (UDA) in segmentation: instead of masking images randomly, mask each target image into two complementary halves, $D \odot X$ and $(1-D) \odot X$, and force the network to produce the same segmentation prediction from both views. It reframes masked reconstruction as sparse signal recovery and claims that complementary masks preserve information better, give tighter generalization bounds, and yield more consistent features than random masks. These claims are packaged as MaskTwins, a UDA framework that adds no learnable parameters: an EMA teacher generates pseudo-labels from the unmasked image, and two consistency losses align the predictions of the complementary views. If the claims are right, complementary masking is a cheap, general-purpose regularizer for cross-domain segmentation, and the paper reports state-of-the-art results on SYNTHIA to Cityscapes as well as on electron-microscopy and synapse benchmarks.

What carries the argument

The load-bearing object is the complementary mask pair $(D, 1-D)$: a patch-wise binary mask $D$ with entries drawn from Bernoulli(0.5), so that every pixel of the image appears in exactly one of the two masked views. The paper's intellectual machinery is the reframing of masked image reconstruction as a sparse-signal recovery problem, importing compressed-sensing guarantees (restricted isometry property, basis pursuit denoising) to argue that complementary masks sample the input space more completely than random masks. In the network, the workhorse is prediction consistency: the student must output the same segmentation from $D \odot X$ and $(1-D) \odot X$, and both outputs must match the teacher pseudo-label, through the losses $L^T_{cl}$ and $L^T_{cm}$; no extra parameters are introduced, and the only added cost is the forward pass on the second masked view.

What would settle it

A direct check is to compute the paper's own information-preservation metric $\mathrm{IP}(x_1, x_2) = \langle x_1, x_2 \rangle / \|x\|^2$ on the masked inputs used in the argument: the paper's appendix (Section A.3) itself computes $\mathbb{E}[\mathrm{IP}] = 0$ for complementary masks and $1/4$ for random masks, which reverses the inequality in Eq. (4) as written; equivalently, measuring this metric on actual network features for a trained MaskTwins model would settle whether the claimed advantage survives the nonlinear feature map or whether a corrected definition of information preservation is needed.

Watch

Extended reading notes

Core claim

On the paper's own terms, the discovery is that the dual form of complementary masks—a Bernoulli(0.5) binary mask $D$ together with its complement $1-D$—is provably superior to independent random masks for extracting domain-agnostic image features. The proof strategy models an image as $X = S + E + N$ (sparse structure, environment, Gaussian noise) and treats masked reconstruction as compressed sensing: Theorem 1 asserts an information-preservation inequality $\mathbb{E}[IP(X_D, X_{1-D})] \ge \mathbb{E}[IP(X_{R_1}, X_{R_2})]$ with lower variance, Theorem 2 claims a tighter generalization bound for complementary masks, Theorem 3 bounds feature-consistency error, and Theorem 5 gives a compressed-sensing recovery guarantee. MaskTwins operationalizes these claims by training a student network so that its predictions on the two complementary views agree with each other and with EMA-teacher pseudo-labels of the unmasked image, using the losses $L^T_{cl}$ and $L^T_{cm}$; the authors report $76.7$ mIoU on SYNTHIA to Cityscapes, $+2.7$ over the previous best, with further gains on mitochondria and 3D synapse detection.

Load-bearing premise

The argument depends on assuming that the mathematical guarantee for recovering a sparse signal from complementary measurements also applies to what a deep segmentation network learns from the two masked images; the paper never connects its sparse-signal model $X = S + E + N$ to the network's learned features or to the actual losses $L^T_{cl}$ and $L^T_{cm}$, so if that bridge is missing, the theory does not cover the method it is meant to justify.

Editorial extensions

If this is right

  • Complementary-mask consistency is a parameter-free regularizer, so it can be dropped into existing UDA pipelines without changing the model architecture.
  • The masking strategy itself carries the gain: the ablation on SYNTHIA to Cityscapes shows that switching from random masks to complementary masks adds roughly +3.9 mIoU over the consistency-learning baseline.
  • The reported gains span natural images, electron microscopy, and 3D synapse detection, so the mechanism is not tied to one image modality or dimension.
  • The theory predicts tighter generalization bounds for complementary than for random masks, which would make masked-image-modeling pre-training unnecessary for the downstream adaptation gains.
  • A mask ratio of 0.5 and a patch size near 1/16 of the input give the best results, consistent with the claim that balanced, fully disjoint views are what matter.

Reading between the lines

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

  • Editorial inference: the printed Theorem 1 appears to conflict with the appendix's own computation—for the inner-product metric, complementary masks have exactly zero overlap while random masks have positive expected overlap—so the 'superiority' probably rests on zero variance and disjoint coverage being reinterpreted as diversity; a reader testing the theorem should ask which metric is meant.
  • Editorial inference: the compressed-sensing guarantee (Theorem 5) is about recovering a sparse coefficient vector from measurements, not about segmentation features; until one shows that the network's feature map or loss satisfies a sparse linear model with a restricted-isometry measurement matrix, the theory is a motivation for, rather than a proof of, the method's behavior.
  • Editorial inference: a clean way to test the mechanism is to interpolate mask overlap between fully complementary and fully independent random masks and measure target-domain mIoU; if performance falls monotonically as overlap rises, the disjoint-view story is supported, whereas a flat or peaked curve would point to other factors such as augmentation strength or pseudo-label quality.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

5 major / 5 minor

Summary. The paper proposes MaskTwins, an unsupervised domain adaptation (UDA) framework for semantic segmentation that enforces consistency between predictions on two complementary masked views of target images, with an EMA teacher generating pseudo-labels. The authors claim a theoretical foundation for complementary masking, restating masked reconstruction as sparse signal recovery and presenting three theorems (information preservation, generalization bound, and feature consistency) that assert complementary masks are provably superior to random masks for extracting domain-invariant features. The empirical section reports state-of-the-art results on SYNTHIA→Cityscapes (+2.7 mIoU over MIC), mitochondria segmentation, and 3D synapse detection, with ablations showing complementary masking outperforms random masking. The central advertised contribution is the theoretical proof, which the paper claims is the first proof of the capability of complementary masking for domain-invariant feature extraction.

Significance. The empirical component of the paper is solid and useful: MaskTwins is simple, introduces no extra learnable parameters, and shows consistent improvements across a broad set of UDA tasks (2D natural images, 2D electron microscopy, and 3D connectomics). The ablation in Table 4 directly compares complementary masking with random masking and reports a substantial gain (+1.5 mIoU on SYNTHIA→Cityscapes), which is a meaningful empirical finding independent of the theory. The addition of VisDA classification results and the three-task evaluation make the experimental study unusually broad. However, the theoretical claims are load-bearing for the paper's stated novelty, and several of them are contradicted by the paper's own appendix or are unproved. If the theory were correct, the paper would provide a clean, parameter-free justification for a simple regularizer; as it stands, the advertised proof is not valid.

major comments (5)
  1. [Section 3, Theorem 1 (Eqs. 4–5) and Appendix A.3] Theorem 1 is contradicted by the paper's own appendix. The appendix (A.2–A.3) defines IP(x1, x2) = ⟨x1, x2⟩/‖x‖² on masked inputs with no feature map f, and derives E[IP(complementary)] = 0 and E[IP(random)] = 1/4, which is the reverse of inequality (4). Furthermore, Appendix Remark 2 and Appendix C bullet 3 claim that complementary masks provide 'guaranteed preservation of a constant fraction of the original information,' which is inconsistent with the same appendix's computation IP = 0 for complementary masks. Because no condition on the feature map f is supplied that would invert the ordering for IP(X_D, X_{1−D}) = ⟨f(D⊙x), f((1−D)⊙x)⟩/‖f(x)‖², Theorem 1 is unproved as stated.
  2. [Section 3, Theorem 2 (Eqs. 6–7) and Appendix A, Theorem 4] The proof of the generalization bound contains a distribution mismatch and algebraic errors. In Theorem 4, case (ii) Step 2, the random-mask computation assumes each diagonal entry 'takes value 1 with probability 1/d,' whereas Definition 2 in both the main text and the appendix specifies independent Bernoulli(0.5) entries. Under the stated Bernoulli(0.5) distribution, E[‖R1 x‖²] = (1/2)‖x‖², not ‖x‖², so the claimed √d penalty for random masks has no valid derivation. The Rademacher complexity step also skips from a bound of 2LβB to 2LβB/√n without a rigorous argument. Independent of these errors, the theorem bounds the empirical risk of a generic objective L(f) = E[ℓ(f(x1), f(x2))]; it does not involve source and target distributions, and it cannot support the paper's claim of tighter domain-invariant generalization bounds.
  3. [Section 3, Theorem 3 (Eqs. 8–9)] The paper states in Section 3 that 'detailed proofs of all results are provided in Appendix A,' but Appendix A contains no proof of Theorem 3. The theorem's statement also introduces a quantity E without defining it in the theorem, and it does not explain how the environmental term from Assumption 1 enters only the random-mask bound. This theorem is asserted without derivation, so the claimed feature-consistency advantage is unsupported.
  4. [Section 3, Theorem 5 and Assumption 1] The compressed-sensing recovery guarantee is never connected to the losses L_T_cl (Eq. 13) and L_T_cm (Eq. 12) used in MaskTwins. The sparse signal S in Assumption 1, the dictionary M, and the recovered coefficient vector z in Theorem 5 are not mapped to network features, pseudo-labels, or the segmentation objective; nor is the RIP condition verified for the patch-wise complementary masks in Eq. (10). As written, Theorem 5 is a standalone compressed-sensing statement rather than a justification of the proposed method's feature consistency losses.
  5. [Section 3, overall] None of the theorems involves a source domain D_S, a target domain D_T, a domain discrepancy measure, or a notion of domain-invariant features. The theoretical section analyzes information preservation and consistency for a generic single-distribution feature-learning objective, so even a repaired version of Theorems 1–3 would not establish the abstract's claim that complementary masks are provably superior at extracting domain-invariant features for UDA. The domain-adaptation theory cited in Appendix A (Ben-David et al.; Zhang et al.) is not used in any proof.
minor comments (5)
  1. [Table 4 and §5.5] The checkmark layout of Table 4 is ambiguous: the columns are not clearly aligned with the row entries, and the main text's statement that 'random masking yields a improvement of +1.2 mIoU' is not directly traceable to the table. Please clarify the table formatting and the exact baseline used for that comparison.
  2. [Eq. (10)] The notation for patch-wise masking, D_{mb+1:(m+1)b}^{nb+1:(n+1)b} ∼ Bernoulli(1−r), appears to be a typesetting error; the intended matrix-index ranges should be written unambiguously.
  3. [§5.2] The text states 'our performance improvement on road is +4.8 from 91.2 to 96.0 IoU,' but the 91.2 baseline is not identifiable in Table 1; the best previous road IoU listed is 87.4 (CAMix) or 86.6 (MIC). Please correct the comparison value.
  4. [Appendix D] Appendix D contains a large number of references (e.g., Li et al. 2025b/c, Chen et al. 2025, Huang et al. 2025, Fu et al. 2025, Ma et al. 2024a/b, Yuan et al. 2024a/b and 2025a/b/c, Yu et al. 2024a/b and 2025a/b) that are not cited in the main text and are not used in the theory or experiments; trimming these would improve readability.
  5. [§5.1] In the synapse-detection paragraph, there is a missing period after 'λ_cm = 0.1 Empirically'; the sentence should be split as 'λ_cm = 0.1. Empirically, we set ...'.

Circularity Check

0 steps flagged · score 1.0 of 10

No circular derivation: complementary-mask consistency is an un-fitted regularizer evaluated on held-out benchmarks; the theoretical claims are internally inconsistent, but the empirical result is not equivalent to its inputs.

full rationale

MaskTwins' central empirical claim is that adding a complementary-masked consistency loss improves UDA segmentation. This loss (Eqs. 12-13, 17) is a fixed regularizer with no parameters fit to the target metric; the reported gains (Tables 1-3) are measured on standard benchmarks after training, so the improvement is not forced by construction. The paper's advertised proof that complementary masks preserve more information (Theorem 1 / Eq. 4) is not circular but is contradicted by the paper's own appendix: Appendix A.2 defines IP on raw masked inputs and A.3.1 computes E[IP] = 0 for complementary masks while A.3.2 computes E[IP] = 1/4 for random masks, the reverse of Eq. (4). Likewise, Theorem 2's proof uses a random-mask activation probability of 1/d inconsistent with Definition 2's Bernoulli(0.5), and Theorem 3 has no proof in the appendix; these are mathematical gaps, not reductions of the result to its assumptions. Several citations are to works with overlapping authorship (e.g., Chen et al. 2024a, Huang et al. 2022b, Yin et al. 2023), but they are used as baselines, dataset protocols, or related work, not as the justification for the central claim. Hyperparameters r = 0.5 and b = 64 are selected on the SYNTHIA→Cityscapes benchmark itself (Table 5), which is a test-set-selection concern, but choosing a mask ratio does not make the reported mIoU a fitted quantity by definition. No load-bearing step reduces by construction to its own inputs, so there is no significant circularity.

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

The paper introduces no new physical or ontological entities. The free parameters are hyperparameters tuned on the target test set; the axioms are the unproved or unconnected assumptions behind the theoretical claims.

free parameters (4)
  • mask ratio r = 0.5 (SYNTHIA→Cityscapes); searched over {0.1,0.2,0.3,0.4,0.5}
    Chosen by grid search on target test set (Table 5a); used across all experiments.
  • patch size b = 64 (2D natural), 32 (mitochondria), 6 (synapse)
    Set to 1/16 of input per task; tuned on SYNTHIA→Cityscapes in Table 5b.
  • complementary loss weight lambda_cm = 0.01 (natural, mitochondria), 0.1 (synapse)
    Hand-selected per task; no sensitivity analysis reported.
  • pseudo-label thresholds = delta=0.7 mito, delta=0.8 SYNTHIA, delta_pre=0.75, delta_post=0.65 synapse
    Set empirically following prior work; affect pseudo-label quality.
assumptions (4)
  • domain assumption Visual data model X = S + E + N with S sparse (Assumption 1)
    Stated but never used in the proofs; masks are applied to X, not to S.
  • standard math The feature extractor f is beta-Lipschitz/smooth and loss l is L-Lipschitz (Theorem 2)
    Standard regularity assumptions for Rademacher bounds, not verified for deep segmentation networks.
  • domain assumption Effective measurement matrix A satisfies RIP of order 2k (Theorem 5)
    Assumed without evidence for complementary mask matrices; no construction or verification.
  • domain assumption Consistency between complementary masked views reduces domain gap
    The core inductive bias of the method; not derived from the theorems.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Dual form Complementary Masking for Domain-Adaptive Image Segmentation." pith.science (2026). https://pith.science/paper/XH2ADCK2

@misc{pith2026250712008,
  author       = {Pith},
  title        = {Pith review of: Dual form Complementary Masking for Domain-Adaptive Image Segmentation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/XH2ADCK2}},
  note         = {Machine review of arXiv:2507.12008}
}
read the original abstract

Recent works have correlated Masked Image Modeling (MIM) with consistency regularization in Unsupervised Domain Adaptation (UDA). However, they merely treat masking as a special form of deformation on the input images and neglect the theoretical analysis, which leads to a superficial understanding of masked reconstruction and insufficient exploitation of its potential in enhancing feature extraction and representation learning. In this paper, we reframe masked reconstruction as a sparse signal reconstruction problem and theoretically prove that the dual form of complementary masks possesses superior capabilities in extracting domain-agnostic image features. Based on this compelling insight, we propose MaskTwins, a simple yet effective UDA framework that integrates masked reconstruction directly into the main training pipeline. MaskTwins uncovers intrinsic structural patterns that persist across disparate domains by enforcing consistency between predictions of images masked in complementary ways, enabling domain generalization in an end-to-end manner. Extensive experiments verify the superiority of MaskTwins over baseline methods in natural and biological image segmentation. These results demonstrate the significant advantages of MaskTwins in extracting domain-invariant features without the need for separate pre-training, offering a new paradigm for domain-adaptive segmentation.

Figures

Figures reproduced from arXiv: 2507.12008 by the authors.

Figure 1
Figure 1. The overall framework of MaskTwins. Given the labeled source data, we calculate the segmentation prediction [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Qualitative segmentation results on SYNTHIA [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Qualitative comparison of MaskTwins with previous methods on VNC III [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: More segmentation results on SYNTHIA→Cityscapes [PITH_FULL_IMAGE:figures/full_fig_p027_4.png]
Figure 5
Figure 5. Figure 5: Visualization of the volume in the WASPSYN dataset. Left to right: sections from X-Y, X-Z, and Y-Z plane. [PITH_FULL_IMAGE:figures/full_fig_p027_5.png]
Figure 6
Figure 6. Figure 6: An example of visualization of the detection results of pre-synapse (left) and post-synapse (right). Dots and lines: [PITH_FULL_IMAGE:figures/full_fig_p027_6.png]
Figure 7
Figure 7. Figure 7: More segmentation results on VNC III→Lucchi Subset1 (row 1 and 2), VNC III→Lucchi Subset2 (row 3 and 4), MitoEM-R→MitoEM-H (row 5 and 6) and MitoEM-H→MitoEM-R (row 7 and 8). The pixels in red and green denote the false-negative and false-positive segmentation results r…

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

115 extracted references · 61 canonical work pages

  1. [1]

    and Roth, S

    Araslanov, N. and Roth, S. Self-supervised augmentation consistency for adapting semantic segmentation. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp.\ 15384--15394, 2021

    2021

  2. [2]

    BEiT : BERT pre-training of image transformers

    Bao, H., Dong, L., Piao, S., and Wei, F. BEiT : BERT pre-training of image transformers. In International Conference on Learning Representations, 2022

    2022

  3. [3]

    Analysis of representations for domain adaptation

    Ben-David, S., Blitzer, J., Crammer, K., and Pereira, F. Analysis of representations for domain adaptation. Advances in neural information processing systems, 19, 2006

    2006

  4. [4]

    Ben-David, S., Blitzer, J., Crammer, K., Kulesza, A., Pereira, F., and Vaughan, J. W. A theory of learning from different domains. Machine learning, 79: 0 151--175, 2010

    2010

  5. [5]

    A domain-adaptive two-stream u-net for electron microscopy image segmentation

    Berm \'u dez-Chac \'o n, R., M \'a rquez-Neila, P., Salzmann, M., and Fua, P. A domain-adaptive two-stream u-net for electron microscopy image segmentation. In ISBI. IEEE, 2018

    2018

  6. [6]

    Brown, T. B. Language models are few-shot learners. arXiv preprint arXiv:2005.14165, 2020

    arXiv 2005

  7. [7]

    Domain adaptation for semantic segmentation with maximum squares loss

    Chen, M., Xue, H., and Cai, D. Domain adaptation for semantic segmentation with maximum squares loss. In ICCV, 2019

    2019

  8. [8]

    Domain adaptive synapse detection with weak point annotations

    Chen, Q., Huang, W., Zhang, Y., and Xiong, Z. Domain adaptive synapse detection with weak point annotations. In ISBI. IEEE, 2024 a

    2024

Show all 115 references
  1. [9]

    Self-supervised neuron segmentation with multi-agent reinforcement learning

    Chen, Y., Huang, W., Zhou, S., Chen, Q., and Xiong, Z. Self-supervised neuron segmentation with multi-agent reinforcement learning. In IJCAI, 2023

    2023

  2. [10]

    Learning multiscale consistency for self-supervised electron microscopy instance segmentation

    Chen, Y., Huang, W., Liu, X., Deng, S., Chen, Q., and Xiong, Z. Learning multiscale consistency for self-supervised electron microscopy instance segmentation. In ICASSP, pp.\ 1566--1570. IEEE, 2024 b

    2024

  3. [11]

    Bimcv-r: A landmark dataset for 3d ct text-image retrieval

    Chen, Y., Liu, C., Liu, X., Arcucci, R., and Xiong, Z. Bimcv-r: A landmark dataset for 3d ct text-image retrieval. In MICCAI, pp.\ 124--134. Springer Nature Switzerland, 2024 c

    2024

  4. [12]

    Tokenunify: Scalable autoregressive visual pre-training with mixture token prediction

    Chen, Y., Shi, H., Liu, X., Shi, T., Zhang, R., Liu, D., Xiong, Z., and Wu, F. Tokenunify: Scalable autoregressive visual pre-training with mixture token prediction. arXiv preprint arXiv:2405.16847, 2024 d

    2024 arXiv

  5. [13]

    Offset: Segmentation-based focus shift revision for composed image retrieval, 2025

    Chen, Z., Hu, Y., Li, Z., Fu, Z., Song, X., and Nie, L. Offset: Segmentation-based focus shift revision for composed image retrieval, 2025. URL https://arxiv.org/abs/2507.05631

    2025

  6. [14]

    Self-ensembling with gan-based data augmentation for domain adaptation in semantic segmentation

    Choi, J., Kim, T., and Kim, C. Self-ensembling with gan-based data augmentation for domain adaptation in semantic segmentation. In Proceedings of the IEEE/CVF international conference on computer vision, pp.\ 6830--6840, 2019

    2019

  7. [15]

    The cityscapes dataset for semantic urban scene understanding

    Cordts, M., Omran, M., Ramos, S., Rehfeld, T., Enzweiler, M., Benenson, R., Franke, U., Roth, S., and Schiele, B. The cityscapes dataset for semantic urban scene understanding. In CVPR, 2016

    2016

  8. [16]

    Unsupervised domain adaptation for em image denoising with invertible networks

    Deng, S., Chen, Y., Huang, W., Zhang, R., and Xiong, Z. Unsupervised domain adaptation for em image denoising with invertible networks. IEEE Transactions on Medical Imaging, 2024

    2024

  9. [17]

    Bert: Pre-training of deep bidirectional transformers for language understanding

    Devlin, J. Bert: Pre-training of deep bidirectional transformers for language understanding. arXiv preprint arXiv:1810.04805, 2018

    2018 arXiv

  10. [18]

    Improved regularization of convolutional neural networks with cutout

    DeVries, T. Improved regularization of convolutional neural networks with cutout. arXiv preprint arXiv:1708.04552, 2017

    2017 arXiv

  11. [19]

    Bootstrapped masked autoencoders for vision bert pretraining

    Dong, X., Bao, J., Zhang, T., Chen, D., Zhang, W., Yuan, L., Chen, D., Wen, F., and Yu, N. Bootstrapped masked autoencoders for vision bert pretraining. In ECCV, 2022

    2022

  12. [20]

    Peco: Perceptual codebook for bert pre-training of vision transformers

    Dong, X., Bao, J., Zhang, T., Chen, D., Zhang, W., Yuan, L., Chen, D., Wen, F., Yu, N., and Guo, B. Peco: Perceptual codebook for bert pre-training of vision transformers. In AAAI, 2023

    2023

  13. [21]

    Donoho, D. L. Compressed sensing. IEEE Transactions on Information Theory, 52 0 (4): 0 1289--1306, 2006

    2006

  14. [22]

    An image is worth 16x16 words: Transformers for image recognition at scale

    Dosovitskiy, A. An image is worth 16x16 words: Transformers for image recognition at scale. arXiv preprint arXiv:2010.11929, 2020

    2010 arXiv

  15. [23]

    Pair: Complementarity-guided disentanglement for composed image retrieval

    Fu, Z., Li, Z., Chen, Z., Wang, C., Song, X., Hu, Y., and Nie, L. Pair: Complementarity-guided disentanglement for composed image retrieval. In Proceedings of the IEEE International Conference on Acoustics, Speech and Signal Processing, pp.\ 1--5. IEEE, 2025

    2025

  16. [24]

    Segmented anisotropic sstem dataset of neural tissue

    Gerhard, S., Funke, J., Martel, J., Cardona, A., and Fetter, R. Segmented anisotropic sstem dataset of neural tissue. figshare, pp.\ 0--0, 2013

    2013

  17. [25]

    Generative adversarial nets

    Goodfellow, I., Pouget-Abadie, J., Mirza, M., Xu, B., Warde-Farley, D., Ozair, S., Courville, A., and Bengio, Y. Generative adversarial nets. Advances in neural information processing systems, 27, 2014

    2014

  18. [26]

    Domain adaptive video segmentation via temporal consistency regularization

    Guan, D., Huang, J., Xiao, A., and Lu, S. Domain adaptive video segmentation via temporal consistency regularization. In ICCV, 2021

    2021

  19. [27]

    Deep residual learning for image recognition

    He, K., Zhang, X., Ren, S., and Sun, J. Deep residual learning for image recognition. In CVPR, 2016

    2016

  20. [28]

    Masked autoencoders are scalable vision learners

    He, K., Chen, X., Xie, S., Li, Y., Doll \'a r, P., and Girshick, R. Masked autoencoders are scalable vision learners. In CVPR, 2022

    2022

  21. [29]

    Daformer: Improving network architectures and training strategies for domain-adaptive semantic segmentation

    Hoyer, L., Dai, D., and Van Gool, L. Daformer: Improving network architectures and training strategies for domain-adaptive semantic segmentation. In CVPR, 2022 a

    2022

  22. [30]

    Hrda: Context-aware high-resolution domain-adaptive semantic segmentation

    Hoyer, L., Dai, D., and Van Gool, L. Hrda: Context-aware high-resolution domain-adaptive semantic segmentation. In ECCV, 2022 b

    2022

  23. [31]

    Mic: Masked image consistency for context-enhanced domain adaptation

    Hoyer, L., Dai, D., Wang, H., and Van Gool, L. Mic: Masked image consistency for context-enhanced domain adaptation. In CVPR, 2023

    2023

  24. [32]

    Median: Adaptive intermediate-grained aggregation network for composed image retrieval

    Huang, Q., Chen, Z., Li, Z., Wang, C., Song, X., Hu, Y., and Nie, L. Median: Adaptive intermediate-grained aggregation network for composed image retrieval. In Proceedings of the IEEE International Conference on Acoustics, Speech and Signal Processing, pp.\ 1--5. IEEE, 2025

    2025

  25. [33]

    Semi-supervised neuron segmentation via reinforced consistency learning

    Huang, W., Chen, C., Xiong, Z., Zhang, Y., Chen, X., Sun, X., and Wu, F. Semi-supervised neuron segmentation via reinforced consistency learning. IEEE Transactions on Medical Imaging, 41 0 (11): 0 3016--3028, 2022 a

    2022

  26. [34]

    Domain adaptive mitochondria segmentation via enforcing inter-section consistency

    Huang, W., Liu, X., Cheng, Z., Zhang, Y., and Xiong, Z. Domain adaptive mitochondria segmentation via enforcing inter-section consistency. In MICCAI, 2022 b

    2022

  27. [35]

    and Belongie, S

    Huang, X. and Belongie, S. Arbitrary style transfer in real-time with adaptive instance normalization. In Proceedings of the IEEE International Conference on Computer Vision, 2017

    2017

  28. [36]

    Prototypical contrast adaptation for domain adaptive semantic segmentation

    Jiang, Z., Li, Y., Yang, C., Gao, P., Wang, Y., Tai, Y., and Wang, C. Prototypical contrast adaptation for domain adaptive semantic segmentation. In ECCV, pp.\ 36--54. Springer, 2022

    2022

  29. [37]

    Minimum class confusion for versatile domain adaptation

    Jin, Y., Wang, X., Long, M., and Wang, J. Minimum class confusion for versatile domain adaptation. In ECCV, 2020

    2020

  30. [38]

    Kingma, D. P. Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980, 2014

    2014 arXiv

  31. [39]

    and Zhang, X

    Kong, X. and Zhang, X. Understanding masked image modeling via learning occlusion invariant feature. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023

    2023

  32. [40]

    Lee, D.-H. et al. Pseudo-label: The simple and efficient semi-supervised learning method for deep neural networks. In ICML, 2013

    2013

  33. [41]

    Lee, K., Zung, J., Li, P., Jain, V., and Seung, H. S. Superhuman accuracy on the snemi3d connectomics challenge. arXiv preprint arXiv:1706.00120, 2017

    2017 arXiv

  34. [42]

    Content-consistent matching for domain adaptive semantic segmentation

    Li, G., Kang, G., Liu, W., Wei, Y., and Yang, Y. Content-consistent matching for domain adaptive semantic segmentation. In ECCV, 2020

    2020

  35. [43]

    Feature re-representation and reliable pseudo label retraining for cross-domain semantic segmentation

    Li, J., Zhou, K., Qian, S., Li, W., Duan, L., and Gao, S. Feature re-representation and reliable pseudo label retraining for cross-domain semantic segmentation. IEEE Transactions on Pattern Analysis and Machine Intelligence, 46 0 (3): 0 1682--1694, 2022

    2022

  36. [44]

    Deviation Wing Loss for High-Performance 2D Pose Estimation

    Li, L., Xing, J., Yu, X., and Zhang, X.-P. Deviation Wing Loss for High-Performance 2D Pose Estimation . In Proceedings of the IEEE International Conference on Multimedia and Expo, pp.\ 1--6, 2024 a

    2024

  37. [45]

    Translating Motion to Notation: Hand Labanotation for Intuitive and Comprehensive Hand Movement Documentation

    Li, L., Yang, W., Yu, X., Xing, J., and Zhang, X.-P. Translating Motion to Notation: Hand Labanotation for Intuitive and Comprehensive Hand Movement Documentation . In Proceedings of the ACM International Conference on Multimedia, pp.\ 4092--4100, 2024 b

    2024

  38. [46]

    Multi-View 3D Human Pose Estimation with Weakly Synchronized Images

    Li, L., Gu, R., Wang, C., Xing, J., Yu, X., and Zhang, X.-P. Multi-View 3D Human Pose Estimation with Weakly Synchronized Images . In Proceedings of the AAAI Conference on Artificial Intelligence, pp.\ 4833--4841, 2025 a

    2025

  39. [47]

    Bidirectional learning for domain adaptation of semantic segmentation

    Li, Y., Yuan, L., and Vasconcelos, N. Bidirectional learning for domain adaptation of semantic segmentation. In CVPR, 2019

    2019

  40. [48]

    Waspsyn: A challenge for domain adaptive synapse detection in microwasp brain connectomes

    Li, Y., Li, W., Chen, Q., Huang, W., Zou, Y., Xiao, X., Shinomiya, K., Gunn, P., Gupta, N., Polilov, A., et al. Waspsyn: A challenge for domain adaptive synapse detection in microwasp brain connectomes. IEEE Transactions on Medical Imaging, 2024 c

    2024

  41. [49]

    Encoder: Entity mining and modification relation binding for composed image retrieval

    Li, Z., Chen, Z., Wen, H., Fu, Z., Hu, Y., and Guan, W. Encoder: Entity mining and modification relation binding for composed image retrieval. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 39, pp.\ 5101--5109, 2025 b

    2025

  42. [50]

    Finecir: Explicit parsing of fine-grained modification semantics for composed image retrieval

    Li, Z., Fu, Z., Hu, Y., Chen, Z., Wen, H., and Nie, L. Finecir: Explicit parsing of fine-grained modification semantics for composed image retrieval. https://arxiv.org/abs/2503.21309, 2025 c

    2025 arXiv

  43. [51]

    C., Ma, L., Fu, J., Guo, Y., Shah, A., Bai, W., and Arcucci, R

    Liu, C., Ouyang, C., Chen, Y., Quilodr \'a n-Casas, C. C., Ma, L., Fu, J., Guo, Y., Shah, A., Bai, W., and Arcucci, R. T3d: Towards 3d medical image understanding through vision-language pre-training. arXiv preprint arXiv:2312.01529, 2023 a

    2023 arXiv

  44. [52]

    Can medical vision-language pre-training succeed with purely synthetic data? In ACL, 2025

    Liu, C., Wan, Z., Wang, H., Chen, Y., Qaiser, T., Jin, C., Yousefi, F., Burlutskiy, N., and Arcucci, R. Can medical vision-language pre-training succeed with purely synthetic data? In ACL, 2025

    2025

  45. [53]

    Pdam: A panoptic-level feature alignment framework for unsupervised domain adaptive instance segmentation in microscopy images

    Liu, D., Zhang, D., Song, Y., Zhang, F., O’Donnell, L., Huang, H., Chen, M., and Cai, W. Pdam: A panoptic-level feature alignment framework for unsupervised domain adaptive instance segmentation in microscopy images. IEEE Transactions on Medical Imaging, 40 0 (1): 0 154--165, 2020 a

    2020

  46. [54]

    The devil is in the frequency: Geminated gestalt autoencoder for self-supervised visual pre-training

    Liu, H., Jiang, X., Li, X., Guo, A., Hu, Y., Jiang, D., and Ren, B. The devil is in the frequency: Geminated gestalt autoencoder for self-supervised visual pre-training. In AAAI, 2023 b

    2023

  47. [55]

    Importance-aware semantic segmentation in self-driving with discrete wasserstein training

    Liu, X., Han, Y., Bai, S., Ge, Y., Wang, T., Han, X., Li, S., You, J., and Lu, J. Importance-aware semantic segmentation in self-driving with discrete wasserstein training. In AAAI, 2020 b

    2020

  48. [56]

    Advancing Video Synchronization with Fractional Frame Analysis: Introducing a Novel Dataset and Model

    Liu, Y., Ai, H., Xing, J., Li, X., Wang, X., and Tao, P. Advancing Video Synchronization with Fractional Frame Analysis: Introducing a Novel Dataset and Model . In Proceedings of the AAAI Conference on Artificial Intelligence, pp.\ 3828--3836, 2024

    2024

  49. [57]

    Long, M., Cao, Z., Wang, J., and Jordan, M. I. Conditional adversarial domain adaptation. Advances in neural information processing systems, 31, 2018

    2018

  50. [58]

    Decoupled weight decay regularization

    Loshchilov, I. Decoupled weight decay regularization. arXiv preprint arXiv:1711.05101, 2017

    2017 arXiv

  51. [59]

    Learning for structured prediction using approximate subgradient descent with working sets

    Lucchi, A., Li, Y., and Fua, P. Learning for structured prediction using approximate subgradient descent with working sets. In CVPR, 2013

    2013

  52. [60]

    Category-level adversarial adaptation for semantic segmentation using purified features

    Luo, Y., Liu, P., Zheng, L., Guan, T., Yu, J., and Yang, Y. Category-level adversarial adaptation for semantic segmentation using purified features. IEEE Transactions on Pattern Analysis and Machine Intelligence, 44 0 (8): 0 3940--3956, 2021

    2021

  53. [61]

    Tinyvim: Frequency decoupling for tiny hybrid vision mamba, 2024 a

    Ma, X., Ni, Z., and Chen, X. Tinyvim: Frequency decoupling for tiny hybrid vision mamba, 2024 a . URL https://arxiv.org/abs/2411.17473

    2024

  54. [62]

    SSA -seg: Semantic and spatial adaptive pixel-level classifier for semantic segmentation

    Ma, X., Ni, Z.-L., and Chen, X. SSA -seg: Semantic and spatial adaptive pixel-level classifier for semantic segmentation. In The Thirty-eighth Annual Conference on Neural Information Processing Systems, 2024 b . URL https://openreview.net/forum?id=RZZo23pQFL

    2024

  55. [63]

    A novel scene coupling semantic mask network for remote sensing image segmentation

    Ma, X., Lian, R., Wu, Z., Guan, R., Hong, T., Zhao, M., Ma, M., Nie, J., Du, Z., Song, S., and Zhang, W. A novel scene coupling semantic mask network for remote sensing image segmentation. ISPRS Journal of Photogrammetry and Remote Sensing, 221: 0 44--63, 2025. ISSN 0924-2716....

    2025 doi

  56. [64]

    Matthews, B. W. Comparison of the predicted and observed secondary structure of t4 phage lysozyme. Biochimica et Biophysica Acta (BBA)-Protein Structure, 405 0 (2): 0 442--451, 1975

    1975

  57. [65]

    Instance adaptive self-training for unsupervised domain adaptation

    Mei, K., Zhu, C., Zou, J., and Zhang, S. Instance adaptive self-training for unsupervised domain adaptation. In ECCV, 2020

    2020

  58. [66]

    and Manrai, A

    Melas-Kyriazi, L. and Manrai, A. K. Pixmatch: Unsupervised domain adaptation via pixelwise consistency training. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp.\ 12435--12445, 2021

    2021

  59. [67]

    Unsupervised mitochondria segmentation in em images via domain adaptive multi-task learning

    Peng, J., Yi, J., and Yuan, Z. Unsupervised mitochondria segmentation in em images via domain adaptive multi-task learning. IEEE Journal of Selected Topics in Signal Processing, 14 0 (6): 0 1199--1209, 2020

    2020

  60. [68]

    Visda: The visual domain adaptation challenge

    Peng, X., Usman, B., Kaushik, N., Hoffman, J., Wang, D., and Saenko, K. Visda: The visual domain adaptation challenge. arXiv preprint arXiv:1710.06924, 2017

    2017 arXiv

  61. [69]

    S., Jin, X., and Fan, D.-P

    Qian, H., Chen, Y., Lou, S., Khan, F. S., Jin, X., and Fan, D.-P. Maskfactory: Towards high-quality synthetic data generation for dichotomous image segmentation. NeurIPS, 2024

    2024

  62. [70]

    K., Mishra, M., Jain, A., and Radhakrishnan, V

    Rangwani, H., Aithal, S. K., Mishra, M., Jain, A., and Radhakrishnan, V. B. A closer look at smoothness in domain adversarial training. In ICML, 2022

    2022

  63. [71]

    Ros, G., Sellart, L., Materzynska, J., Vazquez, D., and Lopez, A. M. The synthia dataset: A large collection of synthetic images for semantic segmentation of urban scenes. In CVPR, 2016

    2016

  64. [72]

    S., and Oh, J

    Shin, U., Lee, K., Kweon, I. S., and Oh, J. Complementary random masking for rgb-thermal semantic segmentation. In ICRA, 2024

    2024

  65. [73]

    and Valpola, H

    Tarvainen, A. and Valpola, H. Mean teachers are better role models: Weight-averaged consistency targets improve semi-supervised deep learning results. Advances in neural information processing systems, 30, 2017

    2017

  66. [74]

    Dacs: Domain adaptation via cross-domain mixed sampling

    Tranheden, W., Olsson, V., Pinto, J., and Svensson, L. Dacs: Domain adaptation via cross-domain mixed sampling. In WACV, 2021

    2021

  67. [75]

    Learning to adapt structured output space for semantic segmentation

    Tsai, Y.-H., Hung, W.-C., Schulter, S., Sohn, K., Yang, M.-H., and Chandraker, M. Learning to adapt structured output space for semantic segmentation. In CVPR, 2018

    2018

  68. [76]

    Dada: Depth-aware domain adaptation in semantic segmentation

    Vu, T.-H., Jain, H., Bucher, M., Cord, M., and P \'e rez, P. Dada: Depth-aware domain adaptation in semantic segmentation. In ICCV, 2019

    2019

  69. [77]

    Masked image modeling with local multi-scale reconstruction

    Wang, H., Tang, Y., Wang, Y., Guo, J., Deng, Z.-H., and Han, K. Masked image modeling with local multi-scale reconstruction. In CVPR, 2023

    2023

  70. [78]

    Uncertainty-aware pseudo label refinery for domain adaptive semantic segmentation

    Wang, Y., Peng, J., and Zhang, Z. Uncertainty-aware pseudo label refinery for domain adaptive semantic segmentation. In ICCV, 2021

    2021

  71. [79]

    Masked feature prediction for self-supervised visual pre-training

    Wei, C., Fan, H., Xie, S., Wu, C.-Y., Yuille, A., and Feichtenhofer, C. Masked feature prediction for self-supervised visual pre-training. In CVPR, 2022

    2022

  72. [80]

    Mitoem dataset: Large-scale 3d mitochondria instance segmentation from em images

    Wei, D., Lin, Z., Franco-Barranco, D., Wendt, N., Liu, X., Yin, W., Huang, X., Gupta, A., Jang, W.-D., Wang, X., et al. Mitoem dataset: Large-scale 3d mitochondria instance segmentation from em images. In MICCAI, 2020

    2020

  73. [81]

    Uncertainty-aware label rectification for domain adaptive mitochondria segmentation

    Wu, S., Chen, C., Xiong, Z., Chen, X., and Sun, X. Uncertainty-aware label rectification for domain adaptive mitochondria segmentation. In MICCAI, 2021

    2021

  74. [82]

    Conditional latent coding with learnable synthesized reference for deep image compression

    Wu, S., Chen, Y., Liu, D., and He, Z. Conditional latent coding with learnable synthesized reference for deep image compression. AAAI, 2025

    2025

  75. [83]

    M., and Luo, P

    Xie, E., Wang, W., Yu, Z., Anandkumar, A., Alvarez, J. M., and Luo, P. Segformer: Simple and efficient design for semantic segmentation with transformers. Advances in neural information processing systems, 34: 0 12077--12090, 2021

    2021

  76. [84]

    S., and Loy, C

    Xie, J., Li, W., Zhan, X., Liu, Z., Ong, Y. S., and Loy, C. C. Masked frequency modeling for self-supervised visual pre-training. arXiv preprint arXiv:2206.07706, 2022 a

    2022 arXiv

  77. [85]

    Simmim: A simple framework for masked image modeling

    Xie, Z., Zhang, Z., Cao, Y., Lin, Y., Bao, J., Yao, Z., Dai, Q., and Hu, H. Simmim: A simple framework for masked image modeling. In CVPR, 2022 b

    2022

  78. [86]

    Cdtrans: Cross-domain transformer for unsupervised domain adaptation

    Xu, T., Chen, W., Wang, P., Wang, F., Li, H., and Jin, R. Cdtrans: Cross-domain transformer for unsupervised domain adaptation. arXiv preprint arXiv:2109.06165, 2021

    2021 arXiv

  79. [87]

    Label-driven reconstruction for domain adaptation in semantic segmentation

    Yang, J., An, W., Wang, S., Zhu, X., Yan, C., and Huang, J. Label-driven reconstruction for domain adaptation in semantic segmentation. In ECCV, 2020

    2020

  80. [88]

    Context-aware domain adaptation in semantic segmentation

    Yang, J., An, W., Yan, C., Zhao, P., and Huang, J. Context-aware domain adaptation in semantic segmentation. In WACV, 2021

    2021

  81. [89]

    Tvt: Transferable vision transformer for unsupervised domain adaptation

    Yang, J., Liu, J., Xu, N., and Huang, J. Tvt: Transferable vision transformer for unsupervised domain adaptation. In WACV, 2023

    2023

  82. [90]

    Micdrop: masking image and depth features via complementary dropout for domain-adaptive semantic segmentation

    Yang, L., Hoyer, L., Weber, M., Fischer, T., Dai, D., Leal-Taix \'e , L., Pollefeys, M., Cremers, D., and Van Gool, L. Micdrop: masking image and depth features via complementary dropout for domain-adaptive semantic segmentation. In European Conference on Computer Vision, pp.\...

    2025

  83. [91]

    Class-aware feature alignment for domain adaptative mitochondria segmentation

    Yin, D., Huang, W., Xiong, Z., and Chen, X. Class-aware feature alignment for domain adaptative mitochondria segmentation. In MICCAI. Springer, 2023

    2023

  84. [92]

    Ich-scnet: Intracerebral hemorrhage segmentation and prognosis classification network using clip-guided sam mechanism

    Yu, X., Elazab, A., Ge, R., Jin, H., Jiang, X., Jia, G., Wu, Q., Shi, Q., and Wang, C. Ich-scnet: Intracerebral hemorrhage segmentation and prognosis classification network using clip-guided sam mechanism. In 2024 IEEE International Conference on Bioinformatics and Biomedicine...

    2024

  85. [93]

    Ichpro: Intracerebral hemorrhage prognosis classification via joint-attention fusion-based 3d cross-modal network

    Yu, X., Li, X., Ge, R., Wu, S., Elazab, A., Zhu, J., Zhang, L., Jia, G., Xu, T., Wan, X., et al. Ichpro: Intracerebral hemorrhage prognosis classification via joint-attention fusion-based 3d cross-modal network. In 2024 IEEE International Symposium on Biomedical Imaging (ISBI)...

    2024

  86. [94]

    Ich-prnet: a cross-modal intracerebral haemorrhage prognostic prediction method using joint-attention interaction mechanism

    Yu, X., Elazab, A., Ge, R., Zhu, J., Zhang, L., Jia, G., Wu, Q., Wan, X., Li, L., and Wang, C. Ich-prnet: a cross-modal intracerebral haemorrhage prognostic prediction method using joint-attention interaction mechanism. Neural Networks, 184: 0 107096, 2025 a

    2025

  87. [95]

    Crisp-sam2: Sam2 with cross-modal interaction and semantic prompting for multi-organ segmentation

    Yu, X., Wang, C., Jin, H., Elazab, A., Jia, G., Wan, X., Zou, C., and Ge, R. Crisp-sam2: Sam2 with cross-modal interaction and semantic prompting for multi-organ segmentation. arXiv preprint arXiv:2506.23121, 2025 b

    2025 arXiv

  88. [96]

    Sd-mvs: Segmentation-driven deformation multi-view stereo with spherical refinement and em optimization

    Yuan, Z., Cao, J., Li, Z., Jiang, H., and Wang, Z. Sd-mvs: Segmentation-driven deformation multi-view stereo with spherical refinement and em optimization. In AAAI, volume 38, pp.\ 6871--6880, 2024 a

    2024

  89. [97]

    Tsar-mvs: Textureless-aware segmentation and correlative refinement guided multi-view stereo

    Yuan, Z., Cao, J., Wang, Z., and Li, Z. Tsar-mvs: Textureless-aware segmentation and correlative refinement guided multi-view stereo. Pattern Recognition, 154: 0 110565, 2024 b

    2024

  90. [98]

    MSP-MVS : Multi-granularity segmentation prior guided multi-view stereo

    Yuan, Z., Liu, C., Shen, F., Li, Z., Luo, J., Mao, T., and Wang, Z. MSP-MVS : Multi-granularity segmentation prior guided multi-view stereo. In Proceedings of the AAAI Conference on Artificial Intelligence , volume 39, pp.\ 9753--9762, 2025 a

    2025

  91. [99]

    DVP-MVS : Synergize depth-edge and visibility prior for multi-view stereo

    Yuan, Z., Luo, J., Shen, F., Li, Z., Liu, C., Mao, T., and Wang, Z. DVP-MVS : Synergize depth-edge and visibility prior for multi-view stereo. In Proceedings of the AAAI Conference on Artificial Intelligence , volume 39, pp.\ 9743--9752, 2025 b

    2025

  92. [100]

    Sed-mvs: Segmentation-driven and edge-aligned deformation multi-view stereo with depth restoration and occlusion constraint

    Yuan, Z., Yang, Z., Cai, Y., Wu, K., Liu, M., Zhang, D., Jiang, H., Li, Z., and Wang, Z. Sed-mvs: Segmentation-driven and edge-aligned deformation multi-view stereo with depth restoration and occlusion constraint. IEEE Transactions on Circuits and Systems for Video Technology, 2025 c

    2025

  93. [101]

    Prototypical pseudo label denoising and target structure learning for domain adaptive semantic segmentation

    Zhang, P., Zhang, B., Zhang, T., Chen, D., Wang, Y., and Wen, F. Prototypical pseudo label denoising and target structure learning for domain adaptive semantic segmentation. In CVPR, 2021

    2021

  94. [102]

    Category anchor-guided unsupervised domain adaptation for semantic segmentation

    Zhang, Q., Zhang, J., Liu, W., and Tao, D. Category anchor-guided unsupervised domain adaptation for semantic segmentation. Advances in neural information processing systems, 32, 2019 a

    2019

  95. [103]

    Bridging theory and algorithm for domain adaptation

    Zhang, Y., Liu, T., Long, M., and Jordan, M. Bridging theory and algorithm for domain adaptation. In ICML, 2019 b

    2019

  96. [104]

    Learning pseudo-relations for cross-domain semantic segmentation

    Zhao, D., Wang, S., Zang, Q., Quan, D., Ye, X., Yang, R., and Jiao, L. Learning pseudo-relations for cross-domain semantic segmentation. In ICCV, 2023

    2023

  97. [105]

    and Yang, Y

    Zheng, Z. and Yang, Y. Rectifying pseudo label learning via uncertainty estimation for domain adaptive semantic segmentation. International Journal of Computer Vision, 129 0 (4): 0 1106--1120, 2021

    2021

  98. [106]

    and Yang, Y

    Zheng, Z. and Yang, Y. Adaptive boosting for domain adaptation: Toward robust predictions in scene segmentation. IEEE Transactions on Image Processing, 31: 0 5371--5382, 2022

    2022

  99. [107]

    ibot: Image bert pre-training with online tokenizer

    Zhou, J., Wei, C., Wang, H., Shen, W., Xie, C., Yuille, A., and Kong, T. ibot: Image bert pre-training with online tokenizer. arXiv preprint arXiv:2111.07832, 2021

    2021 arXiv

  100. [108]

    Uncertainty-aware consistency regularization for cross-domain semantic segmentation

    Zhou, Q., Feng, Z., Gu, Q., Cheng, G., Lu, X., Shi, J., and Ma, L. Uncertainty-aware consistency regularization for cross-domain semantic segmentation. Computer Vision and Image Understanding, 221: 0 103448, 2022 a

    2022

  101. [109]

    Context-aware mixup for domain adaptive semantic segmentation

    Zhou, Q., Feng, Z., Gu, Q., Pang, J., Cheng, G., Lu, X., Shi, J., and Ma, L. Context-aware mixup for domain adaptive semantic segmentation. IEEE Transactions on Circuits and Systems for Video Technology, 33 0 (2): 0 804--817, 2022 b

    2022

  102. [110]

    Affinity space adaptation for semantic segmentation across domains

    Zhou, W., Wang, Y., Chu, J., Yang, J., Bai, X., and Xu, Y. Affinity space adaptation for semantic segmentation across domains. IEEE Transactions on Image Processing, 30: 0 2549--2561, 2020

    2020

  103. [111]

    Unsupervised domain adaptation for semantic segmentation via class-balanced self-training

    Zou, Y., Yu, Z., Kumar, B., and Wang, J. Unsupervised domain adaptation for semantic segmentation via class-balanced self-training. In ECCV, 2018

    2018

  104. [112]

    Confidence regularized self-training

    Zou, Y., Yu, Z., Liu, X., Kumar, B., and Wang, J. Confidence regularized self-training. In CVPR, 2019

    2019

  105. [113]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 gl...

  106. [114]

    J., Romberg, J., & Tao, T

    Candès, E. J., Romberg, J., & Tao, T. (2006). Stable signal recovery from incomplete and inaccurate measurements. Communications on Pure and Applied Mathematics, 59(8), 1207--1223

    2006

  107. [115]

    Laurent, B., & Massart, P. (2000). Adaptive estimation of a quadratic functional by model selection. Annals of Statistics, 28(5), 1302--1338

    2000

Pith tools

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