Pith. sign in

REVIEW 4 major objections 6 minor 30 references

AFRDA: Attentive Feature Refinement for Domain Adaptive Semantic Segmentation

T0 review · 4 major / 6 minor · reviewed 2026-08-06 · deepseek-v4-flash

Pith's one-line read This paper introduces a lightweight attention module, AFR, that refines high-resolution features using low-resolution semantic logits, uncertainty maps, and high-frequency boundary cues, and reports that plugging it into HRDA-based UDA…

desk verdict The plug-in module is well-specified and the ablations are internally consistent, but the 'state-of-the-art' claim collapses when you compare AFRDA's numbers to the officially published MIC results. read the letter →

arxiv 2507.17957 v2 pith:N7PFXCRD submitted 2025-07-23 cs.CV

classification cs.CV
keywords unsuperviseddomainadaptationsemanticsegmentationfeaturerefinementattentionmechanismuncertaintyestimationhigh-frequencyfeaturesself-trainingroboticnavigation
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper argues that unsupervised domain-adaptive semantic segmentation can be improved without changing the training pipeline, by adding a small module that refines high-resolution features using information that already exists in the network: low-resolution class logits, softmax-based uncertainty, and high-frequency residual signals. This matters because models trained on synthetic data often fail on real images, and refinements that improve boundary and small-object accuracy could make such models safer for robot navigation. The authors show the module, called AFR, can be inserted into existing HRDA-style methods, and report consistent gains of about one mIoU point on two standard urban benchmarks as well as qualitative gains in an off-road forest setting with a real robot.

What carries the argument

The load-bearing object is the Attentive Feature Refinement (AFR) module, a dual-attention block with two heads: Class-Aware Logits-Based Attention (CALA) and Uncertainty-Suppressed HR Feature Attention (UHFA). CALA compresses low-resolution logits to an attention map, multiplies it by the sigmoid of high-resolution feature uncertainty, adds the high-frequency residual of the low-resolution logits, and passes the sum through a sigmoid. UHFA globally pools high-resolution features, adds their Gaussian high-frequency residual, applies a 3×3 spatial attention convolution, multiplies by the exponential of the negative low-resolution logit uncertainty, and applies a sigmoid. The two maps are combined as $A_{\text{final}} = \alpha A_1 + (1-\alpha) A_2$ with a learnable $\alpha$, and the refined features are $F_{\text{HR}} \odot A_{\text{final}} + F_{\text{HR}}$. The Gaussian-filter high-frequency extraction removes the need for an explicit boundary head.

What would settle it

Run the GTA V→Cityscapes adaptation with the same settings and batch size as the paper (MIC baseline vs MIC+AFR) across at least three random seeds. If the mean gain over MIC is less than one standard deviation or the intervals overlap, the claimed improvement would not be distinguishable from seed noise. A second check: replace the Gaussian high-frequency extraction in AFR with a Sobel-style edge operator; if mIoU does not drop, the paper's boundary-refinement story is not the active ingredient.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is that low-resolution logits—not just encoder features—can serve as semantic priors to guide high-resolution feature refinement in UDA semantic segmentation. The AFR module computes two attention maps: one (CALA) that multiplies a sigmoid-compressed low-resolution logit map with the uncertainty of high-resolution features, and another (UHFA) that combines global and high-frequency components of the high-resolution features suppressed by low-resolution logit uncertainty. These are fused by a learnable weight alpha and applied to high-resolution features with a residual connection. With this module, MIC+AFR reaches 76.60 mIoU on GTA V→Cityscapes and 68.30 mIoU on Synthia→Cityscapes, gains of +1.05 and +1.04 mIoU over the MIC baseline, while adding only a few percent to training time and no extra labels or boundary heads.

Load-bearing premise

The central load-bearing assumption is that the reported roughly one-point mIoU gains are real improvements rather than random variation, since they come from a single training run at batch size 2 with no reported seeds, and that the softmax-based uncertainty maps guide attention without amplifying pseudo-label noise.

Editorial extensions

If this is right

  • Plugging AFR into other HRDA-based UDA methods should yield similar per-point gains, as demonstrated with HRDA (+0.76 and +0.88 mIoU) and ERF (+0.52 and +0.70 mIoU).
  • Boundary quality and small-object accuracy improve without additional labels or a boundary prediction head, so the approach is usable where annotation is scarce.
  • The module keeps training and inference cost nearly unchanged, making it practical for robot deployment; the paper reports a full navigation pipeline at 0.77 s per frame on an RTX 2060.
  • Because AFR leaves the training loss and pseudo-labeling pipeline untouched, it can be combined with future improvements in pseudo-label filtering or data mixing.

Reading between the lines

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

  • If the reported gains are real, the same logit-guided refinement idea could transfer to monocular depth estimation or instance segmentation under domain shift, where low-resolution semantic logits are also available.
  • The softmax-derived uncertainty maps are a cheap proxy; calibrated or ensemble uncertainties might yield larger gains in high-noise pseudo-label regions, a testable variant.
  • The improvements are reported from single runs with batch size 2 and no seed variance, so the true effect size is uncertain; a multi-seed replication would separate the method's contribution from random variation.
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

4 major / 6 minor

Summary. The paper proposes AFRDA, an Attentive Feature Refinement module for unsupervised domain-adaptive semantic segmentation built on HRDA-based self-training frameworks. The AFR module refines high-resolution features using low-resolution semantic logits, high-frequency boundary cues, and uncertainty-driven attention, with two branches (CALA and UHFA) fused by a learnable parameter. The authors report improvements over MIC of +1.05 mIoU on GTA V→Cityscapes and +1.04 mIoU on Synthia→Cityscapes, as well as plug-in gains on HRDA and ERF, plus qualitative off-road results and a real-robot navigation deployment with the POVNav planner.

Significance. If the reported gains are reproducible, the AFR module is a useful lightweight plug-in: it is described with explicit equations, the ablations are internally consistent, and the code is released. The module's design—using low-resolution logits and softmax-derived uncertainty to refine high-resolution features—is a plausible and potentially transferable idea for HRDA-based UDA methods. However, the central state-of-the-art claim currently rests on a comparison against MIC baselines that are notably below the published MIC numbers, and the experiments are single-run without reported training schedules or seeds. The significance of the contribution therefore depends on resolving the baseline discrepancy and demonstrating that the gains exceed seed-level variance; as presented, the contribution is incremental and not yet validated at the claimed SOTA level.

major comments (4)
  1. [§IV-B, Tables I and II] The reported MIC baselines are not the numbers used by the community. Tables I and II list MIC at 75.55 and 67.26 mIoU, whereas the published MIC paper reports 77.2 and 68.5 mIoU for the same GTA V→Cityscapes and Synthia→Cityscapes settings with the same HRDA protocol (batch size 2, crop 952). Since AFRDA's absolute results (76.60 and 68.30) are lower than those published MIC numbers, the claims that AFRDA 'outperforms the baseline MIC by +1.05 mIoU' and 'outperforms all SOTA methods' are not supported by the comparison as presented. Please either reproduce the official MIC configuration exactly and document the training schedule, or compare against the published MIC numbers and revise the claims accordingly.
  2. [§IV-A.2 and Tables I–II] No training schedule, number of iterations, learning rate, EMA decay, or random seeds are reported, and all mIoU numbers appear to come from single runs with a batch size of 2. A +1.05/+1.04 mIoU improvement over a baseline can easily fall within seed-level variance in this setting. Please report mean and standard deviation over at least three seeds and provide the full training hyperparameters so that the central improvement claim can be evaluated.
  3. [§III-B, Eqs. (9)–(10) and (16)] The uncertainty maps U_HR and U_LR are load-bearing components of the attention equations but are never defined. The text only says that uncertainty maps are 'estimated from softmax probabilities' without giving a formula or specifying the input resolution. Without this definition, the AFR equations are not reproducible, and the ablations in Table IV that remove 'HR Uncertainty' or 'LR Logits Uncertainty' cannot be properly interpreted.
  4. [§IV-D] The off-road and navigation evidence is only qualitative: the MESH dataset has no ground truth, and the navigation mission reports no quantitative metrics such as success rate, number of trials, path deviation, or comparison against a baseline planner. Since the abstract and introduction claim 'accurate, stable perception of traversable ground and obstacles, supporting reliable autonomous navigation', please provide quantitative results or explicitly limit the claim to a qualitative demonstration.
minor comments (6)
  1. [Table I caption] The word 'Quantitive' in the caption of Table I should be corrected to 'Quantitative'.
  2. [§III-B, around Eq. (14)] The sentence stating that G2D_gamma is calculated using Eq. (11) should refer to Eq. (12), where the Gaussian filter is actually defined.
  3. [§III-B, Eq. (12)] The values of the Gaussian filter parameters gamma and kernel size k are not reported anywhere; please state the chosen values and, ideally, a sensitivity study showing that the method is stable across reasonable choices.
  4. [Table IV] The ablation labels such as 'AFR w/o HR Uncertainty (CALA)' and 'AFR w/o LR Logits Uncertainty (UHFA)' do not specify how each component is disabled; please state whether the corresponding term is set to a constant, removed, or replaced.
  5. [Table IV] The counterintuitive ordering in which removing boundary refinement from CALA alone (75.20) hurts more than removing it from both modules (75.58) is explained in the text, but a more rigorous interaction analysis would help readers trust that this is not an artifact of training noise.
  6. [§IV-B] The claim that AFRDA 'outperforms all SOTA methods' is based on a comparison set that includes only ADVENT, DACS, ProDA, DAFormer, HRDA, MIC, and ERF; please compare against the current leaderboard of HRDA-based methods to support the state-of-the-art statement.

Circularity Check

0 steps flagged · score 2.0 of 10

No circular steps; AFR gains are measured against external benchmarks and no equation reduces to a fitted value.

full rationale

The paper's derivation chain is not circular. The AFR module's operations (Eqs. 8–18) are new combinations of learned and softmax-derived quantities; the learnable fusion weight α in Eq. 17 is optimized during training, and no parameter is fitted to the reported target-validation mIoU. The central claims (Tables I and II) compare AFRDA against external baselines MIC, HRDA, and ERF on public Cityscapes validation, so the improvement is independently evaluable rather than forced by construction. The two self-citations, [4] for mixing techniques and [27] for the POVNav planner, are contextual or demo-related and are not load-bearing for the segmentation gains; hence they do not constitute circularity under the stated rules. A separate correctness concern, not a circularity concern, is that the reported MIC baselines (75.55/67.26) are below the MIC paper's published values (77.2/68.5), which weakens the 'state-of-the-art' absolute claim; this is a benchmarking/reproducibility issue, not an equation-level reduction. Score 2 reflects only the presence of minor, non-load-bearing self-citations.

Assumptions & free parameters 1 free parameters · 3 assumptions · 0 invented entities

The method introduces no new physical or algorithmic entities beyond a standard attention block. The main implicit assumptions are the validity of pseudo-label self-training and the utility of high-frequency cues, both inherited or hypothesized by the authors.

free parameters (1)
  • Gaussian filter parameters (std gamma, kernel size k) = not reported
    Eq. (12) defines the Gaussian filter for high-frequency extraction; gamma and k are not stated or swept, so the boundary-enhancement behavior depends on unstated choices.
assumptions (3)
  • domain assumption Teacher-student self-training with pseudo-labels provides a valid training signal for the target domain
    Section III-A, Eqs. (2)-(4); inherited from DAFormer/HRDA, the adaptation pipeline assumes pseudo-labels are reliable enough to train the student.
  • domain assumption Softmax maximum probability is a suitable proxy for prediction uncertainty
    Section III-B and Eqs. (9), (16); no calibration or entropy-based uncertainty is used, yet the attention maps depend on these values.
  • ad hoc to paper High-frequency residuals, computed as the difference between a feature/logit map and its Gaussian-smoothed version, encode boundary information useful for segmentation
    Eqs. (11) and (14); this is the paper's design hypothesis and is not independently justified beyond the ablation results.

how reviews work

0 comments
Cite this review

Pith. "Pith review of AFRDA: Attentive Feature Refinement for Domain Adaptive Semantic Segmentation." pith.science (2026). https://pith.science/paper/N7PFXCRD

@misc{pith2026250717957,
  author       = {Pith},
  title        = {Pith review of: AFRDA: Attentive Feature Refinement for Domain Adaptive Semantic Segmentation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/N7PFXCRD}},
  note         = {Machine review of arXiv:2507.17957}
}
read the original abstract

In Unsupervised Domain Adaptive Semantic Segmentation (UDA-SS), a model is trained on labeled source domain data (e.g., synthetic images) and adapted to an unlabeled target domain (e.g., real-world images) without access to target annotations. Existing UDA-SS methods often struggle to balance fine-grained local details with global contextual information, leading to segmentation errors in complex regions. To address this, we introduce the Adaptive Feature Refinement (AFR) module, which enhances segmentation accuracy by refining highresolution features using semantic priors from low-resolution logits. AFR also integrates high-frequency components, which capture fine-grained structures and provide crucial boundary information, improving object delineation. Additionally, AFR adaptively balances local and global information through uncertaintydriven attention, reducing misclassifications. Its lightweight design allows seamless integration into HRDA-based UDA methods, leading to state-of-the-art segmentation performance. Our approach improves existing UDA-SS methods by 1.05% mIoU on GTA V --> Cityscapes and 1.04% mIoU on Synthia-->Cityscapes. The implementation of our framework is available at: https://github.com/Masrur02/AFRDA

Figures

Figures reproduced from arXiv: 2507.17957 by the authors.

Figure 1
Figure 1. A brief illustration of our proposed Attentive Feature Refinement for Domain Adaptation (AFRDA) model based on a teacher [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Framework overview of the Attentive Feature Refinement Module. AFR consists of two components: the CALA and UHFA [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Qualitative Results for the adaptation of GTA V [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Qualitative Results for the adaptation of RUGD [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 5
Figure 5. Figure 5: Visualization of the attention maps from the AFR module. [PITH_FULL_IMAGE:figures/full_fig_p008_5.png]
Figure 6
Figure 6. Figure 6: Navigation behavior in the forest environment. From left to right, we show the third-person view, segmentation result, navigable [PITH_FULL_IMAGE:figures/full_fig_p009_6.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

30 extracted references · 28 canonical work pages

  1. [1]

    Learning to adapt structured output space for semantic segmentation,

    Y .-H. Tsai, W.-C. Hung, S. Schulter, K. Sohn, M.-H. Yang, and M. Chandraker, “Learning to adapt structured output space for semantic segmentation,” in Proc. IEEE Conf. Comput. Vis. Pattern Recognit. , 2018, pp. 7472–7481

  2. [2]

    Dlow: Domain flow and applications,

    R. Gong, W. Li, Y . Chen, D. Dai, and L. Van Gool, “Dlow: Domain flow and applications,” Int. J. Comput. Vis., vol. 129, no. 10, pp. 2865–2888, 2021

  3. [3]

    Daformer: Improving network architectures and training strategies for domain-adaptive semantic seg- mentation,

    L. Hoyer, D. Dai, and L. Van Gool, “Daformer: Improving network architectures and training strategies for domain-adaptive semantic seg- mentation,” in Proc. IEEE Conf. Comput. Vis. Pattern Recognit. , 2022, pp. 9924–9935

  4. [4]

    Ida: Informed domain adaptive semantic segmentation,

    Z. Chen, Z. Ding, J. M. Gregory, and L. Liu, “Ida: Informed domain adaptive semantic segmentation,” in IEEE/RSJ Int. Conf. Intell. Robots Syst. (IROS). IEEE, 2023, pp. 90–97

  5. [5]

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

    L. Hoyer, D. Dai, and L. Van Gool, “Hrda: Context-aware high- resolution domain-adaptive semantic segmentation,” in European con- ference on computer vision . Springer, 2022, pp. 372–391

  6. [6]

    Playing for data: Ground truth from computer games,

    S. R. Richter, V . Vineet, S. Roth, and V . Koltun, “Playing for data: Ground truth from computer games,” in Proc. Eur. Conf. Comput. Vis. (ECCV). Amsterdam, The Netherlands: Springer, Oct. 2016, pp. 102– 118

  7. [7]

    The cityscapes dataset for semantic urban scene understanding,

    M. Cordts, M. Omran, S. Ramos, T. Rehfeld, M. Enzweiler, R. Benen- son, U. Franke, S. Roth, and B. Schiele, “The cityscapes dataset for semantic urban scene understanding,” in Proc. IEEE Conf. Comput. Vis. Pattern Recognit., 2016, pp. 3213–3223

  8. [8]

    The synthia dataset: A large collection of synthetic images for semantic segmentation of urban scenes,

    G. Ros, L. Sellart, J. Materzynska, D. Vazquez, and A. M. Lopez, “The synthia dataset: A large collection of synthetic images for semantic segmentation of urban scenes,” in Proc. IEEE Conf. Comput. Vis. Pattern Recognit., 2016, pp. 3234–3243

Show all 30 references
  1. [9]

    A rugd dataset for autonomous navigation and visual perception in unstructured outdoor environments,

    M. Wigness, S. Eum, J. G. Rogers, D. Han, and H. Kwon, “A rugd dataset for autonomous navigation and visual perception in unstructured outdoor environments,” in IEEE/RSJ Int. Conf. Intell. Robots Syst. (IROS), 2019, pp. 5000–5007

  2. [10]

    Advent: Adversarial entropy minimization for domain adaptation in semantic segmentation,

    T.-H. Vu, H. Jain, M. Bucher, M. Cord, and P. P ´erez, “Advent: Adversarial entropy minimization for domain adaptation in semantic segmentation,” in Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit. (CVPR), 2019, pp. 2517–2526

  3. [11]

    Sliced wasserstein discrepancy for unsupervised domain adaptation,

    C.-Y . Lee, T. Batra, M. H. Baig, and D. Ulbricht, “Sliced wasserstein discrepancy for unsupervised domain adaptation,” in Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit. , 2019, pp. 10 285–10 295

  4. [12]

    Mic: Masked image consistency for context-enhanced domain adaptation,

    L. Hoyer, D. Dai, H. Wang, and L. Van Gool, “Mic: Masked image consistency for context-enhanced domain adaptation,” in Proc. IEEE Conf. Comput. Vis. Pattern Recognit. , 2023, pp. 11 721–11 732

  5. [13]

    Pseudolabel guided pixels contrast for domain adaptive semantic segmentation,

    J. Xiang, C. Wan, and Z. Cao, “Pseudolabel guided pixels contrast for domain adaptive semantic segmentation,” Scientific Reports , vol. 14, no. 1, p. 31615, 2024

  6. [14]

    Unsupervised domain adaptation for semantic segmentation with pseudo label self-refinement,

    X. Zhao, N. C. Mithun, A. Rajvanshi, H.-P. Chiu, and S. Samarasek- era, “Unsupervised domain adaptation for semantic segmentation with pseudo label self-refinement,” in Proc. IEEE/CVF Winter Conf. Appl. Comput. Vis. (WACV), 2024, pp. 2399–2409

  7. [15]

    Feature pyramid networks for object detection,

    T.-Y . Lin, P. Doll´ar, R. Girshick, K. He, B. Hariharan, and S. Belongie, “Feature pyramid networks for object detection,” in Proc. IEEE Conf. Comput. Vis. Pattern Recognit. (CVPR) , 2017, pp. 2117–2125

  8. [16]

    Deeplab: Semantic image segmentation with deep convolutional nets, atrous convolution, and fully connected crfs,

    L.-C. Chen, G. Papandreou, I. Kokkinos, K. Murphy, and A. L. Yuille, “Deeplab: Semantic image segmentation with deep convolutional nets, atrous convolution, and fully connected crfs,” IEEE Trans. Pattern Anal. Mach. Intell., vol. 40, no. 4, pp. 834–848, 2017

  9. [17]

    Deep high-resolution representation learning for visual recognition,

    J. Wang, K. Sun, T. Cheng, B. Jiang, C. Deng, Y . Zhao, D. Liu, Y . Mu, M. Tan, X. Wang et al. , “Deep high-resolution representation learning for visual recognition,” IEEE Trans. Pattern Anal. Mach. Intell., vol. 43, no. 10, pp. 3349–3364, 2020

  10. [18]

    Cbam: Convolutional block attention module,

    S. Woo, J. Park, J.-Y . Lee, and I. S. Kweon, “Cbam: Convolutional block attention module,” in Proc. Eur. Conf. Comput. Vis. (ECCV) , 2018, pp. 3–19

  11. [19]

    Squeeze-and-excitation networks,

    J. Hu, L. Shen, and G. Sun, “Squeeze-and-excitation networks,” in Proc. IEEE Conf. Comput. Vis. Pattern Recognit. (CVPR) , 2018, pp. 7132– 7141

  12. [20]

    Gated-scnn: Gated shape cnns for semantic segmentation,

    T. Takikawa, D. Acuna, V . Jampani, and S. Fidler, “Gated-scnn: Gated shape cnns for semantic segmentation,” in Proc. IEEE/CVF Int. Conf. Comput. Vis. (ICCV) , 2019, pp. 5229–5238

  13. [21]

    Rethinking bisenet for real-time semantic segmentation,

    M. Fan, S. Lai, J. Huang, X. Wei, Z. Chai, J. Luo, and X. Wei, “Rethinking bisenet for real-time semantic segmentation,” in Proc. IEEE Conf. Comput. Vis. Pattern Recognit. (CVPR) , 2021, pp. 9716–9725

  14. [22]

    Uncertainty-aware pseudo- label filtering for source-free unsupervised domain adaptation,

    X. Chen, Y . Zheng, Y . Wei, and Y . Shen, “Uncertainty-aware pseudo- label filtering for source-free unsupervised domain adaptation,” Neuro- computing, vol. 575, p. 127190, 2024

  15. [23]

    Ga-nav: Efficient terrain segmentation for robot navigation in unstructured outdoor environments,

    T. Guan, D. Kothandaraman, R. Chandra, A. J. Sathyamoorthy, K. Weer- akoon, and D. Manocha, “Ga-nav: Efficient terrain segmentation for robot navigation in unstructured outdoor environments,” IEEE Robot. Autom. Lett., vol. 7, no. 3, pp. 8138–8145, 2022

  16. [24]

    Tns: Terrain traversability mapping and navigation system for autonomous excava- tors,

    T. Guan, Z. He, R. Song, D. Manocha, and L. Zhang, “Tns: Terrain traversability mapping and navigation system for autonomous excava- tors,” arXiv preprint arXiv:2109.06250 , 2021

  17. [25]

    Salon: Self-supervised adaptive learning for off-road navigation,

    M. Sivaprakasam, S. Triest, C. Ho, S. Aich, J. Lew, I. Adu, W. Wang, and S. Scherer, “Salon: Self-supervised adaptive learning for off-road navigation,” arXiv preprint arXiv:2412.07826 , 2024

  18. [26]

    Classmix: Segmentation-based data augmentation for semi-supervised learning,

    V . Olsson, W. Tranheden, J. Pinto, and L. Svensson, “Classmix: Segmentation-based data augmentation for semi-supervised learning,” in Proc. IEEE/CVF Winter Conf. Appl. Comput. Vis. (WACV), 2021, pp. 1369–1378

  19. [27]

    Povnav: A pareto-optimal mapless visual navigator,

    D. Pushp, Z. Chen, C. Luo, J. M. Gregory, and L. Liu, “Povnav: A pareto-optimal mapless visual navigator,” in Int. Symp. Exp. Robot. (ISER). Springer, 2023, pp. 250–263

  20. [28]

    Dacs: Domain adaptation via cross-domain mixed sampling,

    W. Tranheden, V . Olsson, J. Pinto, and L. Svensson, “Dacs: Domain adaptation via cross-domain mixed sampling,” in Proc. IEEE/CVF Winter Conf. Appl. Comput. Vis. (WACV) , 2021, pp. 1379–1389

  21. [29]

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

    P. Zhang, B. Zhang, T. Zhang, D. Chen, Y . Wang, and F. Wen, “Prototypical pseudo label denoising and target structure learning for domain adaptive semantic segmentation,” in Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit. (CVPR) , 2021, pp. 12 414–12 424

  22. [30]

    Extended receptive field uda semantic segmentation based on spatial alignment and knowledge distillation,

    Y . Song, J. Shi, C. Liu, S. Bai, Y . Yuan, X. Shu, Q. Qian, D. Xu, and Y . Sun, “Extended receptive field uda semantic segmentation based on spatial alignment and knowledge distillation,” IEEE Trans. Autom. Sci. Eng., 2025

Pith tools

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