The reviewed record of science sign in
Pith

arxiv: 2606.01818 · v1 · pith:HQ7BPVO2 · submitted 2026-06-01 · cs.CV

Unsupervised Collaborative Domain Adaptation for Driving Scene Parsing

Jiahe Fan , Shaolong Shu , Mingjian Sun , Tiehua Zhang , Bohong Xiao , Hanli Wang , Rui Fan This is my paper

Reviewed by Pith2026-06-28 15:24 UTCgrok-4.3pith:HQ7BPVO2open to challenge →

classification cs.CV
keywords domain adaptationscene parsingdriving scenessource-free adaptationmulti-source collaborationprototype memoryunsupervised learning
0
0 comments X

The pith

Multiple pre-trained source models can be combined to adapt scene parsing to new driving domains without source data.

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

The paper tries to establish that a collaborative framework can transfer knowledge from several source models to improve parsing in target driving scenes even when source data cannot be shared. This would matter because single-model adaptations suffer from biases and poor generalization to varied conditions like different weather or roads. UCDA uses a prototype memory bank to gauge how reliable each model's predictions are by comparing similarities across models. It then refines the models together on the target data and distills the result into one final model. If this works, perception systems for autonomous vehicles could become more robust while respecting privacy constraints on training data.

Core claim

UCDA constructs a class-level prototype memory bank to estimate cross-model prediction reliability through prototype similarity. Based on this, it performs collaborative optimization of multiple source models on unlabeled target data with positive and negative consistency constraints, then distills their expertise into a single target model.

What carries the argument

A class-level prototype memory bank that compares predictions from different models via similarity to generate consistent supervision signals.

If this is right

  • Improves reliability of scene parsing in the target domain.
  • Enhances generalization to diverse driving environments including varying layouts and conditions.
  • Reduces vulnerability to biases from any single source model.
  • Preserves privacy by not requiring access to original source samples.

Where Pith is reading between the lines

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

  • Similar prototype-based reliability estimation could apply to other multi-model fusion tasks in computer vision.
  • Testing on additional real-world autonomous vehicle datasets with extreme conditions would further validate the approach.
  • The method might integrate with online adaptation scenarios where models update continuously.

Load-bearing premise

That cross-model prediction reliability can be accurately estimated from prototype similarity alone, allowing consistent supervision signals to be generated without access to source data or ground-truth labels.

What would settle it

Experiments showing that UCDA achieves no improvement over the best single source-free method on datasets with high variability in weather and traffic would indicate the collaborative benefit does not hold.

Figures

Figures reproduced from arXiv: 2606.01818 by Bohong Xiao, Hanli Wang, Jiahe Fan, Mingjian Sun, Rui Fan, Shaolong Shu, Tiehua Zhang.

Figure 1
Figure 1. Figure 1: An illustration of (a) source-dependent and (b) source-free unsupervised domain adaptation paradigms, in comparison with the proposed (c) [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Overall architecture of the proposed unsupervised collaborative domain adaptation framework. Source model initialization in Step 1 follows a source [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Detailed training pipeline of UCDA. Complementary supervisory signals are derived through cross-model reliability alignment and decoupled consistency [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Qualitative results for (a) 19-class and (b) 16-class parsing tasks on the Cityscapes dataset. [PITH_FULL_IMAGE:figures/full_fig_p009_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Qualitative results on the BDD100K dataset across 16 semantic categories. [PITH_FULL_IMAGE:figures/full_fig_p011_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Qualitative results on the ACDC dataset across 19 semantic categories. [PITH_FULL_IMAGE:figures/full_fig_p012_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: t-SNE visualization of target-domain feature distributions produced by different driving scene adaptation methods. The plots compare feature embeddings [PITH_FULL_IMAGE:figures/full_fig_p012_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Quantitative sensitivity analysis of the confidence threshold [PITH_FULL_IMAGE:figures/full_fig_p013_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Visual comparison of synthesized pseudo labels generated by various approaches for the 16-class parsing task on the Cityscapes dataset. [PITH_FULL_IMAGE:figures/full_fig_p014_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: The NIO ET9 vehicle utilized as the experimental platform for real [PITH_FULL_IMAGE:figures/full_fig_p014_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Qualitative results on the NIO dataset collected in real-world scenarios. [PITH_FULL_IMAGE:figures/full_fig_p015_11.png] view at source ↗
read the original abstract

Reliable driving scene parsing is a fundamental capability for autonomous vehicles operating in open and dynamic driving environments. However, adapting perception models to new deployment domains remains challenging because pixel-level annotations are expensive to obtain, while source-domain data are often inaccessible due to privacy, security, or ownership constraints. Existing source-free unsupervised domain adaptation methods typically rely on a single pre-trained source model, which makes the adapted perception system vulnerable to source-specific biases and limits its robustness under diverse road layouts, illumination conditions, weather patterns, and traffic conditions. This article presents an unsupervised collaborative domain adaptation (UCDA) framework for driving scene parsing in a source-free setting, which transfers complementary knowledge from multiple pre-trained source models to a unified target model without accessing any original source samples. To compare predictions from independently trained models, UCDA constructs a class-level prototype memory bank and estimates cross-model prediction reliability through prototype similarity, reducing the effect of inconsistent confidence scales across source models. Based on the resulting complementary supervision, UCDA adopts a two-stage transfer strategy: multiple source models are first refined on unlabeled target-domain driving data through collaborative optimization with positive and negative consistency constraints, and their validated expertise is then distilled into a single deployable target model. Comprehensive evaluations on public driving-scene datasets and real-world data collected from an autonomous vehicle platform demonstrate that UCDA effectively consolidates complementary multi-source knowledge, improving target-domain scene parsing reliability and generalization across diverse driving environments.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 1 minor

Summary. The paper proposes UCDA, a source-free unsupervised domain adaptation framework for driving scene parsing. It transfers knowledge from multiple pre-trained source models to a target model by building a class-level prototype memory bank to estimate cross-model prediction reliability via prototype similarity (to handle inconsistent confidence scales), then applies a two-stage process: collaborative optimization of the source models on unlabeled target data using positive/negative consistency constraints, followed by distillation of validated expertise into a single deployable target model. Evaluations on public driving datasets and real-world autonomous vehicle data are reported to show improved target-domain parsing reliability and generalization across diverse conditions.

Significance. If the prototype-similarity reliability estimates prove accurate, the framework would offer a practical advance in source-free multi-source domain adaptation for autonomous driving by reducing single-model bias and enabling privacy-preserving adaptation without source samples. The two-stage collaborative optimization plus distillation is a structured way to consolidate complementary knowledge, with potential impact on robustness under varying illumination, weather, and traffic.

major comments (2)
  1. [Abstract] The load-bearing step is the assumption that prototype similarity alone yields accurate reliability estimates for generating supervision signals (abstract, paragraph 3). No correlation analysis, ablation on similarity vs. ground-truth accuracy, or comparison to high-confidence pseudo-label baselines is described, leaving open whether high-similarity predictions reflect true accuracy or shared biases/domain-shift artifacts under driving-scene variations.
  2. [Abstract] The two-stage strategy claims to avoid confirmation bias via collaborative optimization and consistency constraints before distillation, yet the abstract provides no equations, pseudocode, or quantitative evidence that the positive/negative constraints prevent amplification of inconsistent predictions across models.
minor comments (1)
  1. [Abstract] The abstract states 'comprehensive evaluations' but does not name the public datasets, metrics (mIoU, etc.), or baselines, which should be clarified for reproducibility.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments on our manuscript. We address each major comment below with clarifications from the full paper and indicate planned revisions to strengthen the presentation.

read point-by-point responses

  1. Referee: [Abstract] The load-bearing step is the assumption that prototype similarity alone yields accurate reliability estimates for generating supervision signals (abstract, paragraph 3). No correlation analysis, ablation on similarity vs. ground-truth accuracy, or comparison to high-confidence pseudo-label baselines is described, leaving open whether high-similarity predictions reflect true accuracy or shared biases/domain-shift artifacts under driving-scene variations.

    Authors: The abstract is a high-level summary and does not contain the requested analyses. However, Section 3.2 details the prototype memory bank and similarity-based reliability estimation, while Section 5.2 and Table 3 present ablations correlating prototype similarity with ground-truth accuracy on held-out target validation data and direct comparisons against high-confidence pseudo-label baselines. These results show that similarity-based selection outperforms confidence thresholding and reduces bias from domain-shift artifacts. We will revise the abstract to briefly note that supporting ablations and comparisons appear in the experiments. revision: yes

  2. Referee: [Abstract] The two-stage strategy claims to avoid confirmation bias via collaborative optimization and consistency constraints before distillation, yet the abstract provides no equations, pseudocode, or quantitative evidence that the positive/negative constraints prevent amplification of inconsistent predictions across models.

    Authors: The abstract summarizes the overall strategy without equations. Section 3.3 provides the full formulation of the positive and negative consistency constraints (Equations 4-7) together with the collaborative optimization objective. Section 5.3 reports quantitative ablations measuring prediction consistency across models before and after the constraints, demonstrating reduced amplification of inconsistent predictions. We will add a concise clause to the abstract referencing these constraints and their empirical validation in the experiments. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical framework with no derivations or self-referential predictions

full rationale

The paper describes an empirical UCDA framework relying on prototype similarity for cross-model reliability estimation and a two-stage collaborative optimization plus distillation process. No equations, first-principles derivations, or 'predictions' are presented that reduce to fitted inputs or self-citations by construction. The central claims rest on experimental results across public driving datasets and real-world data, which constitute external validation rather than internal redefinition. No load-bearing self-citation chains or ansatzes smuggled via prior work are evident in the provided text.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract provides no explicit free parameters, axioms, or invented entities; all technical details are absent.

pith-pipeline@v0.9.1-grok · 5798 in / 1100 out tokens · 20472 ms · 2026-06-28T15:24:37.514831+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

74 extracted references · 1 canonical work pages · 1 internal anchor

  1. [1]

    Physics-informed neural mapping and motion planning in unknown environments,

    Y . Liuet al., “Physics-informed neural mapping and motion planning in unknown environments,”IEEE Transactions on Robotics, vol. 41, pp. 2200–2212, 2025

    2025

  2. [2]

    OccNeRF: Advancing 3D occupancy prediction in LiDAR-free environments,

    C. Zhanget al., “OccNeRF: Advancing 3D occupancy prediction in LiDAR-free environments,”IEEE Transactions on Image Processing, vol. 34, pp. 3096–3107, 2025

    2025

  3. [3]

    Decouple ego-view motions for predicting pedestrian trajectory and intention,

    Z. Zhanget al., “Decouple ego-view motions for predicting pedestrian trajectory and intention,”IEEE Transactions on Image Processing, vol. 33, pp. 4716–4727, 2024

    2024

  4. [4]

    DALI: Domain adaptive lidar object detection via distribution-level and instance-level pseudolabel denoising,

    X. Lu and H. Radha, “DALI: Domain adaptive lidar object detection via distribution-level and instance-level pseudolabel denoising,”IEEE Transactions on Robotics, vol. 40, pp. 3866–3878, 2024

    2024

  5. [5]

    Toward generative understanding: Incremental few-shot semantic segmentation with diffusion models,

    Q. Liet al., “Toward generative understanding: Incremental few-shot semantic segmentation with diffusion models,”IEEE Transactions on Image Processing, vol. 35, pp. 743–758, 2026

    2026

  6. [6]

    Source-free domain adaptation for semantic segmenta- tion,

    Y . Liuet al., “Source-free domain adaptation for semantic segmenta- tion,” inProceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2021, pp. 1215–1224

    2021

  7. [7]

    C-SFDA: A curriculum learning aided self-training framework for efficient source free domain adaptation,

    N. Karimet al., “C-SFDA: A curriculum learning aided self-training framework for efficient source free domain adaptation,” inProceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2023, pp. 24 120–24 131

    2023

  8. [8]

    Category contrast for unsupervised domain adaptation in visual tasks,

    J. Huanget al., “Category contrast for unsupervised domain adaptation in visual tasks,” inProceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2022, pp. 1193–1204

    2022

  9. [9]

    Adaptive refining-aggregation-separation framework for unsupervised domain adaptation semantic segmentation,

    Y . Caoet al., “Adaptive refining-aggregation-separation framework for unsupervised domain adaptation semantic segmentation,”IEEE Trans- actions on Circuits and Systems for Video Technology, vol. 33, no. 8, pp. 3822–3832, 2023

    2023

  10. [10]

    Prototypical bidirectional adaptation and learning for cross-domain semantic segmentation,

    Q. Renet al., “Prototypical bidirectional adaptation and learning for cross-domain semantic segmentation,”IEEE Transactions on Multime- dia, vol. 26, pp. 501–513, 2024

    2024

  11. [11]

    A curriculum-style self-training approach for source- free semantic segmentation,

    Y . Wanget al., “A curriculum-style self-training approach for source- free semantic segmentation,”IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 46, no. 12, pp. 9890–9907, 2024

    2024

  12. [12]

    Self-mining the confident prototypes for source-free unsupervised domain adaptation in image segmentation,

    Y . Tianet al., “Self-mining the confident prototypes for source-free unsupervised domain adaptation in image segmentation,”IEEE Trans- actions on Multimedia, vol. 26, pp. 7709–7720, 2024

    2024

  13. [13]

    Towards source-free domain adaptive semantic seg- mentation via importance-aware and prototype-contrast learning,

    Y . Caoet al., “Towards source-free domain adaptive semantic seg- mentation via importance-aware and prototype-contrast learning,”IEEE Transactions on Intelligent Vehicles, vol. 10, no. 4, pp. 2736–2747, 2025

    2025

  14. [14]

    Stable neighbor denoising for source-free domain adaptive segmentation,

    D. Zhaoet al., “Stable neighbor denoising for source-free domain adaptive segmentation,” inProceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2024, pp. 23 416– 23 427

    2024

  15. [15]

    HG-SFDA: HyperGraph learning meets source-free un- supervised domain adaptation,

    J. Jianget al., “HG-SFDA: HyperGraph learning meets source-free un- supervised domain adaptation,”IEEE Transactions on Image Processing, vol. 34, pp. 7542–7557, 2025

    2025

  16. [16]

    Unlocking cross-domain synergies for domain adaptive se- mantic segmentation,

    Q. Xuet al., “Unlocking cross-domain synergies for domain adaptive se- mantic segmentation,”IEEE Transactions on Image Processing, vol. 35, pp. 136–149, 2026

    2026

  17. [17]

    Generalize then adapt: Source-free domain adaptive semantic segmentation,

    J. N. Kunduet al., “Generalize then adapt: Source-free domain adaptive semantic segmentation,” inProceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), 2021, pp. 7026–7036

    2021

  18. [18]

    Personalized sleep staging leveraging source-free unsupervised domain adaptation,

    Y . Zhouet al., “Personalized sleep staging leveraging source-free unsupervised domain adaptation,”Proceedings of the AAAI Conference on Artificial Intelligence (AAAI), pp. 14 529–14 537, 2025

    2025

  19. [19]

    CoUDA: Continual unsupervised domain adaptation for industrial fault diagnosis under dynamic working conditions,

    B. Chenet al., “CoUDA: Continual unsupervised domain adaptation for industrial fault diagnosis under dynamic working conditions,”IEEE Transactions on Industrial Informatics, vol. 21, no. 5, pp. 4072–4082, 2025

    2025

  20. [20]

    Multi-granularity class prototype topology distillation for class-incremental source-free unsupervised domain adaptation,

    P. Denget al., “Multi-granularity class prototype topology distillation for class-incremental source-free unsupervised domain adaptation,” in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2025, pp. 30 566–30 576

    2025

  21. [21]

    Robust source-free domain adaptation from non-robust source models,

    Y . Xiaoet al., “Robust source-free domain adaptation from non-robust source models,”IEEE Transactions on Image Processing, vol. 35, pp. 2350–2363, 2026

    2026

  22. [22]

    CLIP-powered domain generalization and domain adapta- tion: A comprehensive survey,

    J. Liet al., “CLIP-powered domain generalization and domain adapta- tion: A comprehensive survey,”IEEE Transactions on Pattern Analysis and Machine Intelligence, pp. 1–20, 2026

    2026

  23. [23]

    Advances in multimodal adaptation and generalization: From traditional approaches to foundation models,

    H. Donget al., “Advances in multimodal adaptation and generalization: From traditional approaches to foundation models,”IEEE Transactions on Pattern Analysis and Machine Intelligence, pp. 1–20, 2026

    2026

  24. [24]

    Leveraging mixture alignment for multi-source domain adaptation,

    A. Dayalet al., “Leveraging mixture alignment for multi-source domain adaptation,”IEEE Transactions on Image Processing, vol. 34, pp. 885– 898, 2025

    2025

  25. [25]

    SimT: Handling open-set noise for domain adaptive semantic segmentation,

    X. Guoet al., “SimT: Handling open-set noise for domain adaptive semantic segmentation,” inProceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2022, pp. 7022– 7031

    2022

  26. [26]

    Adaptive adversarial network for source-free domain adaptation,

    H. Xiaet al., “Adaptive adversarial network for source-free domain adaptation,” inProceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), 2021, pp. 8990–8999

    2021

  27. [27]

    Cal-SFDA: Source-free domain-adaptive semantic seg- mentation with differentiable expected calibration error,

    Z. Wanget al., “Cal-SFDA: Source-free domain-adaptive semantic seg- mentation with differentiable expected calibration error,” inProceedings of the ACM International Conference on Multimedia (ACM MM), 2023, p. 1167–1178

    2023

  28. [28]

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

    T.-H. Vuet al., “ADVENT: Adversarial entropy minimization for domain adaptation in semantic segmentation,” inProceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2019, pp. 2512–2521

    2019

  29. [29]

    Multiple adaptation network for multi-source and multi- target domain adaptation,

    Y . Luet al., “Multiple adaptation network for multi-source and multi- target domain adaptation,”IEEE Transactions on Multimedia, vol. 27, pp. 5731–5745, 2025

    2025

  30. [30]

    Multi-source unsupervised domain adaptation via pseudo target domain,

    C.-X. Renet al., “Multi-source unsupervised domain adaptation via pseudo target domain,”IEEE Transactions on Image Processing, vol. 31, pp. 2122–2135, 2022

    2022

  31. [31]

    Multi-source multi-modal domain adaptation,

    S. Zhaoet al., “Multi-source multi-modal domain adaptation,”Informa- tion Fusion, vol. 117, p. 102862, 2025

    2025

  32. [32]

    Multilevel distribution alignment for multisource uni- versal domain adaptation,

    L. Ninget al., “Multilevel distribution alignment for multisource uni- versal domain adaptation,”IEEE Transactions on Neural Networks and Learning Systems, vol. 36, no. 9, pp. 17 365–17 379, 2025

    2025

  33. [33]

    VDM-DA: Virtual domain modeling for source data- free domain adaptation,

    J. Tianet al., “VDM-DA: Virtual domain modeling for source data- free domain adaptation,”IEEE Transactions on Circuits and Systems for Video Technology, vol. 32, no. 6, pp. 3749–3760, 2022

    2022

  34. [34]

    Uncertainty reduction for model adaptation in semantic segmentation,

    P. T. S and F. Fleuret, “Uncertainty reduction for model adaptation in semantic segmentation,” inProceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2021, pp. 9608– 9618

    2021

  35. [35]

    Generalized source-free domain-adaptive segmentation via reliable knowledge propagation,

    Q. Zanget al., “Generalized source-free domain-adaptive segmentation via reliable knowledge propagation,” inProceedings of the ACM Inter- national Conference on Multimedia (ACM MM), 2024, p. 5967–5976

    2024

  36. [36]

    Cycle self-refinement for multi-source domain adapta- tion,

    C. Zhouet al., “Cycle self-refinement for multi-source domain adapta- tion,” inProceedings of the AAAI Conference on Artificial Intelligence (AAAI), 2024, pp. 17 096–17 104

    2024

  37. [37]

    Multi-source domain adaptation for semantic segmen- tation,

    S. Zhaoet al., “Multi-source domain adaptation for semantic segmen- tation,”Advances in Neural Information Processing Systems (NeurIPS), vol. 32, 2019

    2019

  38. [38]

    Deep cocktail network: Multi-source unsupervised domain adaptation with category shift,

    R. Xuet al., “Deep cocktail network: Multi-source unsupervised domain adaptation with category shift,” inProceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2018, pp. 3964– 3973. 17

    2018

  39. [39]

    Moment matching for multi-source domain adaptation,

    X. Penget al., “Moment matching for multi-source domain adaptation,” inProceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), 2019, pp. 1406–1415

    2019

  40. [40]

    Unsupervised multi-source domain adaptation without access to source data,

    S. M. Ahmedet al., “Unsupervised multi-source domain adaptation without access to source data,” inProceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition (CVPR), 2021, pp. 10 103–10 112

    2021

  41. [41]

    Confident anchor-induced multi-source free domain adaptation,

    J. Donget al., “Confident anchor-induced multi-source free domain adaptation,” inAdvances in Neural Information Processing Systems (NeurIPS), vol. 34, 2021, pp. 2848–2860

    2021

  42. [42]

    Discriminability and transferability estimation: A Bayesian source importance estimation approach for multi-source-free domain adaptation,

    Z. Hanet al., “Discriminability and transferability estimation: A Bayesian source importance estimation approach for multi-source-free domain adaptation,” inProceedings of the AAAI Conference on Artificial Intelligence (AAAI), 2023, pp. 7811–7820

    2023

  43. [43]

    On balancing bias and variance in unsupervised multi- source-free domain adaptation,

    M. Shenet al., “On balancing bias and variance in unsupervised multi- source-free domain adaptation,” inProceedings of the International Conference on Machine Learning (ICML), 2023, pp. 30 976–30 991

    2023

  44. [44]

    Multi-source domain adaptation with collaborative learning for semantic segmentation,

    J. Heet al., “Multi-source domain adaptation with collaborative learning for semantic segmentation,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2021, pp. 11 008– 11 017

    2021

  45. [45]

    RankMatch: Exploring the better consistency regulariza- tion for semi-supervised semantic segmentation,

    H. Maiet al., “RankMatch: Exploring the better consistency regulariza- tion for semi-supervised semantic segmentation,” inProceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2024, pp. 3391–3401

    2024

  46. [46]

    Semi-supervised semantic segmentation with high- and low-level consistency,

    S. Mittalet al., “Semi-supervised semantic segmentation with high- and low-level consistency,”IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 43, no. 4, pp. 1369–1379, 2021

    2021

  47. [47]

    Semi supervised semantic segmentation using genera- tive adversarial network,

    N. Soulyet al., “Semi supervised semantic segmentation using genera- tive adversarial network,” inProceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), 2017, pp. 5689–5697

    2017

  48. [48]

    AllSpark: Reborn labeled features from unlabeled in transformer for semi-supervised semantic segmentation,

    H. Wanget al., “AllSpark: Reborn labeled features from unlabeled in transformer for semi-supervised semantic segmentation,” inProceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2024, pp. 3627–3636

    2024

  49. [49]

    Saliency as pseudo-pixel supervision for weakly and semi-supervised semantic segmentation,

    M. Leeet al., “Saliency as pseudo-pixel supervision for weakly and semi-supervised semantic segmentation,”IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 45, no. 10, pp. 12 341–12 357, 2023

    2023

  50. [50]

    CorrMatch: Label propagation via correlation matching for semi-supervised semantic segmentation,

    B. Sunet al., “CorrMatch: Label propagation via correlation matching for semi-supervised semantic segmentation,” inProceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2024, pp. 3097–3107

    2024

  51. [51]

    Logic-induced diagnostic reasoning for semi-supervised semantic segmentation,

    C. Lianget al., “Logic-induced diagnostic reasoning for semi-supervised semantic segmentation,” inProceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), 2023, pp. 16 197–16 208

    2023

  52. [52]

    CustomKD: Customizing large vision foundation for edge model improvement via knowledge distillation,

    J. Leeet al., “CustomKD: Customizing large vision foundation for edge model improvement via knowledge distillation,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2025, pp. 25 176–25 186

    2025

  53. [53]

    Feature distillation-based uniformity few-shot domain adaptation for cross-domain fault diagnosis with sample shortage,

    Y . Anet al., “Feature distillation-based uniformity few-shot domain adaptation for cross-domain fault diagnosis with sample shortage,”IEEE Transactions on Industrial Informatics, vol. 21, no. 5, pp. 3717–3726, 2025

    2025

  54. [54]

    Hyperbolic insights with knowledge distillation for cross-domain few-shot learning,

    X. Yanget al., “Hyperbolic insights with knowledge distillation for cross-domain few-shot learning,”IEEE Transactions on Image Process- ing, vol. 34, pp. 1921–1933, 2025

    1921

  55. [55]

    Multi-teacher knowledge distillation with reinforcement learning for visual recognition,

    C. Yanget al., “Multi-teacher knowledge distillation with reinforcement learning for visual recognition,” inProceedings of the AAAI Conference on Artificial Intelligence (AAAI), 2025, pp. 9148–9156

    2025

  56. [56]

    Class incremental learning with multi-teacher distilla- tion,

    H. Wenet al., “Class incremental learning with multi-teacher distilla- tion,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2024, pp. 28 443–28 452

    2024

  57. [57]

    Playing for data: Ground truth from computer games,

    S. R. Richteret al., “Playing for data: Ground truth from computer games,” inProceedings of the European Conference on Computer Vision (ECCV), 2016, pp. 102–118

    2016

  58. [58]

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

    G. Roset al., “The SYNTHIA dataset: A large collection of synthetic images for semantic segmentation of urban scenes,” inProceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2016, pp. 3234–3243

    2016

  59. [59]

    Synscapes: A Photorealistic Synthetic Dataset for Street Scene Parsing

    M. Wrenninge and J. Unger, “Synscapes: A photorealistic synthetic dataset for street scene parsing,”Computing Research Repository (CoRR), vol. abs/1810.08705, 2023

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  60. [60]

    The mapillary vistas dataset for semantic under- standing of street scenes,

    G. Neuholdet al., “The mapillary vistas dataset for semantic under- standing of street scenes,” inProceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), 2017, pp. 5000–5009

    2017

  61. [61]

    The Cityscapes dataset for semantic urban scene understanding,

    M. Cordtset al., “The Cityscapes dataset for semantic urban scene understanding,” inProceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2016, pp. 3213–3223

    2016

  62. [62]

    BDD100K: A diverse driving dataset for heterogeneous multitask learning,

    F. Yuet al., “BDD100K: A diverse driving dataset for heterogeneous multitask learning,” inProceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2020, pp. 2633–2642

    2020

  63. [63]

    ACDC: The adverse conditions dataset with corre- spondences for semantic driving scene understanding,

    C. Sakaridiset al., “ACDC: The adverse conditions dataset with corre- spondences for semantic driving scene understanding,” inProceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), 2021, pp. 10 745–10 755

    2021

  64. [64]

    DeepLab: Semantic image segmentation with deep convolutional nets, atrous convolution, and fully connected CRFs,

    L.-C. Chenet al., “DeepLab: Semantic image segmentation with deep convolutional nets, atrous convolution, and fully connected CRFs,”IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 40, no. 4, pp. 834–848, 2018

    2018

  65. [65]

    Deep residual learning for image recognition,

    K. Heet al., “Deep residual learning for image recognition,” inPro- ceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2016, pp. 770–778

    2016

  66. [66]

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

    L. Hoyeret al., “DAFormer: Improving network architectures and training strategies for domain-adaptive semantic segmentation,” inPro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2022, pp. 9924–9935

    2022

  67. [67]

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

    ——, “HRDA: Context-aware high-resolution domain-adaptive semantic segmentation,” inProceedings of the European Conference on Computer Vision (ECCV), 2022, pp. 372–391

    2022

  68. [68]

    Informative data mining for one-shot cross-domain semantic segmentation,

    Y . Wanget al., “Informative data mining for one-shot cross-domain semantic segmentation,” inProceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), 2023, pp. 1064–1074

    2023

  69. [69]

    Model adaptation: Historical contrastive learning for unsupervised domain adaptation without source data,

    J. Huanget al., “Model adaptation: Historical contrastive learning for unsupervised domain adaptation without source data,” inAdvances in Neural Information Processing Systems (NeurIPS), 2021, pp. 3635– 3649

    2021

  70. [70]

    Towards better stability and adaptability: Improve online self-training for model adaptation in semantic segmentation,

    D. Zhaoet al., “Towards better stability and adaptability: Improve online self-training for model adaptation in semantic segmentation,” in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2023, pp. 11 733–11 743

    2023

  71. [71]

    DACS: Domain adaptation via cross-domain mixed sampling,

    W. Tranhedenet al., “DACS: Domain adaptation via cross-domain mixed sampling,” inIEEE Winter Conference on Applications of Computer Vision (WACV), 2021, pp. 1378–1388

    2021

  72. [72]

    VBLC: Visibility boosting and logit-constraint learning for domain adaptive semantic segmentation under adverse conditions,

    M. Liet al., “VBLC: Visibility boosting and logit-constraint learning for domain adaptive semantic segmentation under adverse conditions,” Proceedings of the AAAI Conference on Artificial Intelligence (AAAI), pp. 8605–8613, 2023

    2023

  73. [73]

    Tent: Fully test-time adaptation by entropy mini- mization,

    D. Wanget al., “Tent: Fully test-time adaptation by entropy mini- mization,” inProceedings of the International Conference on Learning Representations (ICLR), 2021

    2021

  74. [74]

    Continual test-time domain adaptation,

    Q. Wanget al., “Continual test-time domain adaptation,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recogni- tion (CVPR), 2022, pp. 7201–7211

    2022