Pith. sign in

REVIEW 4 major objections 5 minor 50 references

Leveraging Contrastive Learning for Semantic Segmentation with Consistent Labels Across Varying Appearances

T0 review · 4 major / 5 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read The paper claims that image pairs showing the same scene under different weather, with pixel-identical labels, make feature-level alignment work for semantic segmentation and improve both domain adaptation and generalization.

desk verdict Useful aligned multi-weather dataset; the claimed gain from feature alignment is real-but-unproven due to a throughput confound with the Random baseline. read the letter →

arxiv 2412.16592 v1 pith:ZIIK6JBK submitted 2024-12-21 cs.CV

Javier Montalvo , Roberto Alcover-Couso , Pablo Carballeira , Álvaro García-Martín , Juan C. SanMiguel , Marcos Escudero-Viñolo This is my paper · ORCID
classification cs.CV
keywords semanticsegmentationdomainadaptationgeneralizationfeaturealignmentsyntheticdataadverseweatherCARLAsimulatorpixel-alignedgroundtruth
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 claims that the main obstacle to feature-level domain adaptation in semantic segmentation is the absence of images that show the same scene layout under different appearances. It removes that obstacle by building CARLA-4AGT, a synthetic urban dataset of roughly 20,000 images: about 5,000 scene layouts, each rendered at noon, sunset, nighttime, and fog, with ground-truth labels that are pixel-identical across the four versions. On top of this dataset, the authors add a feature alignment loss that pulls together, at several network depths, the features produced by two different weather versions of the same layout. Across UDA frameworks (DAFormer, HRDA, MIC, ADVENT) this raises accuracy, for example from 57.2 to 62.8 mIoU for DAFormer on Cityscapes, and improves domain generalization on Cityscapes, ACDC, and Dark Zurich. The paper also answers a data-generation question: fewer layouts with high appearance variability outperform larger, less varied synthetic datasets.

What carries the argument

The load-bearing object is CARLA-4AGT, a synthetic dataset generated by rerunning the CARLA simulator with the same random seed and changing only weather, sun intensity, and sun position, so that each of roughly 5,000 layouts yields four RGB images (Noon, Sunset, Nighttime, Foggy) sharing one pixel-perfect ground truth. The alignment loss then sums, over the feature layers, a distance between features extracted from two different appearances of the same layout, with the distance chosen as L2, MMD, or cosine similarity; for UDA, a MixUp over source instances and target images creates matched pairs on target data as well. The loss is weighted by the inverse of the number of layers, so it scales to any architecture, and cosine similarity computed at all four residual blocks proves best.

What would settle it

Render the same seeded CARLA layout in the four weather setups, run a dense correspondence or optical-flow algorithm between the Noon and Foggy (or Nighttime) versions, and measure the fraction of pixels whose displacement exceeds a small threshold; if substantial misalignment appears at object boundaries or under precipitation, the feature alignment loss is comparing features that do not correspond to the same scene point.

Watch

Extended reading notes

Core claim

The authors' central discovery is that feature-space alignment, which is common in image classification but widely considered harmful for segmentation because pixel features carry spatial context, becomes both feasible and beneficial once the training data contains multiple appearances of the very same scene with the very same labels. They verify this by generating the four-appearance CARLA-4AGT dataset and showing that a lightweight alignment term added to existing UDA objectives, summing L2, maximum mean discrepancy, or cosine similarity between features of two appearances across the network's residual blocks, improves every framework tested. The strongest version uses cosine similarity: DAFormer goes from 57.2 to 62.8 mIoU on Cityscapes, with comparable gains for HRDA, MIC, and ADVENT, and domain-generalization scores rise to 51.0 on Cityscapes, 39.8 on ACDC, and 24.9 on Dark Zurich, surpassing GTA and Synthia under identical training budgets. They further find that aligning all four blocks beats aligning any single block, and that with alignment a model trained on 1,000 layouts (57.5 mIoU) beats a model trained on 4,000 layouts without alignment (56.4 mIoU).

Load-bearing premise

Everything rests on the claim that rerunning CARLA with a fixed seed reproduces the same scene geometry, object poses, and camera settings under every weather setting, so that ground-truth labels are pixel-perfect aligned and the same pixel coordinate always corresponds to the same object across the four appearances; the paper states this is achieved but reports no direct measurement of alignment error.

Editorial extensions

If this is right

  • Adding the alignment loss to any UDA framework that already has a target-domain loss (DAFormer, HRDA, MIC, ADVENT) improves mean IoU by about 4.3% on average, so feature alignment can be used as a drop-in module rather than a competing method.
  • Domain generalization measured on Cityscapes, ACDC, and Dark Zurich improves (51.0 / 39.8 / 24.9 mIoU with cosine similarity), surpassing both GTA and Synthia trained under identical budgets, even though GTA has far more assets per class.
  • Fewer, more varied layouts beat larger datasets: 1,000 layouts with alignment outperform 4,000 layouts without it, and the performance curve keeps rising where the non-aligned baseline flattens.
  • Training on all four appearances together beats any single appearance, including appearances that visually match the target domain, supporting variability over target-domain matching.
  • Aligning all four residual blocks yields the best results, with up to +6 mIoU over single-layer alignment and +8 over no alignment.

Reading between the lines

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

  • A natural extension is to use the same recipe with real data: if pixel correspondence can be obtained from depth, optical flow, or multi-view cameras, feature alignment could be applied to real captured scene pairs instead of only synthetic ones, where the determinism assumption is easier to check.
  • If the trend from Q3 continues, adding even more extreme weather appearances (snow, heavy rain streaks, glare) would improve generalization rather than saturate; this is testable by expanding CARLA-4AGT with additional appearances.
  • Although the title says contrastive, the alignment loss has no negative pairs; adding negatives from different layouts could sharpen features further, or could hurt by over-constraining spatial context, a direct experiment suggested by the paper's setup.
  • Because the gain grows with the number of aligned layers, the method argues that appearance invariance should be enforced throughout the encoder, not just at the output; this could be probed by ablating which layer matters most for each degradation type.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The paper introduces CARLA-4AGT, a synthetic urban dataset in which each scene layout is rendered under four weather/illumination appearances (Sunset, Noon, Nighttime, Foggy) with pixel-aligned ground truth, and proposes a feature-alignment training loss that enforces similarity between features extracted from different appearances of the same layout at multiple network layers. The method is evaluated for unsupervised domain adaptation (UDA) on Cityscapes and for domain generalization (DG) on Cityscapes, ACDC, and Dark Zurich, using DAFormer as the primary backbone and HRDA, MIC, and ADVENT as additional UDA methods. The authors report consistent improvements over single-appearance and random-appearance training, over Synthia and GTA as source datasets, and over data augmentation and diffusion-based style transfer, with the central quantitative claim being that cosine-similarity alignment raises DAFormer from 57.2 to 62.8 mIoU on Cityscapes.

Significance. If the reported gains are correctly attributed to feature alignment, the paper makes a useful contribution: it provides a concrete way to generate pixel-aligned multi-appearance synthetic data, it shows that feature-level alignment can be effective for dense prediction when spatial correspondence is guaranteed, and it demonstrates gains across four UDA methods and three DG targets. The method is simple and has no fitted constants beyond a fixed lambda=1/L, which aids reproducibility. The dataset itself, even without the proposed loss, appears to be a valuable resource. However, the central attribution of the gains to the alignment mechanism is currently confounded by unequal data throughput between the baseline and the proposed training scheme, and several main tables lack error bars, so the strength of the evidence is weaker than the text suggests.

major comments (4)
  1. [Section 4.1.2, Eq. (2), Eq. (6)] The reported gains may be partly or wholly due to increased data throughput rather than to the alignment loss itself. In each training iteration, the Random baseline forwards one appearance per layout, while the proposed method forwards two appearances per layout for the source alignment loss (Eq. 2) and, in UDA, two mixed images per layout for the target alignment loss (Eq. 6). Since all models are trained for the same 40,000 iterations at batch size 2, the feature-aligned models process roughly twice as many source (and mixed) images per iteration. The paper never compares against a control that sees the same number of images per iteration without the alignment loss, such as summing cross-entropy over two randomly chosen appearances. This confound directly undermines the attribution of the improvements to feature alignment, which is the paper's main methodological novelty. A 'double-random' ablation or an equivalent compute-matched comparison is needed before the central claim can be accepted.
  2. [Tables 2-5, 7-8 and Figure 6] The main quantitative tables report single runs without error bars or significance information, while only Figures 6 and 7 use three seeds. Several key comparisons are close enough that run-to-run variance could change the conclusions; for example, Table 6 shows the same ACDC mIoU (39.8) for GTA and the proposed method, and Table 3 shows an 8.9 mIoU gap between CS and MMD that is reported without variance. The authors should provide per-table error bars or standard deviations over at least three seeds for the central comparisons, or explicitly state which results are single runs and temper the corresponding claims.
  3. [Section 3.1, deterministic CARLA replay claim] The load-bearing premise of the method is that CARLA's deterministic replay produces pixel-perfect aligned ground truths across weather changes, so that features at the same spatial location truly correspond to the same scene element. The paper asserts this in Section 3.1 ('we configure CARLA to ensure fully deterministic behavior using a random seed') but provides no quantitative check of alignment error, for example under fog or nighttime rendering where object edges may shift, transparency effects may appear, or weather particles may occlude pixels. A small empirical verification, such as measuring per-pixel correspondence of rendered depth or instance masks across appearances, would substantially strengthen the claim that the aligned labels are exact rather than approximate.
  4. [Figure 6 and Section 4.4, Q1] The data-efficiency claim that 'with just 1K images, the introduction of controlled and known variability yields models that outperform those trained on five times more data' is confounded by the same throughput issue. In Figure 6, the 1,000-layout feature-aligned model uses two appearances per layout, so it sees 2,000 images over the same number of iterations, while the 4,000-layout baseline sees one appearance per layout, or 4,000 distinct images. The comparison mixes dataset size with per-iteration image count; an equal-image or equal-compute comparison is needed to support the efficiency conclusion.
minor comments (5)
  1. [Section 4.1.2, Table 2 caption] The caption describes Fixed as seeing 'two fixed appearances' per layout, while the text in Section 4.1.2 defines Fixed as a single predefined appearance per layout; please reconcile this discrepancy.
  2. [Section 3.2, Eq. (2)] Equation (2) sums from l=0 to L, but the text defines features for layer l in [1, L]; please clarify the indexing and the corresponding definition of lambda=1/L.
  3. [Throughout] There are numerous typographical errors and inconsistencies, including 'TTo', 'Incusion', 'demostrate', 'realiable', 'appereances', 'purpuses', and the double plus sign in Eq. (7) ('= LDG + +LT'). A careful proofreading pass is needed.
  4. [Section 4.3] The text claims 'significant performance improvements on ACDC' for the comparison in Table 6, but the table shows a tie between GTA and the proposed method on ACDC (39.8 mIoU); please soften or correct this statement.
  5. [Tables and text] The comma used as a decimal separator in '4,3%' (Section 4.3) should be replaced with a period for consistency with the rest of the manuscript.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the method's gains are evaluated on external benchmarks and the alignment loss is not fitted to those benchmarks.

full rationale

The central derivation chain is: (i) generate multi-appearance source images with pixel-aligned ground truths by restarting CARLA with the same seed and varying weather/sun parameters; (ii) define a feature alignment loss LA = sum_l a(f^i,j_l, f^i,j'_l) that enforces similarity between features of two appearances of the same layout; (iii) combine it with the cross-entropy loss as LDG = LS + lambda*LA, with lambda = 1/L; and (iv) evaluate on external real benchmarks (Cityscapes, ACDC, Dark Zurich) and standard UDA/DG baselines. None of these equations defines the target metric in terms of the loss, nor is any target result used to set a parameter: lambda is fixed as the inverse of the number of aligned layers, not fitted to validation performance. The reported improvements (e.g., DAFormer 57.2 to 62.8 mIoU, Table 7) are empirical comparisons on held-out real data, not consequences of the definitions. The claim that ground truths are pixel-perfect aligned is a dataset-construction property from deterministic replay with a fixed seed, not a prediction that reduces to an input. Self-citations appear in related work and in the authors' prior segmentation/DA papers, but none is load-bearing for the proposed dataset protocol or loss: the CARLA simulator is cited as an external tool, and the alignment metrics (L2, MMD, CS) are standard and compared empirically rather than imported from a self-citation. The strongest validity concern is experimental, not circular: the feature-aligned model processes two appearances per iteration while the Random baseline processes one, so the gains could partly reflect increased data throughput or effective batch size. That is a confound and an alternative explanation, but it is not a reduction of the claimed result to its inputs by construction, and it does not meet the standard for flagging circularity. The paper's quantitative claims are therefore self-contained against external benchmarks, and no circular step is exhibited.

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

The central claim does not depend on fitted scalar parameters. The fixed lambda=1/L is a design choice, not tuned to target data; the four weather conditions and the preference for cosine similarity are model choices. The main unverified inputs are the simulator determinism assumption and the validity of pixel-level feature alignment as a learning signal. No new physical or conceptual entities are introduced; the CARLA-4AGT dataset and the feature alignment loss are artifacts, not postulated entities.

assumptions (4)
  • domain assumption CARLA deterministic replay with a fixed seed yields pixel-perfect, spatially aligned ground truths across weather conditions.
    Section 3.1 asserts full determinism, but no alignment error is quantified; the multi-layer feature alignment loss in Equation 2 assumes features at the same spatial location correspond to the same object for all appearances.
  • domain assumption Forcing feature similarity across aligned appearances is a valid inductive bias for semantic segmentation.
    Section 3.2 argues this produces appearance-agnostic features; it is plausible but not proven that aligning features does not suppress useful appearance cues (e.g., for night or fog detection).
  • domain assumption The chosen UDA target loss LT is orthogonal to the feature alignment loss and can be combined without interaction artifacts.
    The framework is claimed agnostic to any LT, but experiments only test four UDA objectives (DAFormer, HRDA, MIC, ADVENT) in Section 4.3.
  • domain assumption Cityscapes, ACDC, and Dark Zurich are sufficient benchmarks to establish domain generalization gains.
    Section 4.2 evaluates only these three real datasets; conclusions about general DG capability are based on this limited set.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Leveraging Contrastive Learning for Semantic Segmentation with Consistent Labels Across Varying Appearances." pith.science (2026). https://pith.science/paper/ZIIK6JBK

@misc{pith2026241216592,
  author       = {Pith},
  title        = {Pith review of: Leveraging Contrastive Learning for Semantic Segmentation with Consistent Labels Across Varying Appearances},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ZIIK6JBK}},
  note         = {Machine review of arXiv:2412.16592}
}
read the original abstract

This paper introduces a novel synthetic dataset that captures urban scenes under a variety of weather conditions, providing pixel-perfect, ground-truth-aligned images to facilitate effective feature alignment across domains. Additionally, we propose a method for domain adaptation and generalization that takes advantage of the multiple versions of each scene, enforcing feature consistency across different weather scenarios. Our experimental results demonstrate the impact of our dataset in improving performance across several alignment metrics, addressing key challenges in domain adaptation and generalization for segmentation tasks. This research also explores critical aspects of synthetic data generation, such as optimizing the balance between the volume and variability of generated images to enhance segmentation performance. Ultimately, this work sets forth a new paradigm for synthetic data generation and domain adaptation.

Figures

Figures reproduced from arXiv: 2412.16592 by the authors.

Figure 1
Figure 1. Visual summary of Domain Adaptation research across [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Visual example of artefacts introduced by a Stable Dif [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 4
Figure 4. Visual summary of our proposed Feature Alignment [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗
Figures from the paper (3 more)
Figure 5
Figure 5. Figure 5: Visual summary of our proposed UDA training framework to exploit the different versions of the image. For visualization [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 6
Figure 6. Figure 6: Performance comparison as the dataset size increases [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]
Figure 7
Figure 7. Figure 7: Performance comparison of employing features ex [PITH_FULL_IMAGE:figures/full_fig_p009_7.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

50 extracted references · 43 canonical work pages

  1. [1]

    The cityscapes dataset for semantic urban scene understanding

    Marius Cordts, Mohamed Omran, Sebastian Ramos, Timo Rehfeld, Markus Enzweiler, Rodrigo Benenson, Uwe Franke, Stefan Roth, and Bernt Schiele. The cityscapes dataset for semantic urban scene understanding. In IEEE Conf. Comput. Vis. Pattern Recognit. (CVPR), pages 3212– 3223, 2016. 1, 2, 4, 7

    work page 2016

  2. [2]

    On exploring weakly supervised domain adaptation strategies for seman- tic segmentation using synthetic data

    Roberto Alcover-Couso, Juan C SanMiguel, Marcos Escudero-Vinolo, and Alvaro Garcia-Martin. On exploring weakly supervised domain adaptation strategies for seman- tic segmentation using synthetic data. Multimedia Tools and Applications, pages 35879–35911, 2023. 1

    work page 2023

  3. [3]

    SanMiguel, Marcos Escudero-Vi˜nolo, and Pablo Caballeira

    Roberto Alcover-Couso, Juan C. SanMiguel, Marcos Escudero-Vi˜nolo, and Pablo Caballeira. Per-class curriculum for unsupervised domain adaptation in semantic segmenta- tion. In The Visual Computer, pages 1–19, 2023. 1

    work page 2023

  4. [4]

    Exploiting semantic segmentation to boost reinforcement learning in video game environments

    Javier Montalvo, ´Alvaro Garc´ıa-Mart´ın, and Jes ´us Besc ´os. Exploiting semantic segmentation to boost reinforcement learning in video game environments. Multimedia Tools and Applications, 82(7):10961–10979, 2023. 1

    work page 2023

  5. [5]

    Gradient-based Class Weighting for Unsupervised Domain Adaptation in Dense Prediction Visual Tasks

    Roberto Alcover-Couso, Marcos Escudero-Vi ˜nolo, Juan C. SanMiguel, and Jesus Besc ´os. Gradient-based class weight- ing for unsupervised domain adaptation in dense prediction visual tasks. arXiv:2407.01327, 2024. 1

    work page Pith review arXiv 2024

  6. [6]

    Layer-wise Model Merging for Unsupervised Domain Adaptation in Segmentation Tasks

    Roberto Alcover-Couso, Juan C. SanMiguel, Marcos Escudero-Vi˜nolo, and Jose M Mart ´ınez. Layer-wise model merging for unsupervised domain adaptation in segmenta- tion tasks. arXiv:2409.15813, 2024. 1

    work page Pith review arXiv 2024

  7. [7]

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

    Lukas Hoyer, Dengxin Dai, and Luc Van Gool. DAFormer: Improving network architectures and training strategies for domain-adaptive semantic segmentation. In IEEE Conf. Comput. Vis. Pattern Recognit. (CVPR) , pages 9924–9935,

  8. [8]

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

    Lukas Hoyer, Dengxin Dai, and Luc Van Gool. HRDA: Context-aware high-resolution domain-adaptive semantic segmentation. In IEEE Eur. Conf. Comput. Vis. (ECCV) , pages 372–391, 2022. 1, 3, 4, 10, 11

    work page 2022

Show all 50 references
  1. [9]

    MIC: Masked image consistency for context- enhanced domain adaptation

    Lukas Hoyer, Dengxin Dai, Haoran Wang, and Luc Van Gool. MIC: Masked image consistency for context- enhanced domain adaptation. In IEEE Conf. Comput. Vis. Pattern Recognit. (CVPR), 2023. 1, 3, 4, 10, 11

    2023

  2. [10]

    Dacs: Domain adaptation via cross- domain mixed sampling

    Wilhelm Tranheden, Viktor Olsson, Juliano Pinto, and Lennart Svensson. Dacs: Domain adaptation via cross- domain mixed sampling. IEEE Winter Conf. App. Comp. Vis. (WACV), pages 1378–1388, 2020. 1, 3, 6

    2020

  3. [11]

    Adversarial style discrepancy minimization for unsu- pervised domain adaptation

    Xin Luo, Wei Chen, Zhengfa Liang, Chen Li, and Yusong Tan. Adversarial style discrepancy minimization for unsu- pervised domain adaptation. Neural Networks, pages 216– 225, 2023. 1

    2023

  4. [12]

    Unsupervised pixel-level domain adaptation with generative adversarial networks

    Konstantinos Bousmalis, Nathan Silberman, Google Re- search, New York, David Dohan, Dumitru Erhan, Google Brain, San Francisco, and Dilip Krishnan. Unsupervised pixel-level domain adaptation with generative adversarial networks. In IEEE Conf. Comput. Vis. Pattern Recognit. (CV...

    2019

  5. [13]

    Stage-aware feature alignment network for real-time seman- tic segmentation of street scenes

    Xi Weng, Yan Yan, Si Chen, Jing-Hao Xue, and Hanzi Wang. Stage-aware feature alignment network for real-time seman- tic segmentation of street scenes. IEEE Transactions on Cir- cuits and Systems for Video Technology , pages 4444–4459,

  6. [14]

    Advent: Adversarial entropy mini- mization for domain adaptation in semantic segmentation

    Tuan-Hung Vu, Himalaya Jain, Maxime Bucher, Matthieu Cord, and Patrick P´erez. Advent: Adversarial entropy mini- mization for domain adaptation in semantic segmentation. In IEEE Conf. Comput. Vis. Pattern Recognit. (CVPR) , pages 2512–2521, 2019. 1, 3, 10, 11

    2019

  7. [15]

    SanMiguel, and Jose M

    Roberto Alcover-Couso, Marcos Escudero-Vinolo, Juan C. SanMiguel, and Jose M. Martinez. Soft labelling for se- mantic segmentation: Bringing coherence to label down- sampling. Computational Visual Media, 2025. 1

    2025

  8. [16]

    Encoder-decoder with atrous separable convolution for semantic image segmenta- tion

    Liang-Chieh Chen, Yukun Zhu, George Papandreou, Flo- rian Schroff, and Hartwig Adam. Encoder-decoder with atrous separable convolution for semantic image segmenta- tion. In IEEE Eur. Conf. Comput. Vis. (ECCV), pages 833– 851, 2018. 1

    2018

  9. [17]

    Hierarchi- cal multi-scale attention for semantic segmentation

    Andrew Tao, Karan Sapra, and Bryan Catanzaro. Hierarchi- cal multi-scale attention for semantic segmentation. CoRR, abs/2005.10821, 2020. 1

    2005 arXiv

  10. [18]

    Training deep networks with synthetic data: Bridging the reality gap by domain randomization

    Jonathan Tremblay, Aayush Prakash, David Acuna, Mark Brophy, Varun Jampani, Cem Anil, Thang To, Eric Camer- acci, Shaad Boochoon, and Stan Birchfield. Training deep networks with synthetic data: Bridging the reality gap by domain randomization. IEEE Conf. Comput. Vis. Pattern ...

    2018

  11. [19]

    Stochastic classifiers for unsuper- vised domain adaptation

    Zhihe Lu, Yongxin Yang, Xiatian Zhu, Cong Liu, Yi-Zhe Song, and Tao Xiang. Stochastic classifiers for unsuper- vised domain adaptation. InIEEE Conf. Comput. Vis. Pattern Recog. (CVPR), 2020. 1

    2020

  12. [20]

    Domain adaptive and generalizable network architectures and train- ing strategies for semantic image segmentation

    Lukas Hoyer, Dengxin Dai, and Luc Van Gool. Domain adaptive and generalizable network architectures and train- ing strategies for semantic image segmentation. IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI)., 2024. 1, 6, 10, 11

    2024

  13. [21]

    Sakaridis, D

    C. Sakaridis, D. Dai, and L. Van Gool. Acdc: The adverse conditions dataset with correspondences for semantic driving scene understanding. In IEEE Int. Conf. Comp. Vis. (ICCV), pages 10745–10755, 2021. 1, 4, 8

    2021

  14. [22]

    Map- guided curriculum domain adaptation and uncertainty-aware evaluation for semantic nighttime image segmentation.IEEE Transactions on Pattern Analysis and Machine Intelligence,

    Christos Sakaridis, Dengxin Dai, and Luc Van Gool. Map- guided curriculum domain adaptation and uncertainty-aware evaluation for semantic nighttime image segmentation.IEEE Transactions on Pattern Analysis and Machine Intelligence,

  15. [23]

    Richter, Vibhav Vineet, Stefan Roth, and Vladlen Koltun

    Stephan R. Richter, Vibhav Vineet, Stefan Roth, and Vladlen Koltun. Playing for data: Ground truth from computer games. In IEEE Eur. Conf. Comput. Vis. (ECCV), pages 102– 118, 2016. 1, 2, 3, 6, 10

    2016

  16. [24]

    The mapillary vistas dataset for semantic understanding of street scenes

    Gerhard Neuhold, Tobias Ollmann, Samuel Rota Bul `o, and Peter Kontschieder. The mapillary vistas dataset for semantic understanding of street scenes. In IEEE Int. Conf. Comput. Vis. (ICCV), pages 5000–5009, 2017. 1

    2017

  17. [25]

    German Ros, Laura Sellart, Joanna Materzynska, David Vazquez, and Antonio M. Lopez. The synthia dataset: A large collection of synthetic images for semantic segmen- tation of urban scenes. IEEE Conf. Comput. Vis. Pattern Recog. (CVPR), pages 3234–3243, 2016. 2, 4, 6, 10

    2016

  18. [26]

    SanMiguel, and Marcos Escudero-Vi˜nolo

    Roberto Alcover-Couso, Juan C. SanMiguel, and Marcos Escudero-Vi˜nolo. Biased class disagreement: detection of out of distribution instances by using differently biased se- mantic segmentation models. In Int. Conf. Comput. Vis. (IC- CVW), pages 4580–4588, 2023. 2

    2023

  19. [27]

    The robust semantic segmen- tation uncv2023 challenge results

    Xuanlong Yu, Yi Zuo, Zitao Wang, Xiaowen Zhang, Ji- axuan Zhao, Yuting Yang, Licheng Jiao, Rui Peng, Xinyi Wang, Junpei Zhang, et al. The robust semantic segmen- tation uncv2023 challenge results. In Int. Conf. Comput. Vis. (ICCVW), pages 4618–4628, 2023. 2

    2023

  20. [28]

    Denoising dif- fusion probabilistic models

    Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising dif- fusion probabilistic models. Advances in neural information processing systems, 33:6840–6851, 2020. 2

    2020

  21. [29]

    High-resolution image synthesis with latent diffusion models

    Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Bj ¨orn Ommer. High-resolution image synthesis with latent diffusion models. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 10684–10695, 2022. 2, 10

    2022

  22. [30]

    What the daam: Interpreting stable diffu- sion using cross attention

    Raphael Tang, Linqing Liu, Akshat Pandey, Zhiying Jiang, Gefei Yang, Karun Kumar, Pontus Stenetorp, Jimmy Lin, and Ferhan Ture. What the daam: Interpreting stable diffu- sion using cross attention. arXiv preprint arXiv:2210.04885,

    arXiv

  23. [31]

    SanMiguel, and Jos ´e M

    Pablo Marcos-Manch ´on, Roberto Alcover-Couso, Juan C. SanMiguel, and Jos ´e M. Mart´ınez. Open-vocabulary atten- tion maps with token optimization for semantic segmenta- tion in diffusion models. In IEEE Conf. Comput. Vis. Pattern Recog. (CVPR), 2024. 2

    2024

  24. [32]

    Spin: Spacecraft imagery for navigation

    Javier Montalvo, Juan Ignacio Bravo P ´erez-Villar, ´Alvaro Garc´ıa-Mart´ın, Pablo Carballeira, and Jes ´us Besc ´os. Spin: Spacecraft imagery for navigation. arXiv preprint arXiv:2406.07500, 2024. 3

    2024 arXiv

  25. [33]

    Self-supervised monocular depth esti- mation on unseen synthetic cameras

    Cecilia Diana-Albelda, Juan Ignacio Bravo P ´erez-Villar, Javier Montalvo, ´Alvaro Garc ´ıa-Mart´ın, and Jes ´us Besc´os Cano. Self-supervised monocular depth esti- mation on unseen synthetic cameras. In Iberoamerican Congress on Pattern Recognition, pages 449–463, 2023. 3

    2023

  26. [35]

    Lgsvl simulator: A high fidelity simulator for autonomous driving

    Guodong Rong, Byung Hyun Shin, Hadi Tabatabaee, Qiang Lu, Steve Lemke, M ¯artin ¸ˇs Mo ˇzeiko, Eric Boise, Geehoon Uhm, Mark Gerow, Shalin Mehta, et al. Lgsvl simulator: A high fidelity simulator for autonomous driving. In Int. Conf. Int. Transp. Sys. (ITSC), pages 1–6, 2020. 3

    2020

  27. [36]

    sch ¨afer, Nico M

    Manuel Schwonberg, Joshua Niemeijer, Jan-Aike Term¨ohlen, J ¨org P. sch ¨afer, Nico M. Schmidt, Hanno Gottschalk, and Tim Fingscheidt. Survey on unsuper- vised domain adaptation for semantic segmentation for visual perception in automated driving. IEEE Access , 11:54296–54336, 2023. 3

    2023

  28. [37]

    Nguyen, Trinh V

    Kieu Dang Nam, Tu M. Nguyen, Trinh V . Dieu, Muriel Visani, Thi-Oanh Nguyen, and Dinh Viet Sang. A novel unsupervised domain adaption method for depth-guided se- mantic segmentation using coarse-to-fine alignment. IEEE Access, 10:101248–101262, 2022. 3

    2022

  29. [38]

    Dsp: Dual soft-paste for unsupervised domain adaptive semantic segmentation

    Li Gao, Jing Zhang, Lefei Zhang, and Dacheng Tao. Dsp: Dual soft-paste for unsupervised domain adaptive semantic segmentation. In ACM Int. Conf. Multimedia (MM) , pages 2825–2833, 2021. 3

    2021

  30. [39]

    Weighted and class-specific maxi- mum mean discrepancy for unsupervised domain adaptation

    Hongliang Yan, Zhetao Li, Qilong Wang, Peihua Li, Yong Xu, and Wangmeng Zuo. Weighted and class-specific maxi- mum mean discrepancy for unsupervised domain adaptation. IEEE Transactions on Multimedia, PP:1–1, 2019. 3

    2019

  31. [40]

    H. Yan, Y . Ding, P. Li, Q. Wang, Y . Xu, and W. Zuo. Mind the class weight bias: Weighted maximum mean discrepancy for unsupervised domain adaptation. In IEEE Conf. Comput. Vis. Pattern Recognit. (CVPR), pages 945–954, 2017. 3

    2017

  32. [41]

    Visual domain adaptation through lo- cality information

    Devika A.K., Rakesh Kumar Sanodiya, Babita Roslind Jose, and Jimson Mathew. Visual domain adaptation through lo- cality information. Engineering Applications of Artificial In- telligence, 123:106172, 2023. 3

    2023

  33. [42]

    Learning to adapt structured output space for semantic seg- mentation

    Yi-Hsuan Tsai, Wei-Chih Hung, Samuel Schulter, Ki- hyuk Sohn, Ming-Hsuan Yang, and Manmohan Chandraker. Learning to adapt structured output space for semantic seg- mentation. In IEEE Conf. Comput. Vis. Pattern Recognit. (CVPR), pages 7472–7481, 2018. 3

    2018

  34. [43]

    Road: Reality ori- ented adaptation for semantic segmentation of urban scenes

    Yuhua Chen, Wen Li, and Luc Van Gool. Road: Reality ori- ented adaptation for semantic segmentation of urban scenes. In IEEE Conf. Comput. Vis. Pattern Recognit. (CVPR), pages 7892–7901, 2017. 3

    2017

  35. [44]

    Borgwardt, Malte J

    Arthur Gretton, Karsten M. Borgwardt, Malte J. Rasch, Bernhard Sch ¨olkopf, and Alexander Smola. A kernel two- sample test. Journal of Machine Learning Research, 13:723– 773, 2012. 3

    2012

  36. [45]

    Domain condi- tioned adaptation network

    Shuang Li, Chi Harold Liu, Qiuxia Lin, Binhui Xie, Zheng- ming Ding, Gao Huang, and Jian Tang. Domain condi- tioned adaptation network. InConf. Art. Intell. (AAAI), pages 11386–11393, 2020. 3

    2020

  37. [46]

    Enhanced online test-time adaptation with feature-weight cosine alignment

    WeiQin Chuah, Ruwan Tennakoon, and Alireza Bab- Hadiashar. Enhanced online test-time adaptation with feature-weight cosine alignment. arXiv preprint arXiv:2405.07171, 2024. 3

    2024 arXiv

  38. [47]

    Self-training domain adaptation via weight trans- mission between generators

    Xing Wei, Zhaoxin Ji, Fan Yang, Chong Zhao, Bin Wen, and Yang Lu. Self-training domain adaptation via weight trans- mission between generators. In IEEE International Confer- ence on Acoustics, Speech and Signal Processing (ICASSP), pages 3280–3284, 2024. 3

    2024

  39. [48]

    Pseudo-label assisted optimization of multi-branch net- work for cross-domain person re-identification

    Zhengyang Wang, Shuxiang Guo, Xue Shang, and Xiufen Ye. Pseudo-label assisted optimization of multi-branch net- work for cross-domain person re-identification. In IEEE International Conference on Mechatronics and Automation (ICMA), pages 13–18, 2023. 3

    2023

  40. [49]

    CARLA: An open urban driving simulator

    Alexey Dosovitskiy, German Ros, Felipe Codevilla, Antonio Lopez, and Vladlen Koltun. CARLA: An open urban driving simulator. In Conference on Robot Learning , pages 1–16,

  41. [50]

    The pascal visual object classes (voc) challenge

    Mark Everingham, Luc Van Gool, Christopher KI Williams, John Winn, and Andrew Zisserman. The pascal visual object classes (voc) challenge. International Journal of Computer Vision (IJCV), 88:303–338, 2010. 6

    2010

  42. [51]

    Style-hallucinated dual consistency learning for domain generalized semantic segmentation

    Yuyang Zhao, Zhun Zhong, Na Zhao, Nicu Sebe, and Gim Hee Lee. Style-hallucinated dual consistency learning for domain generalized semantic segmentation. In European conference on computer vision , pages 535–552. Springer,

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.