pith. sign in

arxiv: 2505.09971 · v4 · submitted 2025-05-15 · 💻 cs.CV

APCoTTA: Continual Test-Time Adaptation for Semantic Segmentation of Airborne LiDAR Point Clouds

Yuan Gao , Shaobo Xia , Sheng Nie , Cheng Wang , Xiaohuan Xi , Bisheng Yang This is my paper

Pith reviewed 2026-05-22 15:24 UTC · model grok-4.3

classification 💻 cs.CV
keywords continual test-time adaptationsemantic segmentationairborne LiDAR point cloudsdomain shiftcatastrophic forgetting
0
0 comments X

The pith

APCoTTA lets segmentation models for airborne LiDAR point clouds adapt continuously to new unlabeled scans by selectively updating layers and filtering unreliable data.

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

The paper sets out to solve performance drops in fixed semantic segmentation models when airborne laser scanning point clouds encounter ongoing changes from sensors or environments. It does this with a continual test-time adaptation approach that works on streams of unlabeled data. The method uses gradient signals to pick which layers to adjust while leaving stable ones untouched, drops high-entropy samples to avoid reinforcing mistakes, and randomly mixes current adapted weights back toward the original source weights. A reader should care because real-world large-scale 3D mapping never stops producing new data distributions, yet collecting fresh labels for retraining is costly. The work also supplies two concrete benchmarks so others can measure progress on the same problem.

Core claim

APCoTTA adapts a source model to evolving target domains through three coordinated steps: a gradient-driven layer selection that updates only low-confidence layers to retain source knowledge, an entropy-based consistency loss applied only to reliable samples to curb error buildup, and stochastic interpolation of adapted parameters with source parameters to keep a balance between the two. These steps are tested on the newly built ISPRSC and H3DC benchmarks for ALS point cloud segmentation, where the method raises mean intersection-over-union by roughly 9 percent and 14 percent compared with running the source model directly.

What carries the argument

Gradient-driven layer selection combined with entropy-based consistency filtering on reliable samples and random interpolation of adapted and source parameters.

If this is right

  • Models can keep improving on unlabeled data streams without overwriting earlier training.
  • Discarding uncertain samples during adaptation reduces the chance that mistakes compound over time.
  • The released ISPRSC and H3DC benchmarks give a shared testbed for comparing future continual adaptation techniques on ALS data.

Where Pith is reading between the lines

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

  • The same layer-selection and interpolation ideas could be tried on other 3D sensors such as terrestrial or mobile laser scanners.
  • In production mapping pipelines the approach might cut the frequency of full retraining cycles.
  • Performance under more extreme sensor calibration drifts or seasonal vegetation changes remains an open test.

Load-bearing premise

The three mechanisms together stop catastrophic forgetting and error accumulation for domain shifts beyond the two specific benchmarks that were introduced.

What would settle it

Run the method on a third benchmark collected with a markedly different sensor or in a new geographic region and observe whether mean intersection-over-union still rises or instead falls below the direct-inference baseline.

Figures

Figures reproduced from arXiv: 2505.09971 by Bisheng Yang, Cheng Wang, Shaobo Xia, Sheng Nie, Xiaohuan Xi, Yuan Gao.

Figure 1
Figure 1. Figure 1: Overview of the proposed framework. (a) The overall pipeline at time step t. The target domain batch Xt is processed with both weak (X w t ) and strong (X s t ) augmentations and fed into the model fθt . The “Fire” and “Snowflake” icons explicitly indicate the trainable and frozen status of layers. (b) The Dynamic Selection of Trainable Layers (DSTL) module calculates the gradient norm based on the KL dive… view at source ↗
Figure 2
Figure 2. Figure 2: The LiDAR intensity map of training (a) and testing (b) sets of ISPRS [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: The RGB color map of the training and validation sets of the H3D dataset. The validation set is indicated by the yellow box. [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Visualization of typical corruption types with the largest corruption severity level 5 in our benchmark ISPRSC dataset. [PITH_FULL_IMAGE:figures/full_fig_p009_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Visualization of typical corruption types with the largest corruption severity level 5 in our benchmark H3DC dataset. [PITH_FULL_IMAGE:figures/full_fig_p009_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Local qualitative comparison of semantic segmentation on the ISPRS to ISPRSC CTTA task. 10 [PITH_FULL_IMAGE:figures/full_fig_p010_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Local qualitative comparison of semantic segmentation on the H3D to H3DC CTTA task. 11 [PITH_FULL_IMAGE:figures/full_fig_p011_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Hyperparameter sensitivity analysis. 12 [PITH_FULL_IMAGE:figures/full_fig_p012_8.png] view at source ↗
read the original abstract

Airborne laser scanning (ALS) point cloud semantic segmentation is a fundamental task for large-scale 3D scene understanding. Fixed models deployed in real-world scenarios often suffer from performance degradation due to continuous domain shifts caused by environmental and sensor changes. Continuous Test-Time Adaptation (CTTA) enables adaptation to evolving unlabeled domains, but its application to ALS point clouds remains underexplored, hindered by the lack of benchmarks and the risks of catastrophic forgetting and error accumulation. To address these challenges, we propose APCoTTA (ALS Point cloud Continuous Test-Time Adaptation), a novel CTTA framework tailored for ALS point cloud semantic segmentation. APCoTTA consists of three key components. First, we adapt a gradient-driven layer selection mechanism for ALS point clouds, selectively updating low-confidence layers while freezing stable ones to preserve source knowledge and mitigate catastrophic forgetting. Second, an entropy-based consistency loss discards unreliable samples and enforces consistency regularization solely on reliable ones, effectively reducing error accumulation and improving adaptation stability. Third, a random parameter interpolation mechanism stochastically blends adapted parameters with source model parameters, further balancing target adaptation and source knowledge retention. Finally, we construct two benchmarks, ISPRSC and H3DC, to address the lack of CTTA benchmarks for ALS point cloud segmentation. Extensive experiments demonstrate that APCoTTA achieves superior performance on both benchmarks, improving mIoU by approximately 9\% and 14\% over direct inference. The new benchmarks and code are available at https://github.com/Gaoyuan2/APCoTTA.

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

1 major / 2 minor

Summary. The paper proposes APCoTTA, a continual test-time adaptation (CTTA) framework for semantic segmentation of airborne LiDAR (ALS) point clouds. It introduces three mechanisms—gradient-driven layer selection to update only low-confidence layers while freezing stable ones, an entropy-based consistency loss that filters unreliable samples, and random parameter interpolation to blend adapted and source parameters—to mitigate catastrophic forgetting and error accumulation during continual domain shifts. The authors construct two new benchmarks (ISPRSC and H3DC) by sequencing scans with varying sensors and environments, and report mIoU gains of approximately 9% and 14% over direct inference on these benchmarks, with code and benchmarks released publicly.

Significance. If the central claims hold after addressing the issues below, the work would be a useful contribution to CTTA for ALS point clouds, an area noted as underexplored. The release of two new benchmarks and reproducible code is a clear strength that supports future research. The three explicit mechanisms provide a concrete, empirical approach to balancing adaptation and knowledge retention without relying on self-referential definitions.

major comments (1)
  1. [Section 4] Section 4: The construction of ISPRSC and H3DC is described via sequencing of scans with varying sensors/environments, but no quantitative metrics (e.g., distribution distances, density variation statistics, or comparisons to external ALS corpora) are provided to characterize the domain shifts. This is load-bearing for the central claim, as the reported mIoU superiority could arise from benchmark-specific properties rather than the mechanisms' ability to prevent forgetting and error accumulation under realistic continual shifts.
minor comments (2)
  1. [Abstract] Abstract: The reported improvements are stated as 'approximately 9% and 14%' without specifying which benchmark corresponds to which gain; this should be clarified for precision.
  2. The manuscript would benefit from explicit ablation results isolating the contribution of each of the three mechanisms (layer selection, entropy filtering, parameter interpolation) to the overall gains.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the constructive feedback on our manuscript. We address the major comment below and will incorporate revisions to strengthen the characterization of the domain shifts.

read point-by-point responses

  1. Referee: [Section 4] Section 4: The construction of ISPRSC and H3DC is described via sequencing of scans with varying sensors/environments, but no quantitative metrics (e.g., distribution distances, density variation statistics, or comparisons to external ALS corpora) are provided to characterize the domain shifts. This is load-bearing for the central claim, as the reported mIoU superiority could arise from benchmark-specific properties rather than the mechanisms' ability to prevent forgetting and error accumulation under realistic continual shifts.

    Authors: We agree that quantitative characterization of the domain shifts is important to substantiate that the performance gains arise from the proposed mechanisms rather than benchmark-specific artifacts. In the revised version, we will add explicit metrics in Section 4, including distribution distances such as Earth Mover's Distance computed on point cloud features or normalized coordinates between consecutive domains, statistics on point density and intensity variations across the sequenced scans, and brief comparisons to external ALS corpora (e.g., feature distribution overlap with datasets like Semantic3D or the original ISPRS benchmark). These additions will provide evidence that the shifts in ISPRSC and H3DC are representative of realistic sensor and environmental changes in ALS data. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical mechanisms and benchmark construction are self-contained

full rationale

The paper presents an empirical CTTA framework (APCoTTA) consisting of three explicitly defined components—gradient-driven layer selection, entropy-based sample filtering with consistency loss, and random parameter interpolation—along with two newly constructed benchmarks (ISPRSC and H3DC). Performance improvements are reported from direct experimental comparison on these benchmarks rather than from any first-principles derivation, fitted-parameter prediction, or self-citation chain that reduces the central claims to the inputs by construction. No equations or steps equate outputs to inputs tautologically; the method and evaluation are independent of such reductions.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The framework rests on standard deep-learning assumptions about gradient signals reflecting layer stability and entropy indicating sample reliability; no new physical entities or ad-hoc constants are introduced beyond typical training hyperparameters.

axioms (2)
  • domain assumption Gradient magnitude reliably indicates which layers are unstable under domain shift
    Invoked to justify selective layer updating in the first component.
  • domain assumption Low-entropy predictions on unlabeled target samples are trustworthy for consistency regularization
    Basis for discarding unreliable samples in the entropy-based loss.

pith-pipeline@v0.9.0 · 5827 in / 1279 out tokens · 36004 ms · 2026-05-22T15:24:50.394933+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

61 extracted references · 61 canonical work pages · 2 internal anchors

  1. [1]

    A probabilistic framework for lifelong test-time adaptation

    Brahma, D., Rai, P., 2022. A probabilistic framework for lifelong test-time adaptation. 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , 3582–3591

    work page 2022

  2. [2]

    Multi-modal continual test-time adaptation for 3d seman- tic segmentation

    Cao, H., Xu, Y ., Yang, J., Yin, P., Yuan, S., Xie, L., 2023. Multi-modal continual test-time adaptation for 3d seman- tic segmentation. 2023 IEEE/CVF International Confer- ence on Computer Vision (ICCV) , 18763–18773

    work page 2023

  3. [3]

    Carballo, A., Lambert, J., Monrroy, A., Wong, D., Nark- sri, P., Kitsukawa, Y ., Takeuchi, E., Kato, S., Takeda, K.,

    work page

  4. [4]

    Libre: The multiple 3d lidar dataset, in: 2020 IEEE intelligent vehicles symposium (IV), IEEE. pp. 1094– 1101

    work page 2020

  5. [5]

    Chen, A., Cheng, K., Chen, Y ., Qi, Z., Yang, H., Ren, Y ., Yang, Z., Chen, M., Xu, J., Zhang, Y ., Huang, G., Xiang, T., Zhang, J., Liu, W., Lin, D., Xu, G., Guo, Q.,

    work page

  6. [6]

    Remote Sensing of Environment 329, 114957

    Validating recent global canopy height maps over china’s forests based on uav lidar data. Remote Sensing of Environment 329, 114957

    work page

  7. [7]

    Contrastive test-time adaptation, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp

    Chen, D., Wang, D., Darrell, T., Ebrahimi, S., 2022. Contrastive test-time adaptation, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 295–305

    work page 2022

  8. [8]

    Robust mean teacher for continual and gradual test-time adapta- tion

    Döbler, M., Marsden, R.A., Yang, B., 2022. Robust mean teacher for continual and gradual test-time adapta- tion. 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , 7704–7714

    work page 2022

  9. [9]

    Benchmarking robust- ness of 3d object detection to common corruptions in au- tonomous driving

    Dong, Y ., Kang, C., Zhang, J., Zhu, Z., Wang, Y ., Yang, X., Su, H., Wei, X., Zhu, J., 2023. Benchmarking robust- ness of 3d object detection to common corruptions in au- tonomous driving. 2023 IEEE/CVF Conference on Com- puter Vision and Pattern Recognition (CVPR) , 1022– 1032

    work page 2023

  10. [10]

    Semantic segmentation of airborne lidar point clouds with noisy labels

    Gao, Y ., Xia, S., Wang, C., Xi, X., Yang, B., Xie, C., 2024. Semantic segmentation of airborne lidar point clouds with noisy labels. IEEE Transactions on Geoscience and Re- mote Sensing 62, 1–14

    work page 2024

  11. [11]

    Lidar remote sensing meets weak supervision: Concepts, methods, and perspectives

    Gao, Y ., Xia, S., Wang, P., Xi, X., Nie, S., Wang, C., 2026. Lidar remote sensing meets weak supervision: Concepts, methods, and perspectives. ISPRS Journal of Photogram- metry and Remote Sensing 235, 72–104

    work page 2026

  12. [12]

    Test time adaptation via conjugate pseudo-labels

    Goyal, S., Sun, M., Raghunathan, A., Kolter, J.Z., 2022. Test time adaptation via conjugate pseudo-labels. Ad- vances in Neural Information Processing Systems 35, 6204–6218

    work page 2022

  13. [13]

    Benchmarking Neural Network Robustness to Common Corruptions and Perturbations

    Hendrycks, D., Dietterich, T., 2019. Benchmarking neural network robustness to common corruptions and perturba- tions.arXiv:1903.12261

    work page internal anchor Pith review Pith/arXiv arXiv 2019

  14. [14]

    On the importance of gradients for detecting distributional shifts in the wild

    Huang, R., Geng, A., Li, Y ., 2021. On the importance of gradients for detecting distributional shifts in the wild. Advances in Neural Information Processing Systems 34, 677–689. 14

    work page 2021

  15. [15]

    Pcotta: Con- tinual test-time adaptation for multi-task point cloud un- derstanding

    Jiang, J., Zhou, Q., Li, Y ., Zhao, X., Wang, M., Ma, L., Chang, J., Zhang, J., Lu, X., et al., 2024. Pcotta: Con- tinual test-time adaptation for multi-task point cloud un- derstanding. Advances in Neural Information Processing Systems 37, 96229–96253

    work page 2024

  16. [16]

    Benchmarking the ro- bustness of semantic segmentation models with respect to common corruptions

    Kamann, C., Rother, C., 2021. Benchmarking the ro- bustness of semantic segmentation models with respect to common corruptions. International journal of computer vision 129, 462–483

    work page 2021

  17. [17]

    Ev-tta: Test- time adaptation for event-based object recognition

    Kim, J., Hwang, I.T., Kim, Y ., 2022. Ev-tta: Test- time adaptation for event-based object recognition. 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , 17724–17733

    work page 2022

  18. [18]

    The hes- sigheim 3d (h3d) benchmark on semantic segmentation of high-resolution 3d point clouds and textured meshes from uav lidar and multi-view-stereo

    Kölle, M., Laupheimer, D., Schmohl, S., Haala, N., Rot- tensteiner, F., Wegner, J.D., Ledoux, H., 2021. The hes- sigheim 3d (h3d) benchmark on semantic segmentation of high-resolution 3d point clouds and textured meshes from uav lidar and multi-view-stereo. ISPRS Open Journal of Photogrammetry and Remote Sensing 1, 100001

    work page 2021

  19. [19]

    Robo3d: Towards ro- bust and reliable 3d perception against corruptions

    Kong, L., Liu, Y .C., Li, X., Chen, R., Zhang, W., Ren, J., Pan, L., Chen, K., Liu, Z., 2023. Robo3d: Towards ro- bust and reliable 3d perception against corruptions. 2023 IEEE/CVF International Conference on Computer Vision (ICCV) , 19937–19949

    work page 2023

  20. [20]

    On information and sufficiency

    Kullback, S., Leibler, R.A., 1951. On information and sufficiency. The annals of mathematical statistics 22, 79– 86

    work page 1951

  21. [21]

    Pseudo-label: The simple and efficient semi-supervised learning method for deep neural networks, in: Workshop on challenges in representation learning, ICML, Atlanta

    Lee, D.H., et al., 2013. Pseudo-label: The simple and efficient semi-supervised learning method for deep neural networks, in: Workshop on challenges in representation learning, ICML, Atlanta. p. 896

    work page 2013

  22. [22]

    Adversarially masking synthetic to mimic real: Adaptive noise injection for point cloud segmentation adaptation

    Li, G., Kang, G., Wang, X., Wei, Y ., Yang, Y ., 2023. Adversarially masking synthetic to mimic real: Adaptive noise injection for point cloud segmentation adaptation. 2023 IEEE/CVF Conference on Computer Vision and Pat- tern Recognition (CVPR) , 20464–20474

    work page 2023

  23. [23]

    Revisiting Batch Normalization For Practical Domain Adaptation

    Li, Y ., Wang, N., Shi, J., Liu, J., Hou, X., 2016. Revis- iting batch normalization for practical domain adaptation. arXiv:1603.04779

    work page internal anchor Pith review Pith/arXiv arXiv 2016

  24. [24]

    Neural net- work guided interpolation for mapping canopy height of china’s forests by integrating gedi and icesat-2 data

    Liu, X., Su, Y ., Hu, T., Yang, Q., Liu, B., Deng, Y ., Tang, H., Tang, Z., Fang, J., Guo, Q., 2022. Neural net- work guided interpolation for mapping canopy height of china’s forests by integrating gedi and icesat-2 data. Re- mote Sensing of Environment 269, 112844

    work page 2022

  25. [25]

    Estimating forest aboveground biomass using small-footprint full- waveform airborne lidar data

    Luo, S., Wang, C., Xi, X., Nie, S., Fan, X., Chen, H., Ma, D., Liu, J., Zou, J., Lin, Y ., et al., 2019. Estimating forest aboveground biomass using small-footprint full- waveform airborne lidar data. International Journal of Ap- plied Earth Observation and Geoinformation 83, 101922

    work page 2019

  26. [26]

    Variational continual test-time adaptation.arXiv preprint arXiv:2402.08182, 2024

    Lyu, F., Du, K., Li, Y ., Zhao, H., Zhang, Z., Liu, G., Wang, L., 2024. Variational continual test-time adaptation. arXiv preprint arXiv:2402.08182

    work page arXiv 2024

  27. [27]

    Palm: Pushing adaptive learning rate mechanisms for continual test-time adaptation, in: Proceedings of the AAAI Conference on Artificial Intelligence, pp

    Maharana, S.K., Zhang, B., Guo, Y ., 2025. Palm: Pushing adaptive learning rate mechanisms for continual test-time adaptation, in: Proceedings of the AAAI Conference on Artificial Intelligence, pp. 19378–19386

    work page 2025

  28. [28]

    Benchmarking ro- bustness in object detection: Autonomous driving when win- ter is coming

    Michaelis, C., Mitzkus, B., Geirhos, R., Rusak, E., Bringmann, O., Ecker, A.S., Bethge, M., Brendel, W., 2020. Benchmarking robustness in object de- tection: Autonomous driving when winter is coming. arXiv:1907.07484

    work page arXiv 2020

  29. [29]

    Efficient test-time model adaptation with- out forgetting, in: International conference on machine learning, PMLR

    Niu, S., Wu, J., Zhang, Y ., Chen, Y ., Zheng, S., Zhao, P., Tan, M., 2022. Efficient test-time model adaptation with- out forgetting, in: International conference on machine learning, PMLR. pp. 16888–16905

    work page 2022

  30. [30]

    Qiao, Y ., Xi, X., Nie, S., Wang, P., Guo, H., Wang, C.,

    work page

  31. [31]

    Remote Sensing 14, 4905

    Power pylon reconstruction from airborne lidar data based on component segmentation and model matching. Remote Sensing 14, 4905

    work page

  32. [32]

    Benchmarking and ana- lyzing point cloud classification under corruptions, in: In- ternational Conference on Machine Learning, PMLR

    Ren, J., Pan, L., Liu, Z., 2022. Benchmarking and ana- lyzing point cloud classification under corruptions, in: In- ternational Conference on Machine Learning, PMLR. pp. 18559–18575

    work page 2022

  33. [33]

    Unsupervised domain adaptation in lidar semantic segmentation with self-supervision and gated adapters, in: 2022 International Conference on Robotics and Automation (ICRA), IEEE

    Rochan, M., Aich, S., Corral-Soto, E.R., Nabatchian, A., Liu, B., 2022. Unsupervised domain adaptation in lidar semantic segmentation with self-supervision and gated adapters, in: 2022 International Conference on Robotics and Automation (ICRA), IEEE. pp. 2649–2655

    work page 2022

  34. [34]

    The isprs benchmark on urban object classification and 3d building reconstruc- tion

    Rottensteiner, F., Sohn, G., Jung, J., Gerke, M., Baillard, C., Benitez, S., Breitkopf, U., 2012. The isprs benchmark on urban object classification and 3d building reconstruc- tion. ISPRS Annals of the Photogrammetry, Remote Sens- ing and Spatial Information Sciences; I-3 1, 293–298

    work page 2012

  35. [35]

    Cosmix: Compositional semantic mix for domain adaptation in 3d lidar segmentation, in: Euro- pean Conference on Computer Vision, Springer

    Saltori, C., Galasso, F., Fiameni, G., Sebe, N., Ricci, E., Poiesi, F., 2022a. Cosmix: Compositional semantic mix for domain adaptation in 3d lidar segmentation, in: Euro- pean Conference on Computer Vision, Springer. pp. 586– 602

    work page

  36. [36]

    Gipso: Geometrically informed propagation for online adaptation in 3d lidar segmentation, in: European Con- ference on Computer Vision, Springer

    Saltori, C., Krivosheev, E., Lathuiliére, S., Sebe, N., Galasso, F., Fiameni, G., Ricci, E., Poiesi, F., 2022b. Gipso: Geometrically informed propagation for online adaptation in 3d lidar segmentation, in: European Con- ference on Computer Vision, Springer. pp. 567–585

    work page

  37. [37]

    A mathematical theory of commu- nication

    Shannon, C.E., 1948. A mathematical theory of commu- nication. The Bell system technical journal 27, 379–423

    work page 1948

  38. [38]

    Mm-tta: Multi- modal test-time adaptation for 3d semantic segmentation, in: 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp

    Shin, I., Tsai, Y .H., Zhuang, B., Schulter, S., Liu, B., Garg, S., Kweon, I.S., Yoon, K.J., 2022. Mm-tta: Multi- modal test-time adaptation for 3d semantic segmentation, in: 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp. 16907–16916. 15

    work page 2022

  39. [39]

    Song, J., Lee, J., Kweon, I.S., Choi, S., 2023. Ecotta: Memory-efficient continual test-time adaptation via self- distilled regularization, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 11920–11929

    work page 2023

  40. [40]

    Curved build- ings reconstruction from airborne lidar data by matching and deforming geometric primitives

    Song, J., Xia, S., Wang, J., Chen, D., 2020. Curved build- ings reconstruction from airborne lidar data by matching and deforming geometric primitives. IEEE Transactions on Geoscience and Remote Sensing 59, 1660–1674

    work page 2020

  41. [41]

    Tarvainen, A., Valpola, H., 2017. Mean teachers are bet- ter role models: Weight-averaged consistency targets im- prove semi-supervised deep learning results, in: Proceed- ings of the 31st International Conference on Neural Infor- mation Processing Systems, Curran Associates Inc., Red Hook, NY , USA. p. 1195–1204

    work page 2017

  42. [42]

    Kpconv: Flexible and deformable convolution for point clouds, in: Proceedings of the IEEE/CVF international conference on computer vision, pp

    Thomas, H., Qi, C.R., Deschaud, J.E., Marcotegui, B., Goulette, F., Guibas, L.J., 2019. Kpconv: Flexible and deformable convolution for point clouds, in: Proceedings of the IEEE/CVF international conference on computer vision, pp. 6411–6420

    work page 2019

  43. [43]

    Parameter-selective continual test- time adaptation, in: Proceedings of the Asian Conference on Computer Vision, pp

    Tian, J., Lyu, F., 2024. Parameter-selective continual test- time adaptation, in: Proceedings of the Asian Conference on Computer Vision, pp. 1384–1400

    work page 2024

  44. [44]

    Test time adaptation with regularized loss for weakly supervised salient object detection

    Veksler, O., 2023. Test time adaptation with regularized loss for weakly supervised salient object detection. 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , 7360–7369

    work page 2023

  45. [45]

    Towards online domain adaptive object detection

    Vibashan, V ., Oza, P., Patel, V .M., 2022. Towards online domain adaptive object detection. 2023 IEEE/CVF Winter Conference on Applications of Computer Vision (W ACV) , 478–488

    work page 2022

  46. [46]

    Tent: Fully test-time adaptation by entropy minimization, in: International Conference on Learning Representations

    Wang, D., Shelhamer, E., Liu, S., Olshausen, B.A., Dar- rell, T., 2021. Tent: Fully test-time adaptation by entropy minimization, in: International Conference on Learning Representations

    work page 2021

  47. [47]

    Test-time adap- tation for geospatial point cloud semantic segmentation with distinct domain shifts

    Wang, P., Yao, W., Shao, J., He, Z., 2025. Test-time adap- tation for geospatial point cloud semantic segmentation with distinct domain shifts. ISPRS Journal of Photogram- metry and Remote Sensing 229, 422–435

    work page 2025

  48. [48]

    Continual test-time domain adaptation

    Wang, Q., Fink, O., Gool, L.V ., Dai, D., 2022. Continual test-time domain adaptation. 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , 7191–7201

    work page 2022

  49. [49]

    Wang, W., Zhong, Z., Wang, W., Chen, X., Ling, C., Wang, B., Sebe, N., 2023a. Dynamically instance-guided adaptation: A backward-free approach for test-time do- main adaptive semantic segmentation, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pat- tern Recognition, pp. 24090–24099

    work page

  50. [50]

    Con- tinual test-time domain adaptation via dynamic sample se- lection

    Wang, Y ., Hong, J., Cheraghian, A., Rahman, S., Ahmedt- Aristizabal, D., Petersson, L., Harandi, M., 2023b. Con- tinual test-time domain adaptation via dynamic sample se- lection. 2024 IEEE/CVF Winter Conference on Applica- tions of Computer Vision (W ACV) , 1690–1699

    work page 2024

  51. [51]

    Wu, B., Zhou, X., Zhao, S., Yue, X., Keutzer, K., 2019. Squeezesegv2: Improved model structure and unsuper- vised domain adaptation for road-object segmentation from a lidar point cloud, in: 2019 international confer- ence on robotics and automation (ICRA), IEEE. pp. 4376– 4382

    work page 2019

  52. [52]

    Building instance mapping from als point clouds aided by polygo- nal maps

    Xia, S., Xu, S., Wang, R., Li, J., Wang, G., 2022. Building instance mapping from als point clouds aided by polygo- nal maps. IEEE Transactions on Geoscience and Remote Sensing 60, 1–13

    work page 2022

  53. [53]

    Transfer learning from synthetic to real lidar point cloud for semantic segmentation

    Xiao, A., Huang, J., Guan, D., Zhan, F., Lu, S., 2022. Transfer learning from synthetic to real lidar point cloud for semantic segmentation. Proceedings of the AAAI Conference on Artificial Intelligence , 2795–2803

    work page 2022

  54. [54]

    Benchmarking the robustness of lidar semantic segmen- tation models

    Yan, X., Zheng, C., Xue, Y ., Li, Z., Cui, S., Dai, D., 2024. Benchmarking the robustness of lidar semantic segmen- tation models. International Journal of Computer Vision 132, 2674–2697

    work page 2024

  55. [55]

    Yi, L., Gong, B., Funkhouser, T., 2021. Complete & label: A domain adaptation approach to semantic segmentation of lidar point clouds, in: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp. 15363–15373

    work page 2021

  56. [56]

    Robust test-time adaptation in dynamic scenarios, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp

    Yuan, L., Xie, B., Li, S., 2023. Robust test-time adaptation in dynamic scenarios, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 15922–15932

    work page 2023

  57. [57]

    Yuan, Y ., Xu, B., Hou, L., Sun, F., Shen, H., Cheng, X.,

    work page

  58. [58]

    23901–23911

    Tea: Test-time energy adaptation, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 23901–23911

    work page

  59. [59]

    Aux- adapt: Stable and efficient test-time adaptation for tem- porally consistent video semantic segmentation

    Zhang, Y ., Borse, S., Cai, H., Porikli, F.M., 2021. Aux- adapt: Stable and efficient test-time adaptation for tem- porally consistent video semantic segmentation. 2022 IEEE/CVF Winter Conference on Applications of Com- puter Vision (W ACV) , 2633–2642

    work page 2021

  60. [60]

    Reshaping the online data buffering and organizing mechanism for continual test-time adaptation, in: European Conference on Computer Vision, Springer

    Zhu, Z., Hong, X., Ma, Z., Zhuang, W., Ma, Y ., Dai, Y ., Wang, Y ., 2024. Reshaping the online data buffering and organizing mechanism for continual test-time adaptation, in: European Conference on Computer Vision, Springer. pp. 415–433

    work page 2024

  61. [61]

    Hgl: Hierarchical geometry learning for test- time adaptation in 3d point cloud segmentation, in: Euro- pean Conference on Computer Vision, Springer

    Zou, T., Qu, S., Li, Z., Knoll, A., He, L., Chen, G., Jiang, C., 2024. Hgl: Hierarchical geometry learning for test- time adaptation in 3d point cloud segmentation, in: Euro- pean Conference on Computer Vision, Springer. pp. 19– 36. 16 Appendix .1. Implementation Details for Corruption Robustness Benchmarks In this section, we provide the parameters and e...

    work page 2024