Pith. sign in

REVIEW 3 major objections 5 minor 23 references

Data-Efficient Point Cloud Semantic Segmentation Pipeline for Unimproved Roads

T0 review · 3 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read With only 37 labeled training scans and a frozen pre-trained feature extractor, a LiDAR segmentation pipeline lifts eight-class mean IoU on dirt and gravel roads from 33.5% to 51.8%.

desk verdict A useful low-data LiDAR segmentation case study whose central multi-dataset pretraining claim is confounded by the inclusion of Target scans in the Combined arm. read the letter →

arxiv 2508.20135 v1 pith:OZGHFWT4 submitted 2025-08-26 eess.IV cs.AI

Andrew Yarovoi , Christopher R. Valenta This is my paper · ORCID
classification eess.IVcs.AI
keywords pointcloudsemanticsegmentationfew-shotlearningtransferLiDARunimprovedroadsmulti-datasetpretrainingPromptTrainingManifoldMixup
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper tries to show that a two-stage training recipe lets a LiDAR point-cloud segmentation network learn to segment unimproved roads—dirt, gravel, forest trails—from just 37 labeled scans. First, the FRNet backbone is pretrained on a mixture of large urban datasets (SemanticKITTI, Waymo) plus a small slice of the target data; then the backbone is frozen and only a light MLP prediction head is fine-tuned on the target scans. On the target validation set, the pipeline lifts 8-class mean Intersection-over-Union from 33.5% to 51.8% and overall accuracy from 85.5% to 90.8% relative to training only on the 37 scans. The authors argue that pretraining across multiple datasets, not any single urban dataset, is what makes the transfer work, and they support the recipe with ablations of prompt-normalization, Manifold Mixup, and an extra ambient-intensity channel.

What carries the argument

The central object is the FRNet projection-based LiDAR segmentation network, used as a frozen feature extractor after multi-dataset pretraining. Around it, the paper builds a two-stage transfer mechanism: pretrain on a mixed corpus with dataset-specific prompt-normalization (Point Prompt Training) that adds a learned mean-shift and scaling to each normalization layer, then freeze the backbone and train only a point-wise inverted-bottleneck MLP head on 37 in-domain scans. A separate shared linear layer injects each point's histogram-normalized ambient value into the head, and Manifold Mixup—linear interpolation of hidden feature vectors toward interpolated soft labels—is tested as a regularizer. The pipeline's work is to keep the expensive, generalizable representation fixed while letting the small head adapt to the target domain under extreme label scarcity.

What would settle it

Run the missing control arm: pretrain on SemanticKITTI plus Waymo with Target scans excluded, then fine-tune on the 37 Target scans. If mIoU stays near the reported 51.8%, the multi-dataset diversity explanation is supported; if it falls toward the 36–39% range of single-dataset fine-tuned arms, then in-domain exposure during pretraining, not dataset diversity, is the active ingredient.

Watch

Extended reading notes

Core claim

The paper's central claim is that data-efficient semantic segmentation of a new, sensor-different domain is achievable by decoupling a generic feature extractor from a small task-specific classifier. In concrete terms: pre-training FRNet on a combined corpus of SemanticKITTI, Waymo Open Dataset, and 37 Target scans, then freezing it and fine-tuning only an inverted-bottleneck MLP head on the 37 Target scans, raises validation mIoU to 51.80% and overall accuracy to 90.76%, from 33.51% and 85.52% under Target-only training. The authors interpret the large pre-fine-tuning gain of the Combined model (42.48% mIoU before fine-tuning, versus under 1% for KITTI-only or Waymo-only) as evidence that exposure to diverse sensors and scene geometries produces more transferable, geometry-based features. They also report that Point Prompt Training improves the final fine-tuned model, that Manifold Mixup helps pretraining but hurts after fine-tuning, and that feeding histogram-normalized ambient returns into the head improves mIoU by 3.57 points after fine-tuning.

Load-bearing premise

The load-bearing premise is that the Combined pretraining result isolates dataset diversity; in the experiments, only Combined includes the 37 Target training scans in the pretraining mixture, while KITTI-only and Waymo-only arms do not, and no KITTI+Waymo-without-Target arm was run. If that premise gives way, the measured mIoU gain may still hold but the paper's explanation for why it works collapses.

Editorial extensions

If this is right

  • A deployment team with only a few dozen labeled scans of a new environment can leverage public urban LiDAR datasets to get usable segmentation instead of training from scratch.
  • Freezing the feature extractor removes the need for large in-domain compute during adaptation; only the small MLP head and context embedding are updated.
  • Multi-dataset pretraining is the main lever: before any fine-tuning, the combined-pretraining model already reaches 42.48% mIoU, while single-dataset models stay below 1%.
  • Point Prompt Training should be paired with fine-tuning; without it, the unoptimized target context vector can hurt rare classes.
  • Manifold Mixup should be considered for pretraining generalization but disabled before fine-tuning on the target domain, because it caps the head's specificity there.

Reading between the lines

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

  • A missing control arm—pretraining on SemanticKITTI plus Waymo without any Target scans—would decide whether the headline gain is caused by dataset diversity or simply by the Target scans appearing in the pretraining mixture; the current tables cannot separate these.
  • The ambient-channel result suggests a cheap extension for other sensors that provide extra radiometric channels: feed sensor-specific scalar channels into the prediction head rather than the feature extractor, preserving pretrainability on public benchmarks.
  • Because the Target validation set has only one distant person (14 points), the people-class IoU of 0.0% is not yet informative; a larger target set with more pedestrians would be needed to test whether multi-dataset pretraining actually transfers to rare classes.
  • The failure of Manifold Mixup after fine-tuning hints that interpolation-based regularizers may need to be annealed or applied only during the frozen-feature stage; testing a schedule that turns Mixup off in the final epochs could recover both generalization and specificity.
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

3 major / 5 minor

Summary. This manuscript presents a two-stage training pipeline for 8-class LiDAR point-cloud semantic segmentation of unimproved roads. A frustum-range CNN (FRNet) is pretrained on SemanticKITTI, Waymo Open Dataset, and the 37-scan Target training set, after which the feature extractor is frozen and only a lightweight MLP head is fine-tuned on the Target training set. The paper additionally explores Point Prompt Training applied to batch normalization, Manifold Mixup in the MLP head, and injection of histogram-normalized ambient values into the head. Experiments on a 13-scan validation set report mIoU improving from 33.51% (Target-only training) to 51.80% and overall accuracy from 85.52% to 90.76%. The paper concludes that multi-dataset pretraining is key to generalization.

Significance. If the reported results are reproducible, the pipeline is practically useful for low-data domain adaptation in LiDAR segmentation, and the code release is a concrete asset. The study also provides useful negative evidence about Manifold Mixup in this setting and extends PPT to a convolutional architecture. However, the two central interpretive claims — that multi-dataset diversity rather than in-domain pretraining exposure drives the gain, and that the precise magnitude 51.80/90.76 is an unbiased estimate — are not established by the experiments as reported.

major comments (3)
  1. [IV.A, Table I] The Combined pretraining arm contains the 37 Target training scans, while the KITTI-Only and Waymo-Only arms do not, and no KITTI+Waymo-without-Target arm is reported. The large pre-fine-tune gap (42.48 vs <1% mIoU) and post-fine-tune gap (51.80 vs 39.13 and 36.22 mIoU) can be fully explained by the model having seen the in-domain training set during pretraining, rather than by dataset diversity. The sentence 'Given that the Target data constituted less than 1% of the mixed dataset, this gain stemmed predominantly from improved generalization' is not justified by this design. Please add a Combined-without-Target pretraining arm, or soften the abstract and Section IV.A claims to 'pretraining that includes in-domain data' and explicitly discuss this confound.
  2. [IV.A, Tables I-V] All reported metrics are computed on the same 13-scan validation set that was used for early stopping and for choosing among ablations. The final configuration (PPT on, MM off, ambient on) was selected by maximizing mIoU on that set, and the headline 51.80% mIoU is the maximum of that selection sweep. Thus the reported improvement is a selected estimate, not an unbiased evaluation. Please hold out a test set disjoint from early stopping and model selection, or report selection-adjusted estimates or confidence intervals; at minimum, clearly characterize all numbers as validation-set performance.
  3. [IV.E, Table V] The class-level reliability of the headline metric is limited: the 'people' class has only 14 labeled points in the validation set and per-class IoU 0.0, and the authors note that 'outlier' supervision is inconsistent across source datasets. The single scalar mIoU is therefore dominated by road, ground, and vegetation and should not be read as representative performance over all eight classes. Please report class-wise results together with per-class point counts and frame the 8-class claim accordingly.
minor comments (5)
  1. [III.A] The mapping from SemanticKITTI's 34 classes and Waymo's 23 classes to the 8 target classes is not specified; please provide the exact class mapping table or appendix.
  2. [III.B, III.D] The fine-tuning description is ambiguous about prompt-normalization parameters: the text says the feature extractor is frozen and the linear layers within prompt-normalization are frozen while the context embedding adapts, but earlier it says only the MLP head is trained. Please clarify which parameters are updated during fine-tuning and how this relates to the original PPT procedure.
  3. [Tables II-IV] Each ablation table should state explicitly which other components are enabled or disabled; currently the reader must infer from the text (for example, Table II's no-PPT/no-fine-tune baseline is 45.40 mIoU while Table I's Combined no-fine-tune entry is 42.48).
  4. [II.A] There is a typo in 'a simple linear classifer'; it should read 'classifier'.
  5. [IV.A] The table captions refer to dark green and dark red relative-change formatting, which may be lost in grayscale or inaccessible formats; please ensure the numeric deltas remain readable independently.

Circularity Check

2 steps flagged · score 4.0 of 10

Headline validation numbers are selected on the same 13-scan validation set used for early stopping and ablations, and the 'multi-dataset diversity is key' conclusion is confounded because only the Combined arm contains the Target training scans.

  1. fitted input called prediction [Section IV (Experiments, first paragraph) and Section IV.E / Table V caption]
    "Throughout all experiments, the validation set was used to monitor model performance and to enable early stopping, thereby mitigating the risk of overfitting. ... Final model performance on the Target validation set measured as mean Intersection-over-Union (mIoU), overall accuracy (Acc), mean class accuracy (mAcc), and per-class IoU results."

    The headline numbers (51.80% mIoU and 90.76% Acc) are measured on the 13-scan Target validation set. The same validation set drove early stopping (checkpoint selection) and the ablation studies (configuration selection: PPT enabled, MM disabled, ambient enabled). No held-out test set is reported anywhere in the paper. Consequently, the claimed 18.29-point mIoU improvement is the maximum of a validation sweep, i.e., the outcome of a selection procedure evaluated on the exact labels used to select it, rather than an out-of-sample prediction of the chosen configuration. The reported magnitude is therefore an in-sample, selection-maximized estimate presented as the method's result.

  2. other [Section IV.A (Multi-Dataset Training) and Table I]
    "In contrast, pretraining on the combined corpus of the Waymo Open Dataset, SemanticKITTI, and the Target dataset yielded substantial performance improvements even before fine-tuning. Given that the Target data constituted less than 1% of the mixed dataset, this gain stemmed predominantly from improved generalization."

    Table I defines 'Combined' pretraining as SemanticKITTI plus Waymo plus the 37 Target training scans, whereas 'KITTI Only' and 'Waymo Only' contain no Target data, and no KITTI+Waymo-without-Target arm is run. The Combined arm's advantage (42.48 mIoU before fine-tuning, 51.80 after) over the single-dataset arms is therefore equally consistent with the model having pre-trained on the in-domain training scans as with dataset diversity. The sentence claiming the gain 'stemmed predominantly from improved generalization' credits a mechanism the experiment as constructed cannot isolate; the multi-dataset diversity conclusion is an untested attribution baked into the design rather than a measured result.

full rationale

The paper contains no self-citation chain: every cited work (FRNet, Point Prompt Training, Manifold Mixup, PointNeXt, and the FSL references) is external, so the self-citation, imported-uniqueness, and ansatz-via-citation patterns do not apply, and the transfer-learning pipeline is not a renamed known result. The two-stage framework has independent content: Table I shows that even single-dataset pretraining followed by fine-tuning (KITTI-Only 39.13 mIoU, Waymo-Only 36.22) exceeds Target-only training (33.51), so the transfer strategy itself is not an artifact of the design. However, two load-bearing claims are partially circular or confounded. First, the headline numbers are the selected maximum of the validation sweep: the same 13-scan validation set drove early stopping, ablation selection, and the final reported performance, with no held-out test set, so the precise 18.29-point mIoU gain is not an out-of-sample prediction. Second, the Abstract's and Section IV.A's central attribution that 'pre-training across multiple datasets is key to improving generalization' is not isolated, because the Combined arm includes the 37 Target training scans during pretraining while the single-dataset arms do not, and no multi-dataset-without-Target control exists. The paper itself notes further non-circular limitations (Section V: augmentation ablations were not run; Table V: only 14 labeled 'people' points), which temper the confidence in the exact numbers but do not invalidate the overall pipeline. Net: partial circularity in the magnitude and mechanism claims, not a wholesale reduction of the result to its inputs.

Assumptions & free parameters 6 free parameters · 6 assumptions · 0 invented entities

The paper adds an empirical training recipe rather than a formal derivation. Its free parameters are mostly hand-set hyperparameters and dataset curation choices; the Target split (37/13) and the cross-dataset class mapping most directly shape the reported numbers. The load-bearing axioms are borrowed capabilities (FRNet, PPT) and the unstated assumption that a 13-scan validation set is representative and that class semantics transfer across datasets; the paper itself admits the outlier class is inconsistently supervised.

free parameters (6)
  • Histogram normalization evaluation cutoffs = lower 2%, upper 95%
    Chosen by hand for evaluation; during training the cutoffs are randomly sampled (0-5% lower, 92-97% upper). These cutoffs determine the ambient and intensity features fed to the classifier and affect all reported metrics.
  • Train/validation split of Target dataset = 37 training / 13 validation scans
    Chosen by hand. The validation set is small (13 scans) and is used for early stopping and configuration selection; the people class contains only 14 labeled points in validation, making mIoU unstable for that class.
  • Class mapping from SemanticKITTI (34 classes) and Waymo (23 classes) to 8 target classes = per-class mapping not listed
    The mapping is designed by the authors and is a free choice. The outlier class is not labeled in Waymo and is labeled more aggressively in SemanticKITTI, so the mapping injects inconsistent supervision across datasets.
  • Manual annotation of 50 Target scans = 8-class labels
    Labels were produced by the authors; annotation ambiguity is acknowledged for road and ground boundaries. The dataset is not released, so the labels are not independently checkable.
  • Manifold Mixup Beta distribution shape parameter = not specified
    The mixing coefficient is sampled from a Beta distribution but the shape parameter is never stated. Relevant to the MM ablation, though MM is removed from the final model.
  • Pretraining and fine-tuning schedules = 100k steps, lr 0.002; fine-tune 7.6k steps, lr 0.001
    Hand-chosen training schedules and learning rates. These affect the reported 51.8% mIoU but are not fitted constants in a derivation sense.
assumptions (6)
  • domain assumption FRNet frustum-range network is an adequate feature extractor for LiDAR semantic segmentation
    The paper relies on FRNet as the backbone and does not compare with other backbones. If FRNet features are poor for rural scenes, the pipeline results would change.
  • domain assumption Point Prompt Training prompt-normalization improves multi-dataset training for projection-based CNNs
    Adopted from transformer-based work; the paper tests it and finds a mild positive effect with fine-tuning, but the mechanism is taken from prior work.
  • domain assumption Label correspondences across SemanticKITTI, Waymo, and Target are semantically consistent for the 8 classes
    All datasets are mapped to 8 classes. The inconsistent outlier labeling is admitted in Section IV.E, so cross-dataset supervision is imperfect.
  • domain assumption The 13 validation scans are representative of the target domain
    Scans are selectively sampled from long sequences; people appear only as 14 validation points, and no held-out test set exists, so the reported mIoU may not generalize.
  • ad hoc to paper Injecting ambient values only at the MLP head, with zeros for datasets lacking ambients, is a valid fusion strategy
    This design choice is specific to this paper. It means the feature extractor never sees ambient data, and the classifier must learn to ignore zero-valued ambient features from other datasets.
  • standard math Standard AdamW and OneCycleLR optimization assumptions
    Relies on standard optimizers and learning-rate scheduling; no formal convergence guarantees are needed, which is common practice in deep learning.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Data-Efficient Point Cloud Semantic Segmentation Pipeline for Unimproved Roads." pith.science (2026). https://pith.science/paper/OZGHFWT4

@misc{pith2026250820135,
  author       = {Pith},
  title        = {Pith review of: Data-Efficient Point Cloud Semantic Segmentation Pipeline for Unimproved Roads},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/OZGHFWT4}},
  note         = {Machine review of arXiv:2508.20135}
}
read the original abstract

In this case study, we present a data-efficient point cloud segmentation pipeline and training framework for robust segmentation of unimproved roads and seven other classes. Our method employs a two-stage training framework: first, a projection-based convolutional neural network is pre-trained on a mixture of public urban datasets and a small, curated in-domain dataset; then, a lightweight prediction head is fine-tuned exclusively on in-domain data. Along the way, we explore the application of Point Prompt Training to batch normalization layers and the effects of Manifold Mixup as a regularizer within our pipeline. We also explore the effects of incorporating histogram-normalized ambients to further boost performance. Using only 50 labeled point clouds from our target domain, we show that our proposed training approach improves mean Intersection-over-Union from 33.5% to 51.8% and the overall accuracy from 85.5% to 90.8%, when compared to naive training on the in-domain data. Crucially, our results demonstrate that pre-training across multiple datasets is key to improving generalization and enabling robust segmentation under limited in-domain supervision. Overall, this study demonstrates a practical framework for robust 3D semantic segmentation in challenging, low-data scenarios. Our code is available at: https://github.com/andrewyarovoi/MD-FRNet.

Figures

Figures reproduced from arXiv: 2508.20135 by the authors.

Figure 1
Figure 1. An overview of the proposed training methodology for data-efficient training of a semantic segmentation pipeline. Note that the pre-training phase is trained on a mixture of multiple datasets, while the fine-tuning phase only trains on our 37 Target training samples. Image Credit: Manifold Mixup from [18]. and ambient dropout, randomized histogram equalization, and minor road/ground-specific rotations. We applied in… view at source ↗
Figure 3
Figure 3. Diagram showing how a normalization layer (e.g. Batch Norm) can be adapted into a prompt-normalization layer. The context embedding is shared across prompt-normalization layers, specific to each dataset, and is learned for each dataset. E. Handling Ambient Values During the manual annotation process of the Target dataset, it was noted that the ambient return values, captured by the OS-1 LiDAR, were particularly effe… view at source ↗
Figure 4
Figure 4. An example of the predicted labels for a diverse scene within the Target validation set. The colors correspond to class labels as follows: red is road, orange is ground, green is vegetation, yellow is structure, cyan is vehicle, magenta is object, and white is outlier. This scene does not contain any points labeled as people. point clouds with limited annotated in-domain data. Through experiments, we demonstrated th… view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

23 extracted references · 21 canonical work pages

  1. [1]

    Scalability in Percep- tion for Autonomous Driving: Waymo Open Dataset,

    P. Sun, H. Kretzschmar, X. Dotiwalla, A. Chouard, V . Patnaik, P. Tsui, J. Guo, Y . Zhou, Y . Chai, B. Caine, V . Vasudevan, W. Han, J. Ngiam, H. Zhao, A. Timofeev, S. Ettinger, M. Krivokon, A. Gao, A. Joshi, Y . Zhang, J. Shlens, Z. Chen, and D. Anguelov, “Scalability in Percep- tion for Autonomous Driving: Waymo Open Dataset,” in Proceedings of the IEEE...

    work page 2020

  2. [2]

    SemanticKITTI: A Dataset for Semantic Scene Under- standing of LiDAR Sequences,

    J. Behley, M. Garbade, A. Milioto, J. Quenzel, S. Behnke, C. Stachniss, and J. Gall, “SemanticKITTI: A Dataset for Semantic Scene Under- standing of LiDAR Sequences,” in Proc. of the IEEE/CVF International Conf.˜on Computer Vision (ICCV) , 2019

    work page 2019

  3. [3]

    A Closer Look at Few-shot Classification,

    W.-Y . Chen, Y .-C. Liu, Z. Kira, Y .-C. Wang, and J.-B. Huang, “A Closer Look at Few-shot Classification,” in International Conference on Learning Representations , 2019

    work page 2019

  4. [4]

    Prototypical Networks for Few- shot Learning,

    J. Snell, K. Swersky, and R. Zemel, “Prototypical Networks for Few- shot Learning,” Advances in Neural Information Processing Systems , vol. 2017-December, pp. 4078–4088, 3 2017

    work page 2017

  5. [5]

    SimpleShot: Revisiting Nearest-Neighbor Classification for Few-Shot Learning,

    Y . Wang, W.-L. Chao, K. Q. Weinberger, and L. van der Maaten, “SimpleShot: Revisiting Nearest-Neighbor Classification for Few-Shot Learning,” arXiv preprint arXiv:1911.04623 , 11 2019

    arXiv 1911

  6. [6]

    Self- Supervised Learning For Few-Shot Image Classification,

    D. Chen, Y . Chen, Y . Li, F. Mao, Y . Hey, and H. Xue, “Self- Supervised Learning For Few-Shot Image Classification,” ICASSP , IEEE International Conference on Acoustics, Speech and Signal Processing - Proceedings, vol. 2021-June, pp. 1745–1749, 11 2019

    work page 2021

  7. [7]

    Pushing the Limits of Simple Pipelines for Few-Shot Learning: External Data and Fine-Tuning Make a Difference,

    S. X. Hu, D. Li, J. Stuhmer, M. Kim, and T. M. Hospedales, “Pushing the Limits of Simple Pipelines for Few-Shot Learning: External Data and Fine-Tuning Make a Difference,” Proceedings of the IEEE Com- puter Society Conference on Computer Vision and Pattern Recognition , vol. 2022-June, pp. 9058–9067, 2022

    work page 2022

  8. [8]

    Re- thinking Few-Shot Image Classification: a Good Embedding Is All You Need?,

    Y . Tian, Y . Wang, D. Krishnan, J. B. Tenenbaum, and P. Isola, “Re- thinking Few-Shot Image Classification: a Good Embedding Is All You Need?,” Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) , vol. 12359 LNCS, pp. 266–282, 3 2020

    work page 2020

Show all 23 references
  1. [9]

    PANet: Few-shot image semantic segmentation with prototype alignment,

    K. Wang, J. H. Liew, Y . Zou, D. Zhou, and J. Feng, “PANet: Few-shot image semantic segmentation with prototype alignment,” Proceedings of the IEEE International Conference on Computer Vision , vol. 2019- October, pp. 9196–9205, 10 2019

    2019

  2. [10]

    GPr-Net: Geometric Prototypical Network for Point Cloud Few-Shot Learning,

    T. Anvekar and D. Bazazian, “GPr-Net: Geometric Prototypical Network for Point Cloud Few-Shot Learning,” IEEE Computer Society Confer- ence on Computer Vision and Pattern Recognition Workshops, vol. 2023- June, pp. 4179–4188, 2023

    2023

  3. [11]

    3d shapenets: A deep representation for volumetric shapes,

    Z. Wu, S. Song, A. Khosla, F. Yu, L. Zhang, X. Tang, and J. Xiao, “3d shapenets: A deep representation for volumetric shapes,” in Proceedings of the IEEE conference on computer vision and pattern recognition , pp. 1912–1920, 2015

    1912

  4. [12]

    A Closer Look at Few-Shot 3D Point Cloud Classification,

    C. Ye, H. Zhu, B. Zhang, and T. Chen, “A Closer Look at Few-Shot 3D Point Cloud Classification,” International Journal of Computer Vision , vol. 131, pp. 772–795, 3 2023

    2023

  5. [13]

    Unified 3D Segmenter As Prototypical Classifiers,

    Z. Qin, C. Han, Q. Wang, X. Nie, Y . Yin, and L. Xiankai, “Unified 3D Segmenter As Prototypical Classifiers,” in Advances in Neural Information Processing Systems (A. Oh, T. Naumann, A. Globerson, K. Saenko, M. Hardt, and S. Levine, eds.), vol. 36, pp. 46419–46432, Curran Asso...

    2023

  6. [14]

    ProtoSeg: A Prototype-Based Point Cloud Instance Segmentation Method,

    R. Royen, L. Denis, and A. Munteanu, “ProtoSeg: A Prototype-Based Point Cloud Instance Segmentation Method,” 2024

    2024

  7. [15]

    Few-shot 3D Point Cloud Semantic Segmentation,

    N. Zhao, T. S. Chua, and G. H. Lee, “Few-shot 3D Point Cloud Semantic Segmentation,” Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition , pp. 8869–8878, 2021

    2021

  8. [16]

    Towards Large-Scale 3D Representation Learning with Multi-Dataset Point Prompt Training,

    X. Wu, Z. Tian, X. Wen, B. Peng, X. Liu, K. Yu, and H. Zhao, “Towards Large-Scale 3D Representation Learning with Multi-Dataset Point Prompt Training,” in 2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , pp. 19551–19562, IEEE, 6 2024

    2024

  9. [17]

    Point Transformer V3: Simpler, Faster, Stronger,

    X. Wu, L. Jiang, P.-S. Wang, Z. Liu, X. Liu, Y . Qiao, W. Ouyang, T. He, and H. Zhao, “Point Transformer V3: Simpler, Faster, Stronger,” in 2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp. 4840–4851, IEEE, 6 2024

    2024

  10. [18]

    Manifold Mixup: Better Representations by Interpolat- ing Hidden States,

    V . Verma, A. Lamb, C. Beckham, A. Najafi, I. Mitliagkas, D. Lopez-Paz, and Y . Bengio, “Manifold Mixup: Better Representations by Interpolat- ing Hidden States,” in Proceedings of the 36th International Conference on Machine Learning (K. Chaudhuri and R. Salakhutdinov, eds.),...

    2019

  11. [19]

    Charting the Right Manifold: Manifold Mixup for Few-shot Learning,

    P. Mangla, M. Singh, A. Sinha, N. Kumari, V . N. Balasubramanian, and B. Krishnamurthy, “Charting the Right Manifold: Manifold Mixup for Few-shot Learning,” Proceedings - 2020 IEEE Winter Conference on Applications of Computer Vision, WACV 2020 , pp. 2207–2216, 7 2019

    2020

  12. [20]

    PointMixup: Augmentation for Point Clouds,

    Y . Chen, V . T. Hu, E. Gavves, T. Mensink, P. Mettes, P. Yang, and C. G. Snoek, “PointMixup: Augmentation for Point Clouds,” Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) , vol. 12348 LNCS,...

    2020

  13. [21]

    SimpliMix: A Sim- plified Manifold Mixup for Few-shot Point Cloud Classification,

    M. Yang, W. Chai, J. Wang, and S. Velipasalar, “SimpliMix: A Sim- plified Manifold Mixup for Few-shot Point Cloud Classification,” Pro- ceedings - 2024 IEEE Winter Conference on Applications of Computer Vision, WACV 2024, pp. 3656–3665, 1 2024

    2024

  14. [22]

    FRNet: Frustum-Range Networks for Scalable LiDAR Segmentation,

    X. Xu, L. Kong, H. Shuai, and Q. Liu, “FRNet: Frustum-Range Networks for Scalable LiDAR Segmentation,” IEEE Transactions on Image Processing, 2025

    2025

  15. [23]

    PointNeXt: Revisiting PointNet++ with Improved Training and Scaling Strategies,

    G. Qian, Y . Li, H. Peng, J. Mai, H. Abed Al Kader Hammoud, M. Elhoseiny, and B. Ghanem, “PointNeXt: Revisiting PointNet++ with Improved Training and Scaling Strategies,” in Advances in Neural Information Processing Systems (NeurIPS) , 2022

    2022

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.