Pith. sign in

REVIEW 3 major objections 5 minor 51 references

SAM-guided Pseudo Label Enhancement for Multi-modal 3D Semantic Segmentation

T0 review · 3 major / 5 minor · reviewed 2026-08-09 · deepseek-v4-flash

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

Pith's one-line read This paper proposes that SAM-mask grouping and geometry-aware propagation turn sparse pseudo-labels into dense, accurate training signals for multi-modal 3D semantic segmentation under domain shift, improving cross-domain performance over…

desk verdict Useful SAM-guided pseudo-label densification for multi-modal 3D UDA, but the central 'high-quality' claim is contradicted by the paper's own Table II. read the letter →

arxiv 2502.00960 v1 pith:GITBSEI4 submitted 2025-02-02 cs.CV

Mingyu Yang , Jitong Lu , Hun-Seok Kim This is my paper · ORCID
classification cs.CV
keywords 3Dsemanticsegmentationdomainadaptationpseudo-labelsSegmentAnythingModelself-trainingmulti-modalfusionpointcloudsource-free
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 tries to establish that the notoriously sparse pseudo-labels produced by strict reliability filters in multi-modal 3D semantic segmentation can be made much denser, without sacrificing accuracy, by leveraging the 2D object-grouping prior of the Segment Anything Model (SAM). The concrete claim is this: when points whose image projections fall inside one SAM mask are labeled by majority vote—after the mask passes size, purity, and representativity filters—and the mask label is then propagated only to geometrically connected 3D neighbors, the number of correct pseudo-labels grows substantially and cross-domain mIoU rises consistently. The authors demonstrate this on three domain-shift benchmarks (USA-to-Singapore, A2D2-to-SemanticKITTI, Singapore-to-SemanticKITTI) under both unsupervised and source-free adaptation, using xMUDA and SUMMIT as baselines. A sympathetic reader cares because dense, trustworthy pseudo-labels are the bottleneck for adaptation methods that must retrain on target data with no ground truth; if the claim holds, a lightweight image-segmentation prior can be bolted onto existing self-training loops to push adaptation performance higher.

What carries the argument

The load-bearing mechanism is the SAM mask as an object-grouping prior, coupled with two operations: Mask Label Assignment (MLA), which takes the majority class of the existing pseudo-labels inside a mask and rejects the mask unless its relative area is below a threshold, its label purity is high, and its dominant class covers enough points; and Geometry-Aware Progressive Propagation (GAPP), which spreads a valid mask label outward in rounds, adding only points whose distance to the current seed set is within a scaled nearest-neighbor bound. This two-step design is what lets the method fill in unlabeled points while leaving out occluded background points that appear over an object due to lidar-camera misalignment.

What would settle it

Replace the SAM masks with random 2D superpixels of matched area distribution and rerun the A2D2-to-SemanticKITTI experiment; if the mIoU gain over xMUDA-pl or SUMMIT persists, then the object-grouping prior is not the active ingredient. Alternatively, artificially bias the seed pseudo-labels inside a car mask toward 'truck' and verify whether the purity and representativity constraints reject the mask or accept the wrong label.

Watch

Extended reading notes

Core claim

The central discovery is that SAM's class-agnostic masks, which are generated purely from 2D images, carry a grouping signal that transfers to 3D point clouds through the projection function: points sharing a mask are likely part of the same semantic object even under domain shift. The paper shows that labeling a mask by the majority class of its existing reliable pseudo-labels, then propagating that label only to points within a growing 3D exploration distance, produces a denser pseudo-label set with comparable order-level accuracy, yielding consistent mIoU improvements in every tested adaptation setting. The paper's own evidence is the accuracy and count statistics (Table II) and the ablations (Table III), which isolate the contribution of the masks from brute-force direct propagation.

Load-bearing premise

The load-bearing premise is that the majority class among the already-pruned pseudo-labels inside a SAM mask is the true class of that mask's object; if the domain shift biases those seed labels toward a wrong class, the mask receives a wrong label and none of the three tuned filtering thresholds can catch it.

Editorial extensions

If this is right

  • Self-training loops for 3D semantic segmentation can be upgraded by bolting a 2D mask-grouping step on top of any existing pseudo-label generator, without retraining the backbone.
  • Source-free domain adaptation, where no source data is available, benefits from the enhancement just as much as standard UDA, since the method relies only on target-side images, point clouds, and SAM.
  • Because GAPP propagates only along 3D proximity, the method specifically reduces errors caused by lidar-camera occlusion, a known failure mode of direct projection-based fusion.
  • The reported pseudo-label counts more than double for the SFDA baseline, implying that density of training signal, not just precision, is a major factor in adaptation performance.

Reading between the lines

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

  • Editorial inference: the same mask-enhancement recipe should transfer to other dense 2D segmenters with object-aware masks (e.g., in-the-wild segmenters or even supervised panoptic models), since the paper's mechanism depends on object coherence rather than SAM specifically.
  • Editorial inference: the three thresholds (λs=0.2, λp=0.8, λr=0.1) and β=2 are fixed across all experiments; a natural stress test is whether they transfer to setups with different object scales (e.g., drones or indoor scenes) or to larger calibration errors, where the purity threshold may need to be tightened.
  • Editorial inference: since the method enhances pseudo-labels before each self-training iteration, its benefits might compound; the paper evaluates only the final model, but an iterative version could be tested by re-running the enhancement on updated pseudo-labels.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The paper proposes a pseudo-label enhancement method for multi-modal 3D semantic segmentation under domain shift. Starting from pruned pseudo-labels generated by xMUDA or SUMMIT, the method projects SAM masks from paired images onto the LiDAR point cloud, assigns each mask a class label by majority voting over the valid pseudo-labels inside it, filters masks by size, purity, and representativity constraints, and then propagates the mask label to nearby unlabeled points using a Geometry-Aware Progressive Propagation (GAPP) procedure. Experiments are reported for USA-to-Singapore, A2D2-to-SemanticKITTI, and Singapore-to-SemanticKITTI, in both unsupervised and source-free domain adaptation settings, with mIoU gains over the xMUDA-pl and SUMMIT baselines and an ablation study reporting pseudo-label statistics.

Significance. If the reported results are reproducible, the paper addresses a real bottleneck in self-training-based domain adaptation: the sparsity of pruned pseudo-labels. The method is simple and clearly specified in Algorithm 1 and Equations (2)-(13), the hyperparameters are held fixed across all experiments, and GAPP is a sensible mechanism for reducing the impact of lidar-camera misalignment during label propagation. The evaluation covers three benchmarks and two adaptation tasks, and the paper ships both mIoU results and pseudo-label statistics rather than only final accuracies. The main weakness is that the reported pseudo-label statistics do not support the paper's headline claim that the method increases the quantity of 'high-quality' pseudo-labels, and the absence of repeated-run statistics tempers the strength of the empirical claims.

major comments (3)
  1. [§IV-C, Table II] The claim that the method 'significantly increases the quantity of high-quality pseudo-labels' is not supported by Table II as presented. In every configuration, the average pseudo-label accuracy over the four listed classes decreases after the full enhancement (xMUDA-2D: 56.83% to 53.09%; xMUDA-3D: 60.60% to 55.95%; SUMMIT: 82.41% to 76.20%). The 'Total Inc.' column appears to report the percentage increase in the number of correct pseudo-labels, but the caption never defines it, and a count of correct labels cannot by itself establish quality. The text in §IV-C even states that the final accuracy is 'comparable' to the original, which the table contradicts. Please define the statistic, report the accuracy of the newly added labels separately from the accuracy of the union, and revise the 'high-quality' claim or provide a formal definition of quality.
  2. [§IV-C, Tables I and III] All mIoU numbers appear to come from single training runs. Some gains are small (e.g., SUMMIT+ours on USA-to-Singapore 2D+3D is +0.63 mIoU), and no error bars, standard deviations, or significance tests are provided. Given that the adaptation procedures involve stochastic optimization and iterative pseudo-label updates, a few runs with mean and standard deviation, or at least a statement that the reported numbers are medians over seeds, are needed to establish that the improvements are not run-to-run noise.
  3. [§III-C, Eqs. (4)-(8)] The mask-label assignment assumes that the dominant class among the already-pruned pseudo-labels inside a SAM mask is the true class of the object covered by the mask. The filters in Eqs. (5)-(7) check mask size, label purity, and representativity, but they cannot detect a confidently wrong dominant class induced by domain shift. Because the full pipeline measurably lowers pseudo-label accuracy in Table II, the mechanism by which the method helps is not shown to be 'more reliable pseudo-labels'; it may instead be denser, noisier supervision. An analysis of mask-level label precision (e.g., the fraction of masks whose assigned label matches ground truth) would directly test this assumption and would make the central claim credible.
minor comments (5)
  1. [§III-B] The masks are said to be sorted in descending order, yet the text says the iteration begins with the smallest mask M1. Please make the ordering consistent with Fig. 2 and the stated motivation that smaller masks are processed first.
  2. [Table II caption] Please define 'Total Inc.' and state whether the reported average accuracy is computed over all classes or only the four listed classes.
  3. [Algorithm 1] The title contains a typo ('PLE NHANCEMENT'), and line 8 uses 'Rsize' while Eq. (5) defines the quantity as 'Rs'.
  4. [§IV-C, first paragraph] The text states that the mIoU for 2D+3D improves by 0.87% in the USA-to-Singapore scenario, but Table I shows +1.44 for xMUDA+ours and +0.63 for SUMMIT+ours; please identify the baseline used for the reported relative improvement.
  5. [§IV-B] The hyperparameters lambda_s, lambda_p, lambda_r, and beta are fixed across all experiments; a short sensitivity study varying one at a time would increase confidence that the results are not tied to a specific manual tuning.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: the method's outputs are evaluated against ground-truth mIoU and pseudo-label accuracy, not against its own inputs.

full rationale

The paper makes an empirical claim that SAM-mask-guided enhancement of sparse pseudo-labels improves multi-modal 3D semantic segmentation under domain shift. The generation chain (Equations 2-13) is a heuristic label-propagation procedure: mask membership via projection (Eq. 2), majority-vote label (Eq. 4), filtering thresholds (Eqs. 5-7), and progressive geometric propagation (Eqs. 10-13). None of these equations is defined in terms of the final reported quantity. The final mIoU values in Table I are computed against ground-truth annotations, and the pseudo-label accuracy statistics in Table II are likewise computed against ground truth. Thus the central claim does not reduce by construction to its inputs. The hyperparameters lambda_s, lambda_p, lambda_r, and beta are fixed constants stated in Section IV-B, not fitted to target-domain labels or to the reported metric, so there is no fitted-input-called-prediction loop. The only self-citation (reference [37], Yang et al., ECCV 2022) appears in a catalog of multi-modal methods in Related Work and is not load-bearing for the proposed algorithm. The fact that Table II shows lower average pseudo-label accuracy after enhancement is a substantive correctness concern about the 'high-quality' claim, but it is not circularity: the accuracy is measured against ground truth. No uniqueness theorem or ansatz is smuggled in via self-citation. Verdict: no significant circularity; score 1 reflects only the mild, non-load-bearing tuning freedom in the fixed thresholds.

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

No new physical entities or model components beyond the algorithmic modules MLA and GAPP; the method reuses SAM, a pretrained external model, and the baselines' network architectures.

free parameters (4)
  • lambda_s (mask size threshold) = 0.2
    Ratio threshold in Equation 5; set manually and stated in Section IV-B. No sensitivity analysis is reported.
  • lambda_p (purity threshold) = 0.8
    Minimum purity in Equation 6; set manually.
  • lambda_r (representativity threshold) = 0.1
    Minimum dominant-class coverage in Equation 7; set manually.
  • beta (exploration distance scaling) = 2
    Scales the exploration distance in Equation 12; set manually.
assumptions (4)
  • domain assumption LiDAR-camera projection proj() is calibrated and paired images are temporally aligned for all target frames.
    Invoked in Equation 2 to map 3D points into SAM masks; the paper cites 2D-3D misalignment as a problem but assumes the projection function is available and valid.
  • domain assumption SAM masks are semantically coherent: each mask mostly covers one object or class.
    Invoked in Section III-B: smaller masks are more likely to correspond to individual objects sharing a common label; the whole method relies on this grouping property.
  • domain assumption Initial pseudo-labels are accurate enough that majority voting within a mask identifies the true class.
    Invoked in Section III-C, Equations 3 and 4; if the dominant class among confident pseudo-labels is systematically wrong, MLA produces wrong mask labels.
  • domain assumption Object points form a single spatially connected component separated from outliers by a distance gap large enough for the beta-scaled exploration distance.
    Invoked in Section III-D and Equations 11 and 12; GAPP's distance-based region growing assumes background outliers are not connected to the object's points.

how reviews work

0 comments
Cite this review

Pith. "Pith review of SAM-guided Pseudo Label Enhancement for Multi-modal 3D Semantic Segmentation." pith.science (2026). https://pith.science/paper/GITBSEI4

@misc{pith2026250200960,
  author       = {Pith},
  title        = {Pith review of: SAM-guided Pseudo Label Enhancement for Multi-modal 3D Semantic Segmentation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/GITBSEI4}},
  note         = {Machine review of arXiv:2502.00960}
}
read the original abstract

Multi-modal 3D semantic segmentation is vital for applications such as autonomous driving and virtual reality (VR). To effectively deploy these models in real-world scenarios, it is essential to employ cross-domain adaptation techniques that bridge the gap between training data and real-world data. Recently, self-training with pseudo-labels has emerged as a predominant method for cross-domain adaptation in multi-modal 3D semantic segmentation. However, generating reliable pseudo-labels necessitates stringent constraints, which often result in sparse pseudo-labels after pruning. This sparsity can potentially hinder performance improvement during the adaptation process. We propose an image-guided pseudo-label enhancement approach that leverages the complementary 2D prior knowledge from the Segment Anything Model (SAM) to introduce more reliable pseudo-labels, thereby boosting domain adaptation performance. Specifically, given a 3D point cloud and the SAM masks from its paired image data, we collect all 3D points covered by each SAM mask that potentially belong to the same object. Then our method refines the pseudo-labels within each SAM mask in two steps. First, we determine the class label for each mask using majority voting and employ various constraints to filter out unreliable mask labels. Next, we introduce Geometry-Aware Progressive Propagation (GAPP) which propagates the mask label to all 3D points within the SAM mask while avoiding outliers caused by 2D-3D misalignment. Experiments conducted across multiple datasets and domain adaptation scenarios demonstrate that our proposed method significantly increases the quantity of high-quality pseudo-labels and enhances the adaptation performance over baseline methods.

Figures

Figures reproduced from arXiv: 2502.00960 by the authors.

Figure 1
Figure 1. Overview of the proposed method. The initial pseudo-labels tend to be accurate but sparse. Our proposed method aims to enhance the pseudo￾labels by utilizing the grouping information of SAM masks. significant area of research for both 2D and 3D semantic segmentation [10], [11], [12], [13], [14]. Recently, such focus has extended to multi-modal 3D semantic segmentation and has shown impressive performance across vari… view at source ↗
Figure 2
Figure 2. Visual explanation of the proposed method. [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Visualization of the proposed GAPP. Note the green points on the Ground Truth images (rightmost) belong to the background, not the car. The lidar paths of these green points are occluded in the camera image, thus they ‘appear’ to be on the car due to the sensor location mismatch between the lidar and camera. Unlike Direct Propagation (DP), GAPP successfully avoids those incorrect mask labels assigned to the car. con… view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Visualization of the SAM masks and enhanced pseudo-labels. The proposed method significantly improves the pseudo-label quality. scenarios (2) and (3). The Adam optimizer [51] is used with a batch size of 8 for all experiments. All training and testing are conducted usi…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

51 extracted references · 41 canonical work pages

  1. [1]

    Squeezeseg: Convolutional neural nets with recurrent crf for real-time road-object segmentation from 3d lidar point cloud,

    B. Wu, A. Wan, X. Yue, and K. Keutzer, “Squeezeseg: Convolutional neural nets with recurrent crf for real-time road-object segmentation from 3d lidar point cloud,” in 2018 IEEE international conference on robotics and automation (ICRA) . IEEE, 2018, pp. 1887–1893. 1

    work page 2018

  2. [2]

    Rpvnet: A deep and efficient range-point-voxel fusion network for lidar point cloud segmentation,

    J. Xu, R. Zhang, J. Dou, Y . Zhu, J. Sun, and S. Pu, “Rpvnet: A deep and efficient range-point-voxel fusion network for lidar point cloud segmentation,” in Proceedings of the IEEE/CVF international conference on computer vision , 2021, pp. 16 024–16 033. 1

    work page 2021

  3. [3]

    4d spatio-temporal convnets: Minkowski convolutional neural networks,

    C. Choy, J. Gwak, and S. Savarese, “4d spatio-temporal convnets: Minkowski convolutional neural networks,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , 2019, pp. 3075–3084. 1

    work page 2019

  4. [4]

    Kpconv: Flexible and deformable convolution for point clouds,

    H. Thomas, C. R. Qi, J.-E. Deschaud, B. Marcotegui, F. Goulette, and L. J. Guibas, “Kpconv: Flexible and deformable convolution for point clouds,” in Proceedings of the IEEE/CVF international conference on computer vision, 2019, pp. 6411–6420. 1

    work page 2019

  5. [5]

    Pointnet++: Deep hierarchical feature learning on point sets in a metric space,

    C. R. Qi, L. Yi, H. Su, and L. J. Guibas, “Pointnet++: Deep hierarchical feature learning on point sets in a metric space,” Advances in neural information processing systems , vol. 30, 2017. 1

    work page 2017

  6. [6]

    nuscenes: A multimodal dataset for autonomous driving,

    H. Caesar, V . Bankiti, A. H. Lang, S. V ora, V . E. Liong, Q. Xu, A. Krishnan, Y . Pan, G. Baldan, and O. Beijbom, “nuscenes: A multimodal dataset for autonomous driving,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , 2020, pp. 11 621–11 631. 1, 2, 5

    work page 2020

  7. [7]

    Semantickitti: A dataset for semantic scene un- derstanding of lidar sequences,

    J. Behley, M. Garbade, A. Milioto, J. Quenzel, S. Behnke, C. Stach- niss, and J. Gall, “Semantickitti: A dataset for semantic scene un- derstanding of lidar sequences,” in Proceedings of the IEEE/CVF international conference on computer vision , 2019, pp. 9297–9307. 1, 2, 5

    work page 2019

  8. [8]

    Splatnet: Sparse lattice networks for point cloud processing,

    H. Su, V . Jampani, D. Sun, S. Maji, E. Kalogerakis, M.-H. Yang, and J. Kautz, “Splatnet: Sparse lattice networks for point cloud processing,” in Proceedings of the IEEE conference on computer vision and pattern recognition , 2018, pp. 2530–2539. 1

    work page 2018

Show all 51 references
  1. [9]

    Sensor fusion for joint 3d object detection and semantic segmentation,

    G. P. Meyer, J. Charland, D. Hegde, A. Laddha, and C. Vallespi- Gonzalez, “Sensor fusion for joint 3d object detection and semantic segmentation,” in Proceedings of the IEEE/CVF conference on com- puter vision and pattern recognition workshops , 2019, pp. 0–0. 1, 2

    2019

  2. [10]

    Cycada: Cycle-consistent adversarial domain adaptation,

    J. Hoffman, E. Tzeng, T. Park, J.-Y . Zhu, P. Isola, K. Saenko, A. Efros, and T. Darrell, “Cycada: Cycle-consistent adversarial domain adaptation,” in International conference on machine learning . Pmlr, 2018, pp. 1989–1998. 1, 2

    2018

  3. [11]

    Bidirectional learning for domain adaptation of semantic segmentation,

    Y . Li, L. Yuan, and N. Vasconcelos, “Bidirectional learning for domain adaptation of semantic segmentation,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition , 2019, pp. 6936–6945. 1

    2019

  4. [12]

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

    T.-H. Vu, H. Jain, M. Bucher, M. Cord, and P. P ´erez, “Advent: Adversarial entropy minimization for domain adaptation in semantic segmentation,” in Proceedings of the IEEE/CVF conference on com- puter vision and pattern recognition , 2019, pp. 2517–2526. 1, 2

    2019

  5. [13]

    Squeezesegv2: Im- proved model structure and unsupervised domain adaptation for road- object segmentation from a lidar point cloud,

    B. Wu, X. Zhou, S. Zhao, X. Yue, and K. Keutzer, “Squeezesegv2: Im- proved model structure and unsupervised domain adaptation for road- object segmentation from a lidar point cloud,” in 2019 international conference on robotics and automation (ICRA) . IEEE, 2019, pp. 4376–4382. 1, 2

    2019

  6. [14]

    Complete & label: A domain adaptation approach to semantic segmentation of lidar point clouds,

    L. Yi, B. Gong, and T. Funkhouser, “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, 2021, pp. 15 363–15 373. 1, 2

    2021

  7. [15]

    xmuda: Cross-modal unsupervised domain adaptation for 3d semantic seg- mentation,

    M. Jaritz, T.-H. Vu, R. d. Charette, E. Wirbel, and P. P ´erez, “xmuda: Cross-modal unsupervised domain adaptation for 3d semantic seg- mentation,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , 2020, pp. 12 605–12 614. 1, 2, 3, 5, 6

    2020

  8. [16]

    Mopa: Multi- modal prior aided domain adaptation for 3d semantic segmentation,

    H. Cao, Y . Xu, J. Yang, P. Yin, S. Yuan, and L. Xie, “Mopa: Multi- modal prior aided domain adaptation for 3d semantic segmentation,” in 2024 IEEE International Conference on Robotics and Automation (ICRA). IEEE, 2024, pp. 9463–9470. 1, 2

    2024

  9. [17]

    Mm-tta: multi-modal test-time adaptation for 3d semantic segmentation,

    I. Shin, Y .-H. Tsai, B. Zhuang, S. Schulter, B. Liu, S. Garg, I. S. Kweon, and K.-J. Yoon, “Mm-tta: multi-modal test-time adaptation for 3d semantic segmentation,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , 2022, pp. 16 928–16 937. 1, 2, 3

    2022

  10. [18]

    Summit: Source-free adaptation of uni-modal models to multi-modal targets,

    C. Simons, D. S. Raychaudhuri, S. M. Ahmed, S. You, K. Karydis, and A. K. Roy-Chowdhury, “Summit: Source-free adaptation of uni-modal models to multi-modal targets,” in Proceedings of the IEEE/CVF International Conference on Computer Vision , 2023, pp. 1239–1249. 1, 2, 3, 5

    2023

  11. [19]

    Segment anything,

    A. Kirillov, E. Mintun, N. Ravi, H. Mao, C. Rolland, L. Gustafson, T. Xiao, S. Whitehead, A. C. Berg, W.-Y . Lo, et al. , “Segment anything,” in Proceedings of the IEEE/CVF International Conference on Computer Vision , 2023, pp. 4015–4026. 1

    2023

  12. [20]

    Sam 2: Segment anything in images and videos,

    N. Ravi, V . Gabeur, Y .-T. Hu, R. Hu, C. Ryali, T. Ma, H. Khedr, R. R¨adle, C. Rolland, L. Gustafson, et al., “Sam 2: Segment anything in images and videos,” arXiv preprint arXiv:2408.00714 , 2024. 1, 2

    2024 arXiv

  13. [21]

    Push the boundary of sam: A pseudo-label correction framework for medical segmentation,

    Z. Huang, H. Liu, H. Zhang, X. Li, H. Liu, F. Xing, A. Laine, E. Angelini, C. Hendon, and Y . Gan, “Push the boundary of sam: A pseudo-label correction framework for medical segmentation,” arXiv preprint arXiv:2308.00883, 2023. 1, 2

    2023 arXiv

  14. [22]

    Segment anything in medical images,

    J. Ma, Y . He, F. Li, L. Han, C. You, and B. Wang, “Segment anything in medical images,” Nature Communications, vol. 15, no. 1, p. 654,

  15. [23]

    Segment anything model for medical images?

    Y . Huang, X. Yang, L. Liu, H. Zhou, A. Chang, X. Zhou, R. Chen, J. Yu, J. Chen, C. Chen, et al., “Segment anything model for medical images?” Medical Image Analysis , vol. 92, p. 103061, 2024. 1, 2

    2024

  16. [24]

    Segment anything model (sam) enhanced pseudo labels for weakly supervised semantic seg- mentation,

    T. Chen, Z. Mai, R. Li, and W.-l. Chao, “Segment anything model (sam) enhanced pseudo labels for weakly supervised semantic seg- mentation,” arXiv preprint arXiv:2305.05803 , 2023. 1, 2

    2023 arXiv

  17. [25]

    Sam-guided unsupervised domain adaptation for 3d segmentation,

    X. Peng, R. Chen, F. Qiao, L. Kong, Y . Liu, T. Wang, X. Zhu, and Y . Ma, “Sam-guided unsupervised domain adaptation for 3d segmentation,” arXiv preprint arXiv:2310.08820 , 2023. 1, 2

    2023 arXiv

  18. [26]

    Segment any point cloud sequences by distilling vision foun- dation models,

    Y . Liu, L. Kong, J. Cen, R. Chen, W. Zhang, L. Pan, K. Chen, and Z. Liu, “Segment any point cloud sequences by distilling vision foun- dation models,” Advances in Neural Information Processing Systems , vol. 36, 2024. 1, 2

    2024

  19. [27]

    Geometric calibration for lidar-camera system fusing 3d-2d and 3d-3d point correspondences,

    P. An, T. Ma, K. Yu, B. Fang, J. Zhang, W. Fu, and J. Ma, “Geometric calibration for lidar-camera system fusing 3d-2d and 3d-3d point correspondences,” Optics express, vol. 28, no. 2, pp. 2122–2141, 2020. 2, 4

    2020

  20. [28]

    Learning to adapt structured output space for seman- tic segmentation,

    Y .-H. Tsai, W.-C. Hung, S. Schulter, K. Sohn, M.-H. Yang, and M. Chandraker, “Learning to adapt structured output space for seman- tic segmentation,” in Proceedings of the IEEE conference on computer vision and pattern recognition , 2018, pp. 7472–7481. 2

    2018

  21. [29]

    Conditional generative adversarial network for structured domain adaptation,

    W. Hong, Z. Wang, M. Yang, and J. Yuan, “Conditional generative adversarial network for structured domain adaptation,” in Proceedings of the IEEE conference on computer vision and pattern recognition , 2018, pp. 1335–1344. 2

    2018

  22. [30]

    Class-balanced pixel-level self-labeling for domain adaptive semantic segmentation,

    R. Li, S. Li, C. He, Y . Zhang, X. Jia, and L. Zhang, “Class-balanced pixel-level self-labeling for domain adaptive semantic segmentation,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2022, pp. 11 593–11 603. 2

    2022

  23. [31]

    Instance adaptive self- training for unsupervised domain adaptation,

    K. Mei, C. Zhu, J. Zou, and S. Zhang, “Instance adaptive self- training for unsupervised domain adaptation,” in Computer Vision– ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Proceedings, Part XXVI 16 . Springer, 2020, pp. 415–430. 2

    2020

  24. [32]

    Unsuper- vised intra-domain adaptation for semantic segmentation through self- supervision,

    F. Pan, I. Shin, F. Rameau, S. Lee, and I. S. Kweon, “Unsuper- vised intra-domain adaptation for semantic segmentation through self- supervision,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , 2020, pp. 3764–3773. 2

    2020

  25. [33]

    Learning pseudo-relations for cross-domain semantic segmentation,

    D. Zhao, S. Wang, Q. Zang, D. Quan, X. Ye, R. Yang, and L. Jiao, “Learning pseudo-relations for cross-domain semantic segmentation,” in Proceedings of the IEEE/CVF International Conference on Com- puter Vision, 2023, pp. 19 191–19 203. 2

    2023

  26. [34]

    Cosmix: Compositional semantic mix for domain adaptation in 3d lidar segmentation,

    C. Saltori, F. Galasso, G. Fiameni, N. Sebe, E. Ricci, and F. Poiesi, “Cosmix: Compositional semantic mix for domain adaptation in 3d lidar segmentation,” in European Conference on Computer Vision . Springer, 2022, pp. 586–602. 2

    2022

  27. [35]

    A2d2: Audi autonomous driving dataset,

    J. Geyer, Y . Kassahun, M. Mahmudi, X. Ricou, R. Durgesh, A. S. Chung, L. Hauswald, V . H. Pham, M. M¨uhlegg, S. Dorn, et al., “A2d2: Audi autonomous driving dataset,” arXiv preprint arXiv:2004.06320 ,

    2004 arXiv

  28. [36]

    Transfusion: Robust lidar-camera fusion for 3d object detection with transformers,

    X. Bai, Z. Hu, X. Zhu, Q. Huang, Y . Chen, H. Fu, and C.-L. Tai, “Transfusion: Robust lidar-camera fusion for 3d object detection with transformers,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , 2022, pp. 1090–1099. 2

    2022

  29. [37]

    Efficient deep visual and iner- tial odometry with adaptive visual modality selection,

    M. Yang, Y . Chen, and H.-S. Kim, “Efficient deep visual and iner- tial odometry with adaptive visual modality selection,” in European Conference on Computer Vision . Springer, 2022, pp. 233–250. 2

    2022

  30. [38]

    Sparse-to-dense feature matching: Intra and inter domain cross-modal learning in domain adaptation for 3d semantic segmentation,

    D. Peng, Y . Lei, W. Li, P. Zhang, and Y . Guo, “Sparse-to-dense feature matching: Intra and inter domain cross-modal learning in domain adaptation for 3d semantic segmentation,” in Proceedings of the IEEE/CVF International Conference on Computer Vision , 2021, pp. 7108–7117. 2

    2021

  31. [39]

    Cross-modal contrastive learning for domain adaptation in 3d semantic segmenta- tion,

    B. Xing, X. Ying, R. Wang, J. Yang, and T. Chen, “Cross-modal contrastive learning for domain adaptation in 3d semantic segmenta- tion,” in Proceedings of the AAAI Conference on Artificial Intelligence, vol. 37, no. 3, 2023, pp. 2974–2982. 2

    2023

  32. [40]

    Reliable spatial-temporal voxels for multi-modal test-time adaptation,

    H. Cao, Y . Xu, J. Yang, P. Yin, X. Ji, S. Yuan, and L. Xie, “Reliable spatial-temporal voxels for multi-modal test-time adaptation,” arXiv preprint arXiv:2403.06461, 2024. 2

    2024

  33. [41]

    Weakly-supervised concealed object segmentation with sam- based pseudo labeling and multi-scale feature grouping,

    C. He, K. Li, Y . Zhang, G. Xu, L. Tang, Y . Zhang, Z. Guo, and X. Li, “Weakly-supervised concealed object segmentation with sam- based pseudo labeling and multi-scale feature grouping,” Advances in Neural Information Processing Systems , vol. 36, 2024. 2

    2024

  34. [42]

    Sam3d: Segment anything in 3d scenes,

    Y . Yang, X. Wu, T. He, H. Zhao, and X. Liu, “Sam3d: Segment anything in 3d scenes,” arXiv preprint arXiv:2306.03908 , 2023. 2

    2023 arXiv

  35. [43]

    Track anything: Segment anything meets videos,

    J. Yang, M. Gao, Z. Li, S. Gao, F. Wang, and F. Zheng, “Track anything: Segment anything meets videos,” arXiv preprint arXiv:2304.11968, 2023. 2

    2023 arXiv

  36. [44]

    Segment and track anything,

    Y . Cheng, L. Li, Y . Xu, X. Li, Z. Yang, W. Wang, and Y . Yang, “Segment and track anything,”arXiv preprint arXiv:2305.06558, 2023. 2

    2023 arXiv

  37. [45]

    Personalize segment anything model with one shot,

    R. Zhang, Z. Jiang, Z. Guo, S. Yan, J. Pan, X. Ma, H. Dong, P. Gao, and H. Li, “Personalize segment anything model with one shot,” arXiv preprint arXiv:2305.03048, 2023. 2

    2023 arXiv

  38. [46]

    Medical sam adapter: Adapting segment anything model for medical image segmentation,

    J. Wu, W. Ji, Y . Liu, H. Fu, M. Xu, Y . Xu, and Y . Jin, “Medical sam adapter: Adapting segment anything model for medical image segmentation,” arXiv preprint arXiv:2304.12620 , 2023. 2

    2023 arXiv

  39. [47]

    A new local distance-based outlier detection approach for scattered real-world data,

    K. Zhang, M. Hutter, and H. Jin, “A new local distance-based outlier detection approach for scattered real-world data,” in Advances in Knowledge Discovery and Data Mining: 13th Pacific-Asia Conference, PAKDD 2009 Bangkok, Thailand, April 27-30, 2009 Proceedings 13 . Springer, ...

    2009

  40. [48]

    An efficient outlier removal method for scattered point cloud data,

    X. Ning, F. Li, G. Tian, and Y . Wang, “An efficient outlier removal method for scattered point cloud data,” PloS one , vol. 13, no. 8, p. e0201280, 2018. 4

    2018

  41. [49]

    Deep residual learning for image recognition,

    K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learning for image recognition,” in Proceedings of the IEEE conference on computer vision and pattern recognition , 2016, pp. 770–778. 5

    2016

  42. [50]

    3d semantic segmentation with submanifold sparse convolutional networks,

    B. Graham, M. Engelcke, and L. Van Der Maaten, “3d semantic segmentation with submanifold sparse convolutional networks,” in Proceedings of the IEEE conference on computer vision and pattern recognition, 2018, pp. 9224–9232. 5

    2018

  43. [51]

    Adam: A method for stochastic optimization,

    D. P. Kingma, “Adam: A method for stochastic optimization,” arXiv preprint arXiv:1412.6980, 2014. 6

    2014 arXiv

Pith tools

Reviewed August 9, 2026 · model on record in the stance chip above.