Pith. sign in

REVIEW 5 major objections 5 minor 112 references

Beyond One Shot, Beyond One Perspective: Cross-View and Long-Horizon Distillation for Better LiDAR Representations

T0 review · 5 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read LiMA aggregates six frames of cross-view image features and distills them into LiDAR, improving segmentation and detection over single-frame pretraining.

desk verdict Solid subfield contribution with a genuine but fixable reproducibility gap in the temporal warping module. read the letter →

arxiv 2507.05260 v1 pith:G23WUXQQ submitted 2025-07-07 cs.CV cs.LGcs.RO

Xiang Xu , Lingdong Kong , Song Wang , Chuanwei Zhou , Qingshan Liu This is my paper
classification cs.CVcs.LGcs.RO
keywords LiDARrepresentationlearningimage-to-Lidistillationlong-termtemporalmodelingcross-viewaggregationmemorybanksemanticsegmentation3Dobjectdetectionautonomousdriving
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 LiDAR representation learning benefits from distilling long-term image context, not just spatial alignment or two-frame consistency. It proposes LiMA, which first averages features from overlapping cameras into one unified feature map, then warps and pools image features from the previous six frames into a temporally enriched teacher, and finally distills this teacher into a LiDAR backbone using an $\ell^2$ loss. If correct, this means a pretrained LiDAR model can absorb motion and scene evolution from image sequences without any additional labels, and it generalizes better to low-annotation regimes, new datasets, and corruptions. The strongest evidence is Table 1, where a ViT-B teacher gives 56.65 mIoU linear probing and 51.29 mIoU at 1% fine-tuning on nuScenes, against 51.90 and 48.90 for the single-frame ScaLR baseline.

What carries the argument

The load-bearing object is the memory bank: a FIFO buffer of unified image features from the past $k$ frames. At each timestamp, features from overlapping cameras are averaged into a single unified map, historical maps are warped into the current ego-vehicle frame via temporal calibration, and the bank's contents are average-pooled with the current features to form the teacher signal distilled into LiDAR points with an $\ell^2$ loss. The bank is what lets LiMA capture long-range dependencies without recomputing features from raw images every step, and it is what Table 7 shows to have an optimal size around six frames.

What would settle it

Run LiMA pretraining with the ego-motion warp replaced by identity alignment but keep the same six-frame averaging; if the gain over the single-frame baseline largely remains, temporal alignment is not what carries the result, while if the gain disappears, the alignment assumption is confirmed as load-bearing.

Watch

Extended reading notes

Core claim

The central claim is that modeling long-range spatiotemporal structure during image-to-LiDAR pretraining produces better LiDAR features than spatial-only or short-term methods. LiMA does this by maintaining a FIFO memory bank of unified cross-view image features, using ego-motion transforms to align historical frames into the current coordinate system, average-pooling them with the current frame, and distilling the result into the LiDAR backbone. The paper further claims that averaging across overlapping camera views removes the optimization conflicts that arise when one LiDAR point matches inconsistent pixels from multiple cameras, and that mixing LiDAR scenes from different sequences during pretraining improves cross-domain generalization. On the paper's evidence, each of the three components contributes: cross-view aggregation adds about 1.8 mIoU in linear probing, long-term propagation adds about 1.5, and cross-sequence alignment adds about 1.4, with the full system beating prior pretraining methods on segmentation and detection benchmarks.

Load-bearing premise

The method assumes that ego-motion calibration is accurate enough that warping and averaging historical image features into the current frame produces a clean teacher signal; when calibration drifts or too many frames accumulate, the averaged features add noise instead of information.

Editorial extensions

If this is right

  • Pretraining for LiDAR perception no longer needs to treat scans as independent; a memory bank of image features from past frames gives the LiDAR learner motion-aware targets at no extra downstream cost.
  • Longer temporal context, up to about six frames, improves linear probing and few-shot fine-tuning across nuScenes, SemanticKITTI, and Waymo, suggesting temporal consistency is a transferable pretraining signal.
  • Distilling temporally fused image features improves 3D object detection in low-annotation regimes such as 5% to 20% labels, so one pretrained backbone can serve both segmentation and detection tasks.
  • Cross-sequence mixing during pretraining improves robustness under corruptions on nuScenes-C, implying domain generalization can be built into the representation before downstream labels are seen.
  • The FIFO memory bank keeps total pretraining under 20 hours even with eight frames, so long-horizon pretraining remains computationally affordable.

Reading between the lines

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

  • Editorial inference: the paper's own Table 7 shows gains peaking at six frames and dropping at seven to eight, so an untested extension is to adapt memory length to scene dynamics, such as vehicle speed or the number of moving objects, rather than fixing it.
  • Editorial inference: because the teacher signal is entirely image-derived, a natural transfer test is whether the same memory-aggregation recipe works for LiDAR-only self-supervised pretraining, which would remove the dependence on cameras and calibration.
  • Editorial inference: the cross-sequence mixing component suggests that segment-level domain randomization, rather than only scene-level mixing, could push generalization further; a testable variant is mixing more than two sequences per pretraining step.
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

5 major / 5 minor

Summary. The paper proposes LiMA, a long-term image-to-LiDAR distillation framework that combines cross-view aggregation of multi-camera image features, temporal feature propagation through a FIFO memory bank, and cross-sequence mixing for domain generalization. The image teacher is a frozen DINOv2 network, and the distillation target is produced by aggregating aligned historical image features and the current frame's features. The authors pretrain MinkUNet and VoxelNet backbones on nuScenes and evaluate on a broad set of downstream segmentation and detection benchmarks, including low-data fine-tuning, cross-domain transfer, and corrupted-input robustness. The main claims are that LiMA outperforms prior image-to-LiDAR pretraining methods, especially under 1% fine-tuning and linear probing, and that the long-term memory mechanism is the source of these gains.

Significance. If the claims hold, the paper would make a useful empirical contribution to LiDAR representation learning by showing that multi-frame image features can be distilled into LiDAR models more effectively than single-frame or short-term methods. The benchmark coverage is unusually broad: nuScenes, SemanticKITTI, Waymo, seven additional segmentation datasets, robustness evaluation on nuScenes-C, and two detection frameworks. The code is released, the teacher is a frozen external model, and the evaluation protocol follows prior work, which reduces circularity concerns. The paper also reports pretraining time and memory trade-offs, which is a practical strength. The main risk is attribution: the long-term propagation module, which accounts for the headline improvements, is not specified precisely enough to be reproduced or to cleanly separate temporal coherence from implementation choices.

major comments (5)
  1. [Sec. 4.2] The geometric operation underlying Long-Term Feature Propagation is underspecified. The text states that historical unified image features are 'warped into the current frame's coordinate space through temporal calibration [79]', but a 2D image feature map cannot be transported by an ego-motion transform alone without per-pixel depth or an explicit projection into a shared 3D or BEV space. Reference [79] is an object-centric 3D detection method and does not provide a dense 2D feature-warping recipe. Please specify the exact algorithm—e.g., whether historical image features are sampled at the projections of ego-motion-compensated LiDAR points—and state how occlusions and independently moving objects are handled. Without this specification, the gains in Tables 1 and 7 cannot be cleanly attributed to long-term temporal coherence rather than to a particular implementation choice or multi-frame averaging of static-scene features.
  2. [Sec. 5.2, Table 3] The robustness claim is overstated. The text says LiMA 'consistently outperforms recent pretraining methods across most corruption types,' but in the Full fine-tuning regime LiMA is worse than SuperFlow on mRR (82.57 vs. 83.17), on Rain (73.38 vs. 75.77), on Beam (66.71 vs. 68.09), on Cross (47.66 vs. 60.02), on Sensor (48.65 vs. 50.41), and on Average mIoU (62.93 vs. 63.68). Only in the LP regime is LiMA uniformly better on the average. Please remove or qualify the 'consistently outperforms' wording and discuss the mixed results, which include a large 12-point deficit on the Cross corruption.
  3. [Sec. 4.3] Cross-Sequence Memory Alignment has no formal objective. The mechanism is described only verbally as making mixed features 'optimized to maintain structural coherence with the original sequence memory banks'; no loss function, pseudocode, or implementation detail is provided. Since Table 5 attributes a 1.39 mIoU LP gain and a 0.9 mIoU fine-tuning gain to this component, please give the exact loss, how the mixed scene is generated in the feature space, and how the memory banks of the two source sequences are used.
  4. [Sec. 5.1 and Tables 1-7] No error bars or number of seeds are reported. The headline improvements over ScaLR are often 2-3 mIoU (e.g., Table 1, ViT-B 1%: 51.29 vs. 48.90), and the memory-size ablation in Table 7 varies by less than 2 mIoU between 5 and 8 frames. Please report the mean and standard deviation over at least three fine-tuning seeds for the central comparisons, or justify why a single seed is sufficient.
  5. [Sec. 4.1] The cross-view aggregation rule is incompletely specified. It says 'for each LiDAR point appearing in multiple views, we extract the corresponding pixel-aligned features... and aggregate them with an averaging operation.' A LiDAR point does not necessarily have a visible projection in every camera whose frustum contains it; please specify how visibility and occlusion are determined and whether averaging includes features from cameras where the point is occluded. This matters because the ablation in Table 6 shows a 1.82 mIoU LP improvement from cross-view aggregation.
minor comments (5)
  1. [Abstract] The sentence 'The code has be made publicly accessible' should read 'The code has been made publicly accessible.'
  2. [Sec. 1] The sentence 'Notably, Our framework ensures high pretraining efficiency' has an inconsistent capitalization: 'Our' should be lowercase.
  3. [Sec. 5.2] The phrase 'under limited annotation settings (e.g., 1%, 5%, and 10%)' is imprecise for linear probing, which uses no annotations for the backbone; consider distinguishing 'labeled-data regimes' from 'LP'.
  4. [Sec. 7.2] The image preprocessing paragraph says 'Following ScaLR [63], in this work, we do not utilize any Vision Foundation Models (VFMs) to generate superpixels in this work.' The sentence is redundant and appears to accidentally retain a note-to-self; please revise.
  5. [Sec. 8.3, Eq. (14)] The contrastive loss in Eq. (14) pools over all j in the denominator, which corresponds to a global softmax over features; please clarify whether the negatives are sampled from the same batch or from the whole point cloud and whether hard negatives are used, since Table 8 compares this objective against ℓ2 and cosine losses.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the distillation target is an external frozen teacher, design choices are validated empirically, and self-citations are non-load-bearing.

full rationale

The paper's central claims are empirical: LiMA's three modules (cross-view aggregation, long-term feature propagation, and cross-sequence memory alignment) improve downstream LiDAR segmentation and detection. The teacher signal is DINOv2, a frozen external model, and the distillation objective is a standard L2 loss over image-LiDAR feature pairs (Eq. 3). Downstream evaluation is performed on independent benchmarks (nuScenes, SemanticKITTI, Waymo, and others), so the reported gains are not defined in terms of the method's own fitted quantities. The memory-bank size k=6, mean aggregation, and L2 distillation are selected through ablations (Tables 5–8) and are presented as ablations rather than as predictions derived from first principles. The cross-sequence alignment uses established mixing strategies (LaserMix and PolarMix) as an additional training signal, not as a restatement of the test outcome. Self-citations appear in related work and as baselines (SuperFlow, Seal, FRNet, LiMoE, SuperFlow++), but none is load-bearing for the method's correctness or for forbidding alternatives; there is no uniqueness theorem, ansatz, or fitted-parameter renaming imported from the authors' prior work. The paper even acknowledges a limitation (sensitivity to sensor calibration, Sec. 10.2), which is consistent with the mechanism being fragile, but fragility is not circularity. Consequently, the derivation chain is self-contained and the circularity score is 0.

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

The central claim depends on calibration, ego-motion warping, DINOv2 teacher quality, and the validity of mixed-scene alignment. These are domain assumptions typical of this literature. The free parameters are design choices selected by validation, not quantities derived from data through a theoretical model.

free parameters (4)
  • Memory bank length k = 6 frames (Table 7)
    Chosen by sweeping 2-8 frames; 6 is best on nuScenes LP, though 7-8 continue improving KITTI 1%.
  • Cross-view aggregation function = Average pooling (Table 6)
    Average beats max and an attention module in the ablation.
  • Distillation objective = L2 distance (Table 8)
    L2 beats cosine, contrastive, and KL in the comparison.
  • Mixing strategy for cross-sequence alignment = LaserMix/PolarMix
    Borrowed from prior work, not ablated; no ablation shown for which mix or mixing ratio.
assumptions (4)
  • domain assumption Camera-LiDAR calibration matrices are accurate and time-synchronized
    Eq. (1) projects points through intrinsics/extrinsics; Sec. 10.2 lists misalignment as a limitation.
  • domain assumption Ego-motion transform warps historical image features into the current frame
    Sec. 4.2 says features are warped 'through temporal calibration'; errors compound over k frames, and Table 7 shows degradation at 7-8 frames.
  • domain assumption DINOv2 features are a transferable semantic prior for LiDAR points
    The teacher is a frozen DINOv2; the whole distillation objective assumes these 2D features are the right target.
  • domain assumption Mixed LiDAR scenes retain the semantics needed for alignment
    Sec. 4.3 relies on LaserMix/PolarMix scene mixing, assuming labels and geometry of mixed scenes line up with memory banks.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Beyond One Shot, Beyond One Perspective: Cross-View and Long-Horizon Distillation for Better LiDAR Representations." pith.science (2026). https://pith.science/paper/G23WUXQQ

@misc{pith2026250705260,
  author       = {Pith},
  title        = {Pith review of: Beyond One Shot, Beyond One Perspective: Cross-View and Long-Horizon Distillation for Better LiDAR Representations},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/G23WUXQQ}},
  note         = {Machine review of arXiv:2507.05260}
}
read the original abstract

LiDAR representation learning aims to extract rich structural and semantic information from large-scale, readily available datasets, reducing reliance on costly human annotations. However, existing LiDAR representation strategies often overlook the inherent spatiotemporal cues in LiDAR sequences, limiting their effectiveness. In this work, we propose LiMA, a novel long-term image-to-LiDAR Memory Aggregation framework that explicitly captures longer range temporal correlations to enhance LiDAR representation learning. LiMA comprises three key components: 1) a Cross-View Aggregation module that aligns and fuses overlapping regions across neighboring camera views, constructing a more unified and redundancy-free memory bank; 2) a Long-Term Feature Propagation mechanism that efficiently aligns and integrates multi-frame image features, reinforcing temporal coherence during LiDAR representation learning; and 3) a Cross-Sequence Memory Alignment strategy that enforces consistency across driving sequences, improving generalization to unseen environments. LiMA maintains high pretraining efficiency and incurs no additional computational overhead during downstream tasks. Extensive experiments on mainstream LiDAR-based perception benchmarks demonstrate that LiMA significantly improves both LiDAR semantic segmentation and 3D object detection. We hope this work inspires more effective pretraining paradigms for autonomous driving. The code has be made publicly accessible for future research.

Figures

Figures reproduced from arXiv: 2507.05260 by the authors.

Figure 1
Figure 1. Illustrative examples of image-to-LiDAR pretrain￾ing paradigms. (a) Spatial Alignment aligns LiDAR features with corresponding image features in the spatial domain without considering temporal consistency. (b) Short-Term methods propa￾gate LiDAR features frame by frame, ensuring feature consistency across neighboring frames, but fail to capture long-term depen￾dencies. (c) Our approach leverages Long-Term image sequ… view at source ↗
Figure 2
Figure 2. Overview of the LiMA framework. At each timestamp t, multi-view image features are first extracted and unified through the Cross-View Aggregation module, producing F t u (Sec. 4.1). A Memory Bank is introduced to maintain unified image features from the past k frames, enabling temporal feature propagation and fusion with F t u to capture Long-Term Motion Patterns F t d (Sec. 4.2). The enriched temporal features are … view at source ↗
Figure 3
Figure 3. Illustration of cross-sequence memory alignment. Ex￾isting mixed strategies [34, 84] are utilized to blend two LiDAR scenes from distinct sequences, generating out-of-distribution variations. This process encourages feature alignment across se￾quences by leveraging memory banks from the original sequences. yields a set of frame-aligned features {Fbt−k u , . . . , Fbt−1 u }, effectively compensating for vehicle motio… view at source ↗
Figures from the paper (8 more)
Figure 4
Figure 4. Figure 4: Cosine similarity between a query point (marked as red dot) and: (1) image features, and (2) LiDAR features projected onto the image. Colors range from red (indicating high similarity) to blue (indicating low similarity). Best viewed in colors [PITH_FULL_IMAGE:figures…
Figure 5
Figure 5. Figure 5: Qualitative assessments of state-of-the-art methods, pretrained on nuScenes [5] and fine-tuned on nuScenes [18] with 1% annotations. The error maps depict correct and incorrect pre￾dictions in gray and red, respectively. Best viewed in colors. ScaLR LiMA (Ours) Ground …
Figure 6
Figure 6. Figure 6: Visual comparisons from the 3D object detection task, where methods are pretrained on nuScenes [5] and fine-tuned on nuScenes [5] with 5% annotations. Best viewed in colors [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]
Figure 8
Figure 8. Figure 8: Qualitative assessments of state-of-the-art methods, pretrained on nuScenes [5] and fine-tuned on nuScenes [18] with 1% annotations. The error maps depict correct and incorrect predictions in gray and red, respectively. Best viewed in colors. 16 [PITH_FULL_IMAGE:figur…
Figure 9
Figure 9. Figure 9: Qualitative assessments of state-of-the-art methods, pretrained on nuScenes [5] and fine-tuned on SemanticKITTI [2] with 1% annotations. The error maps depict correct and incorrect predictions in gray and red, respectively. Best viewed in colors. 17 [PITH_FULL_IMAGE:f…
Figure 10
Figure 10. Figure 10: Qualitative assessments of state-of-the-art methods, pretrained on nuScenes [5] and fine-tuned on Waymo [74] with 1% annotations. The error maps depict correct and incorrect predictions in gray and red, respectively. Best viewed in colors. 18 [PITH_FULL_IMAGE:figures…
Figure 11
Figure 11. Figure 11: Qualitative assessments of object detection, pretrained on nuScenes [5] and fine-tuned on nuScenes [5] with 5% annotations. The groundtruth / predicted results are highlighted with blue / red boxes, respectively. Best viewed in colors. 19 [PITH_FULL_IMAGE:figures/ful…
Figure 12
Figure 12. Figure 12: Cosine similarity between a query point (marked as the red dot) and: (1) image features, and (2) LiDAR point features projected onto the image. Colors range from red (indicating high similarity) to blue (indicating low similarity). Best viewed in colors. 20 [PITH_FUL…

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

112 extracted references · 71 canonical work pages

  1. [79]

    Exploring object-centric temporal modeling for efficient multi-view 3d object detection

    Shihao Wang, Yingfei Liu, Tiancai Wang, Ying Li, and Xi- angyu Zhang. Exploring object-centric temporal modeling for efficient multi-view 3d object detection. In IEEE/CVF International Conference on Computer Vision, pages 3621– 3631, 2023. 2, 4

    work page 2023

  2. [1]

    Rangevit: Towards vision transformers for 3d semantic segmentation in au- tonomous driving

    Angelika Ando, Spyros Gidaris, Andrei Bursuc, Gilles Puy, Alexandre Boulch, and Renaud Marlet. Rangevit: Towards vision transformers for 3d semantic segmentation in au- tonomous driving. In IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 5240–5250, 2023. 2

    2023

  3. [2]

    Se- mantickitti: A dataset for semantic scene understanding of lidar sequences

    Jens Behley, Martin Garbade, Andres Milioto, Jan Quen- zel, Sven Behnke, Cyrill Stachniss, and Jurgen Gall. Se- mantickitti: A dataset for semantic scene understanding of lidar sequences. In IEEE/CVF International Conference on Computer Vision, pages 9297–9307, 2019. 5, 6, 8, 9, 12, 17

    2019

  4. [3]

    Dynamiccity: Large-scale 4d oc- cupancy generation from dynamic scenes

    Hengwei Bian, Lingdong Kong, Haozhe Xie, Liang Pan, Yu Qiao, and Ziwei Liu. Dynamiccity: Large-scale 4d oc- cupancy generation from dynamic scenes. In International Conference on Learning Representations, 2025. 1

    2025

  5. [4]

    Also: Automotive lidar self- supervision by occupancy estimation

    Alexandre Boulch, Corentin Sautier, Bj ¨orn Michele, Gilles Puy, and Renaud Marlet. Also: Automotive lidar self- supervision by occupancy estimation. In IEEE/CVF Con- ference on Computer Vision and Pattern Recognition, pages 13455–13465, 2023. 14, 15

    2023

  6. [5]

    nuscenes: A multi- modal dataset for autonomous driving

    Holger Caesar, Varun Bankiti, Alex H Lang, Sourabh V ora, Venice Erin Liong, Qiang Xu, Anush Krishnan, Yu Pan, Giancarlo Baldan, and Oscar Beijbom. nuscenes: A multi- modal dataset for autonomous driving. In IEEE/CVF Con- ference on Computer Vision and Pattern Recognition, pages 11621–11631, 2020. 5, 6, 7, 8, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19

    2020

  7. [6]

    Building a strong pre- training baseline for universal 3d large-scale perception

    Haoming Chen, Zhizhong Zhang, Yanyun Qu, Ruixin Zhang, Xin Tan, and Yuan Xie. Building a strong pre- training baseline for universal 3d large-scale perception. In IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 19925–19935, 2024. 5, 6, 7, 14, 15

    2024

  8. [7]

    Polarstream: Streaming object detection and segmentation with polar pil- lars

    Qi Chen, Sourabh V ora, and Oscar Beijbom. Polarstream: Streaming object detection and segmentation with polar pil- lars. In Advances in Neural Information Processing Sys- tems, pages 26871–26883, 2021. 2

    2021

Show all 112 references
  1. [8]

    Towards label-free scene understanding by vi- sion foundation models

    Runnan Chen, Youquan Liu, Lingdong Kong, Nenglun Chen, Xinge Zhu, Yuexin Ma, Tongliang Liu, and Wen- ping Wang. Towards label-free scene understanding by vi- sion foundation models. In Advances in Neural Information Processing Systems, pages 75896–75910, 2023. 2

    2023

  2. [9]

    Clip2scene: Towards label-efficient 3d scene under- standing by clip

    Runnan Chen, Youquan Liu, Lingdong Kong, Xinge Zhu, Yuexin Ma, Yikang Li, Yuenan Hou, Yu Qiao, and Wenping Wang. Clip2scene: Towards label-efficient 3d scene under- standing by clip. In IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 7020–7030, 2023. 2

    2023

  3. [10]

    4dcontrast: Contrastive learning with dynamic correspondences for 3d scene understanding

    Yujin Chen, Matthias Nießner, and Angela Dai. 4dcontrast: Contrastive learning with dynamic correspondences for 3d scene understanding. InEuropean Conference on Computer Vision, pages 543–560, 2022. 2

    2022

  4. [11]

    Af2-s3net: Attentive feature fusion with adaptive feature selection for sparse semantic segmentation network

    Ran Cheng, Ryan Razani, Ehsan Taghavi, Enxu Li, and Bingbing Liu. Af2-s3net: Attentive feature fusion with adaptive feature selection for sparse semantic segmentation network. In IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 12547–12556, 2021. 2

    2021

  5. [12]

    4d spatio-temporal convnets: Minkowski convolutional neural networks

    Christopher Choy, JunYoung Gwak, and Silvio Savarese. 4d spatio-temporal convnets: Minkowski convolutional neural networks. In IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 3075–3084, 2019. 2, 5, 6

    2019

  6. [13]

    MMDetection3D: Open- MMLab next-generation platform for general 3D object detection

    MMDetection3D Contributors. MMDetection3D: Open- MMLab next-generation platform for general 3D object detection. https://github.com/open- mmlab/ mmdetection3d, 2020. 5

    2020

  7. [14]

    An im- age is worth 16x16 words: Transformers for image recog- nition at scale

    Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Sylvain Gelly, Jakob Uszkoreit, and Neil Houlsby. An im- age is worth 16x16 words: Transformers for image recog- nitio...

    2021

  8. [15]

    Self-supervised learning for pre-training 3d point clouds: A survey

    Ben Fei, Weidong Yang, Liwen Liu, Tianyue Luo, Rui Zhang, Yixuan Li, and Ying He. Self-supervised learning for pre-training 3d point clouds: A survey. arXiv preprint arXiv:2305.04691, 2023. 1

    2023 arXiv

  9. [16]

    A survey of world models for autonomous driving

    Tuo Feng, Wenguan Wang, and Yi Yang. A survey of world models for autonomous driving. arXiv preprint arXiv:2501.11260, 2025. 2

    2025 arXiv

  10. [17]

    Random sample consensus: a paradigm for model fitting with applications to image analysis and automated cartography

    Martin A Fischler and Robert C Bolles. Random sample consensus: a paradigm for model fitting with applications to image analysis and automated cartography. Communica- tions of the ACM, 24(6):381–395, 1981. 3

    1981

  11. [18]

    Panoptic nuscenes: A large-scale benchmark for lidar panoptic segmentation and tracking

    Whye Kit Fong, Rohit Mohan, Juana Valeria Hurtado, Lub- ing Zhou, Holger Caesar, Oscar Beijbom, and Abhinav Val- ada. Panoptic nuscenes: A large-scale benchmark for lidar panoptic segmentation and tracking. IEEE Robotics and Automation Letters, 7(2):3795–3802, 2022. 5, 6, 8, 9...

    2022

  12. [19]

    Are we hungry for 3d lidar data for semantic segmentation? a survey of datasets and methods

    Biao Gao, Yancheng Pan, Chengkun Li, Sibo Geng, and Huijing Zhao. Are we hungry for 3d lidar data for semantic segmentation? a survey of datasets and methods. IEEE Transactions on Intelligent Transportation Systems, 23(7): 6063–6081, 2021. 1

    2021

  13. [20]

    Are we ready for autonomous driving? the kitti vision benchmark suite

    Andreas Geiger, Philip Lenz, and Raquel Urtasun. Are we ready for autonomous driving? the kitti vision benchmark suite. In IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 3354–3361, 2012. 9

    2012

  14. [21]

    Deep learning for 3d point clouds: A survey

    Yulan Guo, Hanyun Wang, Qingyong Hu, Hao Liu, Li Liu, and Mohammed Bennamoun. Deep learning for 3d point clouds: A survey. IEEE Transactions on Pattern Analysis and Machine Intelligence, 43(12):4338–4364, 2020. 1

    2020

  15. [22]

    Is your hd map construc- tor reliable under sensor corruptions? In Advances in Neu- ral Information Processing Systems , pages 22441–22482,

    Xiaoshuai Hao, Mengchuan Wei, Yifan Yang, Haimei Zhao, Hui Zhang, Yi Zhou, Qiang Wang, Weiming Li, Lingdong Kong, and Jing Zhang. Is your hd map construc- tor reliable under sensor corruptions? In Advances in Neu- ral Information Processing Systems , pages 22441–22482,

  16. [23]

    Deep residual learning for image recognition

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. InIEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 770–778, 2016. 6

    2016

  17. [24]

    Lidar-based panoptic segmentation via dynamic shifting network

    Fangzhou Hong, Hui Zhou, Xinge Zhu, Hongsheng Li, and Ziwei Liu. Lidar-based panoptic segmentation via dynamic shifting network. In IEEE/CVF Conference on Computer 22 Vision and Pattern Recognition, pages 13090–13099, 2021. 2

    2021

  18. [25]

    Unified 3d and 4d panoptic segmentation via dynamic shifting networks

    Fangzhou Hong, Lingdong Kong, Hui Zhou, Xinge Zhu, Hongsheng Li, and Ziwei Liu. Unified 3d and 4d panoptic segmentation via dynamic shifting networks. IEEE Trans- actions on Pattern Analysis and Machine Intelligence , 46 (5):3480–3495, 2024. 2

    2024

  19. [26]

    Randla-net: Efficient semantic segmentation of large-scale point clouds

    Qingyong Hu, Bo Yang, Linhai Xie, Stefano Rosa, Yulan Guo, Zhihua Wang, Niki Trigoni, and Andrew Markham. Randla-net: Efficient semantic segmentation of large-scale point clouds. In IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 11108–11117, 2020. 2

    2020

  20. [27]

    St-p3: End-to-end vision- based autonomous driving via spatial-temporal feature learning

    Shengchao Hu, Li Chen, Penghao Wu, Hongyang Li, Junchi Yan, and Dacheng Tao. St-p3: End-to-end vision- based autonomous driving via spatial-temporal feature learning. In European Conference on Computer Vision , pages 533–549, 2022. 1

    2022

  21. [28]

    Ponder: Point cloud pre-training via neural rendering

    Di Huang, Sida Peng, Tong He, Honghui Yang, Xiaowei Zhou, and Wanli Ouyang. Ponder: Point cloud pre-training via neural rendering. In IEEE/CVF International Confer- ence on Computer Vision, pages 16089–16098, 2023. 2

    2023

  22. [29]

    Spatio-temporal self-supervised representation learn- ing for 3d point clouds

    Siyuan Huang, Yichen Xie, Song-Chun Zhu, and Yixin Zhu. Spatio-temporal self-supervised representation learn- ing for 3d point clouds. In IEEE/CVF International Con- ference on Computer Vision, pages 6535–6545, 2021. 2

    2021

  23. [30]

    Rellis-3d dataset: Data, benchmarks and anal- ysis

    Peng Jiang, Philip Osteen, Maggie Wigness, and Srikanth Saripalli. Rellis-3d dataset: Data, benchmarks and anal- ysis. In IEEE International Conference on Robotics and Automation, pages 1110–1116, 2021. 5, 6, 9

    2021

  24. [31]

    Daps3d: Domain adaptive projective segmen- tation of 3d lidar point clouds

    Alexey A Klokov, Di Un Pak, Aleksandr Khorin, Dmitry A Yudin, Leon Kochiev, Vladimir D Luchinskiy, and Vitaly D Bezuglyj. Daps3d: Domain adaptive projective segmen- tation of 3d lidar point clouds. IEEE Access, 11:79341– 79356, 2023. 5, 6, 10

    2023

  25. [32]

    Rethinking range view representation for lidar segmentation

    Lingdong Kong, Youquan Liu, Runnan Chen, Yuexin Ma, Xinge Zhu, Yikang Li, Yuenan Hou, Yu Qiao, and Zi- wei Liu. Rethinking range view representation for lidar segmentation. In IEEE/CVF International Conference on Computer Vision, pages 228–240, 2023. 2

    2023

  26. [33]

    Robo3d: Towards robust and reliable 3d perception against corruptions

    Lingdong Kong, Youquan Liu, Xin Li, Runnan Chen, Wen- wei Zhang, Jiawei Ren, Liang Pan, Kai Chen, and Ziwei Liu. Robo3d: Towards robust and reliable 3d perception against corruptions. In IEEE/CVF International Confer- ence on Computer Vision , pages 19994–20006, 2023. 5, 6, 7, 10, 11

    2023

  27. [34]

    Lasermix for semi-supervised lidar semantic segmentation

    Lingdong Kong, Jiawei Ren, Liang Pan, and Ziwei Liu. Lasermix for semi-supervised lidar semantic segmentation. In IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 21705–21715, 2023. 2, 5

    2023

  28. [35]

    Robodepth: Robust out-of-distribution depth estimation under corruptions

    Lingdong Kong, Shaoyuan Xie, Hanjiang Hu, Lai Xing Ng, Benoit Cottereau, and Wei Tsang Ooi. Robodepth: Robust out-of-distribution depth estimation under corruptions. In Advances in Neural Information Processing Systems, pages 21298–21342, 2023. 2

    2023

  29. [36]

    Calib3d: Calibrating model preferences for reliable 3d scene understanding

    Lingdong Kong, Xiang Xu, Jun Cen, Wenwei Zhang, Liang Pan, Kai Chen, and Ziwei Liu. Calib3d: Calibrating model preferences for reliable 3d scene understanding. In IEEE/CVF Winter Conference on Applications of Computer Vision, pages 1965–1978, 2025. 1

    1965

  30. [37]

    Largead: Large-scale cross-sensor data pretraining for au- tonomous driving

    Lingdong Kong, Xiang Xu, Youquan Liu, Jun Cen, Runnan Chen, Wenwei Zhang, Liang Pan, Kai Chen, and Ziwei Liu. Largead: Large-scale cross-sensor data pretraining for au- tonomous driving. arXiv preprint arXiv:2501.04005, 2025. 2

    2025

  31. [38]

    Multi-modal data-efficient 3d scene understanding for au- tonomous driving

    Lingdong Kong, Xiang Xu, Jiawei Ren, Wenwei Zhang, Liang Pan, Kai Chen, Wei Tsang Ooi, and Ziwei Liu. Multi-modal data-efficient 3d scene understanding for au- tonomous driving. IEEE Transactions on Pattern Analysis and Machine Intelligence, 47(5):3748–3765, 2025. 2

    2025

  32. [39]

    Pointpillars: Fast encoders for object detection from point clouds

    Alex H Lang, Sourabh V ora, Holger Caesar, Lubing Zhou, Jiong Yang, and Oscar Beijbom. Pointpillars: Fast encoders for object detection from point clouds. In IEEE/CVF Con- ference on Computer Vision and Pattern Recognition, pages 12697–12705, 2019. 2

    2019

  33. [40]

    Less is more: Reducing task and model complexity for 3d point cloud se- mantic segmentation

    Li Li, Hubert PH Shum, and Toby P Breckon. Less is more: Reducing task and model complexity for 3d point cloud se- mantic segmentation. In IEEE/CVF Conference on Com- puter Vision and Pattern Recognition , pages 9361–9371,

  34. [41]

    Seeground: See and ground for zero-shot open- vocabulary 3d visual grounding

    Rong Li, Shijie Li, Lingdong Kong, Xulei Yang, and Jun- wei Liang. Seeground: See and ground for zero-shot open- vocabulary 3d visual grounding. In IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 3707– 3717, 2025. 2

    2025

  35. [42]

    Is your lidar placement optimized for 3d scene understanding? In Advances in Neural Information Processing Systems, pages 34980–35017, 2024

    Ye Li, Lingdong Kong, Hanjiang Hu, Xiaohao Xu, and Xi- aonan Huang. Is your lidar placement optimized for 3d scene understanding? In Advances in Neural Information Processing Systems, pages 34980–35017, 2024. 1

    2024

  36. [43]

    Exploring geometry-aware contrast and cluster- ing harmonization for self-supervised 3d object detection

    Hanxue Liang, Chenhan Jiang, Dapeng Feng, Xin Chen, Hang Xu, Xiaodan Liang, Wei Zhang, Zhenguo Li, and Luc Van Gool. Exploring geometry-aware contrast and cluster- ing harmonization for self-supervised 3d object detection. In IEEE/CVF International Conference on Computer Vi- s...

    2021

  37. [44]

    Sparse4d: Multi-view 3d object detec- tion with sparse spatial-temporal fusion

    Xuewu Lin, Tianwei Lin, Zixiang Pei, Lichao Huang, and Zhizhong Su. Sparse4d: Multi-view 3d object detec- tion with sparse spatial-temporal fusion. arXiv preprint arXiv:2211.10581, 2022. 2, 4

    2022 arXiv

  38. [45]

    Lidar-based 4d occu- pancy completion and forecasting

    Xinhao Liu, Moonjun Gong, Qi Fang, Haoyu Xie, Yim- ing Li, Hang Zhao, and Chen Feng. Lidar-based 4d occu- pancy completion and forecasting. In IEEE/RSJ Interna- tional Conference on Intelligent Robots and Systems, pages 11102–11109, 2024. 1

    2024

  39. [46]

    Box2seg: Learning semantics of 3d point clouds with box-level supervision

    Yan Liu, Qingyong Hu, Yinjie Lei, Kai Xu, Jonathan Li, and Yulan Guo. Box2seg: Learning semantics of 3d point clouds with box-level supervision. arXiv preprint arXiv:2201.02963, 2022. 2

    2022 arXiv

  40. [47]

    Leaf: Learning frames for 4d point cloud sequence understanding

    Yunze Liu, Junyu Chen, Zekai Zhang, Jingwei Huang, and Li Yi. Leaf: Learning frames for 4d point cloud sequence understanding. In IEEE/CVF International Conference on Computer Vision, pages 604–613, 2023. 2

    2023

  41. [48]

    Uniseg: A unified multi-modal li- dar segmentation network and the openpcseg codebase

    Youquan Liu, Runnan Chen, Xin Li, Lingdong Kong, Yuchen Yang, Zhaoyang Xia, Yeqi Bai, Xinge Zhu, Yuexin Ma, Yikang Li, et al. Uniseg: A unified multi-modal li- dar segmentation network and the openpcseg codebase. In 23 IEEE/CVF International Conference on Computer Vision , pag...

    2023

  42. [49]

    Seg- ment any point cloud sequences by distilling vision founda- tion models

    Youquan Liu, Lingdong Kong, Jun Cen, Runnan Chen, Wenwei Zhang, Liang Pan, Kai Chen, and Ziwei Liu. Seg- ment any point cloud sequences by distilling vision founda- tion models. In Advances in Neural Information Processing Systems, pages 37193–37229, 2023. 1, 2, 3, 5, 6, 7, 8, 14, 15

    2023

  43. [50]

    Petrv2: A unified framework for 3d perception from multi-camera images

    Yingfei Liu, Junjie Yan, Fan Jia, Shuailin Li, Aqi Gao, Tiancai Wang, and Xiangyu Zhang. Petrv2: A unified framework for 3d perception from multi-camera images. In IEEE/CVF International Conference on Computer Vision , pages 3262–3272, 2023. 2, 4

    2023

  44. [51]

    Multi- space alignments towards universal lidar segmentation

    Youquan Liu, Lingdong Kong, Xiaoyang Wu, Runnan Chen, Xin Li, Liang Pan, Ziwei Liu, and Yuexin Ma. Multi- space alignments towards universal lidar segmentation. In IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 14648–14661, 2024. 1

    2024

  45. [52]

    Learning from 2d: Contrastive pixel-to-point knowledge transfer for 3d pretraining

    Yueh-Cheng Liu, Yu-Kai Huang, Hung-Yueh Chiang, Hung-Ting Su, Zhe-Yu Liu, Chin-Tang Chen, Ching-Yu Tseng, and Winston H Hsu. Learning from 2d: Contrastive pixel-to-point knowledge transfer for 3d pretraining. arXiv preprint arXiv:2104.04687, 2021. 3, 12, 14, 15

    2021 arXiv

  46. [53]

    Decoupled weight decay regularization

    Ilya Loshchilov and Frank Hutter. Decoupled weight decay regularization. arXiv preprint arXiv:1711.05101, 2017. 5, 10

    2017 arXiv

  47. [54]

    Self- supervised image-to-point distillation via semantically tol- erant contrastive loss

    Anas Mahmoud, Jordan SK Hu, Tianshu Kuai, Ali Harakeh, Liam Paull, and Steven L Waslander. Self- supervised image-to-point distillation via semantically tol- erant contrastive loss. In IEEE/CVF Conference on Com- puter Vision and Pattern Recognition , pages 7102–7110,

  48. [55]

    Towards a weakly supervised framework for 3d point cloud object detection and annotation

    Qinghao Meng, Wenguan Wang, Tianfei Zhou, Jianbing Shen, Yunde Jia, and Luc Van Gool. Towards a weakly supervised framework for 3d point cloud object detection and annotation. IEEE Transactions on Pattern Analysis and Machine Intelligence, 44(8):4454–4468, 2021. 2

    2021

  49. [56]

    Rangenet++: Fast and accurate lidar semantic segmentation

    Andres Milioto, Ignacio Vizzo, Jens Behley, and Cyrill Stachniss. Rangenet++: Fast and accurate lidar semantic segmentation. In IEEE/RSJ International Conference on Intelligent Robots and Systems, pages 4213–4220, 2019. 2

    2019

  50. [57]

    Tempo- ral consistent 3d lidar representation learning for semantic perception in autonomous driving

    Lucas Nunes, Louis Wiesmann, Rodrigo Marcuzzi, Xieyuanli Chen, Jens Behley, and Cyrill Stachniss. Tempo- ral consistent 3d lidar representation learning for semantic perception in autonomous driving. In IEEE/CVF Confer- ence on Computer Vision and Pattern Recognition , pages 5...

    2023

  51. [58]

    Dinov2: Learning robust visual features without supervi- sion

    Maxime Oquab, Timoth ´ee Darcet, Th ´eo Moutakanni, Huy V o, Marc Szafraniec, Vasil Khalidov, Pierre Fernandez, Daniel Haziza, Francisco Massa, Alaaeldin El-Nouby, et al. Dinov2: Learning robust visual features without supervi- sion. arXiv preprint arXiv:2304.07193, 2023. 5

    2023 arXiv

  52. [59]

    Semanticposs: A point cloud dataset with large quantity of dynamic instances

    Yancheng Pan, Biao Gao, Jilin Mei, Sibo Geng, Chengkun Li, and Huijing Zhao. Semanticposs: A point cloud dataset with large quantity of dynamic instances. In IEEE Intelli- gent Vehicles Symposium, pages 687–693, 2020. 5, 6, 10

    2020

  53. [60]

    Unsupervised 3d point cloud representation learning by triangle constrained contrast for autonomous driving

    Bo Pang, Hongchi Xia, and Cewu Lu. Unsupervised 3d point cloud representation learning by triangle constrained contrast for autonomous driving. In IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 5229– 5239, 2023. 1, 3, 6, 7, 14, 15

    2023

  54. [61]

    Masked autoencoders for point cloud self-supervised learning

    Yatian Pang, Wenxiao Wang, Francis EH Tay, Wei Liu, Yonghong Tian, and Li Yuan. Masked autoencoders for point cloud self-supervised learning. In European Confer- ence on Computer Vision, pages 604–621, 2022. 2

    2022

  55. [62]

    Learning to adapt sam for segmenting cross- domain point clouds

    Xidong Peng, Runnan Chen, Feng Qiao, Lingdong Kong, Youquan Liu, Yujing Sun, Tai Wang, Xinge Zhu, and Yuexin Ma. Learning to adapt sam for segmenting cross- domain point clouds. InEuropean Conference on Computer Vision, pages 54–71, 2024. 2

    2024

  56. [63]

    Three pillars improving vision founda- tion model distillation for lidar

    Gilles Puy, Spyros Gidaris, Alexandre Boulch, Oriane Sim´eoni, Corentin Sautier, Patrick P ´erez, Andrei Bursuc, and Renaud Marlet. Three pillars improving vision founda- tion model distillation for lidar. In IEEE/CVF Conference on Computer Vision and Pattern Recognition, page...

    2024

  57. [64]

    Pointnet: Deep learning on point sets for 3d classification and segmentation

    Charles R Qi, Hao Su, Kaichun Mo, and Leonidas J Guibas. Pointnet: Deep learning on point sets for 3d classification and segmentation. In IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 652–660, 2017. 2

    2017

  58. [65]

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

    Charles Ruizhongtai Qi, Li Yi, Hao Su, and Leonidas J Guibas. Pointnet++: Deep hierarchical feature learning on point sets in a metric space. In Advances in Neural Infor- mation Processing Systems, pages 5105–5114, 2017. 2

    2017

  59. [66]

    Joint long-term prediction of human motion using a planning- based social force approach

    Andrey Rudenko, Luigi Palmieri, and Kai O Arras. Joint long-term prediction of human motion using a planning- based social force approach. In IEEE International Confer- ence on Robotics and Automation, pages 4571–4577, 2018. 2

    2018

  60. [67]

    Gipso: Geometrically informed propa- gation for online adaptation in 3d lidar segmentation

    Cristiano Saltori, Evgeny Krivosheev, St´ephane Lathuili´ere, Nicu Sebe, Fabio Galasso, Giuseppe Fiameni, Elisa Ricci, and Fabio Poiesi. Gipso: Geometrically informed propa- gation for online adaptation in 3d lidar segmentation. In European Conference on Computer Vision, pages...

  61. [68]

    Image-to-lidar self-supervised distillation for autonomous driving data

    Corentin Sautier, Gilles Puy, Spyros Gidaris, Alexandre Boulch, Andrei Bursuc, and Renaud Marlet. Image-to-lidar self-supervised distillation for autonomous driving data. In IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 9891–9901, 2022. 1, 2, 3, 5, 6, 7...

    2022

  62. [69]

    Bevcontrast: Self-supervision in bev space for automotive lidar point clouds

    Corentin Sautier, Gilles Puy, Alexandre Boulch, Renaud Marlet, and Vincent Lepetit. Bevcontrast: Self-supervision in bev space for automotive lidar point clouds. In Interna- tional Conference on 3D Vision, pages 559–568, 2024. 2, 14, 15

    2024

  63. [70]

    Pointr- cnn: 3d object proposal generation and detection from point cloud

    Shaoshuai Shi, Xiaogang Wang, and Hongsheng Li. Pointr- cnn: 3d object proposal generation and detection from point cloud. In IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 770–779, 2019. 2

    2019

  64. [71]

    Backward atten- tive fusing network with local aggregation classifier for 3d point cloud semantic segmentation

    Hui Shuai, Xiang Xu, and Qingshan Liu. Backward atten- tive fusing network with local aggregation classifier for 3d point cloud semantic segmentation. IEEE Transactions on Image Processing, 30:4973–4984, 2021. 2

    2021

  65. [72]

    Super-convergence: Very fast training of neural networks using large learn- 24 ing rates

    Leslie N Smith and Nicholay Topin. Super-convergence: Very fast training of neural networks using large learn- 24 ing rates. In Artificial Intelligence and Machine Learn- ing for Multi-Domain Operations Applications, pages 369– 386, 2019. 5, 10

    2019

  66. [73]

    An empirical study of training state-of-the-art lidar segmentation models

    Jiahao Sun, Chunmei Qing, Xiang Xu, Lingdong Kong, Youquan Liu, Li Li, Chenming Zhu, Jingwei Zhang, Zeqi Xiao, Runnan Chen, et al. An empirical study of training state-of-the-art lidar segmentation models. arXiv preprint arXiv:2405.14870, 2024. 1

    2024 arXiv

  67. [74]

    Scalability in perception for autonomous driving: Waymo open dataset

    Pei Sun, Henrik Kretzschmar, Xerxes Dotiwalla, Aure- lien Chouard, Vijaysai Patnaik, Paul Tsui, James Guo, Yin Zhou, Yuning Chai, Benjamin Caine, et al. Scalability in perception for autonomous driving: Waymo open dataset. In IEEE/CVF Conference on Computer Vision and Pattern ...

    2020

  68. [75]

    Searching efficient 3d architectures with sparse point-voxel convolution

    Haotian Tang, Zhijian Liu, Shengyu Zhao, Yujun Lin, Ji Lin, Hanrui Wang, and Song Han. Searching efficient 3d architectures with sparse point-voxel convolution. In Eu- ropean Conference on Computer Vision , pages 685–702,

  69. [76]

    Prediction-uncertainty- aware decision-making for autonomous vehicles

    Xiaolin Tang, Kai Yang, Hong Wang, Jiahang Wu, Yechen Qin, Wenhao Yu, and Dongpu Cao. Prediction-uncertainty- aware decision-making for autonomous vehicles. IEEE Transactions on Intelligent Vehicles , 7(4):849–862, 2022. 2

    2022

  70. [77]

    Kpconv: Flexible and deformable convolution for point clouds

    Hugues Thomas, Charles R Qi, Jean-Emmanuel Deschaud, Beatriz Marcotegui, Franc ¸ois Goulette, and Leonidas J Guibas. Kpconv: Flexible and deformable convolution for point clouds. In IEEE/CVF International Conference on Computer Vision, pages 6411–6420, 2019. 2

    2019

  71. [78]

    Scribble- supervised lidar semantic segmentation

    Ozan Unal, Dengxin Dai, and Luc Van Gool. Scribble- supervised lidar semantic segmentation. InIEEE/CVF Con- ference on Computer Vision and Pattern Recognition, pages 2697–2707, 2022. 2, 5, 6, 9

    2022

  72. [80]

    Pointlora: Low-rank adaptation with token selection for point cloud learning

    Song Wang, Xiaolu Liu, Lingdong Kong, Jianyun Xu, Chunyong Hu, Gongfan Fang, Wentong Li, Jianke Zhu, and Xinchao Wang. Pointlora: Low-rank adaptation with token selection for point cloud learning. In IEEE/CVF Confer- ence on Computer Vision and Pattern Recognition , pages 6605...

    2025

  73. [81]

    Nuc-net: Non-uniform cylindrical partition network for ef- ficient lidar semantic segmentation

    Xuzhi Wang, Wei Feng, Lingdong Kong, and Liang Wan. Nuc-net: Non-uniform cylindrical partition network for ef- ficient lidar semantic segmentation. IEEE Transactions on Circuits and Systems for Video Technology, 2025. 2

    2025

  74. [82]

    Semi-supervised 3d object detection with patchteacher and pillarmix

    Xiaopei Wu, Liang Peng, Liang Xie, Yuenan Hou, Bin- bin Lin, Xiaoshui Huang, Haifeng Liu, Deng Cai, and Wanli Ouyang. Semi-supervised 3d object detection with patchteacher and pillarmix. In AAAI Conference on Artifi- cial Intelligence, pages 6153–6161, 2024. 2

    2024

  75. [83]

    Spatiotemporal self-supervised learn- ing for point clouds in the wild

    Yanhao Wu, Tong Zhang, Wei Ke, Sabine S ¨usstrunk, and Mathieu Salzmann. Spatiotemporal self-supervised learn- ing for point clouds in the wild. In IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 5251– 5260, 2023. 2

    2023

  76. [84]

    Polarmix: A general data augmen- tation technique for lidar point clouds

    Aoran Xiao, Jiaxing Huang, Dayan Guan, Kaiwen Cui, Shi- jian Lu, and Ling Shao. Polarmix: A general data augmen- tation technique for lidar point clouds. In Advances in Neu- ral Information Processing Systems , pages 11035–11048,

  77. [85]

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

    Aoran Xiao, Jiaxing Huang, Dayan Guan, Fangneng Zhan, and Shijian Lu. Transfer learning from synthetic to real lidar point cloud for semantic segmentation. In AAAI Con- ference on Artificial Intelligence , pages 2795–2803, 2022. 5, 6, 10

    2022

  78. [86]

    Unsupervised point cloud rep- resentation learning with deep neural networks: A survey

    Aoran Xiao, Jiaxing Huang, Dayan Guan, Xiaoqin Zhang, Shijian Lu, and Ling Shao. Unsupervised point cloud rep- resentation learning with deep neural networks: A survey. IEEE Transactions on Pattern Analysis and Machine Intel- ligence, 45(9):11321–11339, 2023. 1

    2023

  79. [87]

    3d semantic segmenta- tion in the wild: Learning generalized models for adverse- condition point clouds

    Aoran Xiao, Jiaxing Huang, Weihao Xuan, Ruijie Ren, Kangcheng Liu, Dayan Guan, Abdulmotaleb El Saddik, Shijian Lu, and Eric P Xing. 3d semantic segmenta- tion in the wild: Learning generalized models for adverse- condition point clouds. In IEEE/CVF Conference on Com- puter Vis...

  80. [88]

    A survey of label-efficient deep learning for 3d point clouds

    Aoran Xiao, Xiaoqin Zhang, Ling Shao, and Shijian Lu. A survey of label-efficient deep learning for 3d point clouds. IEEE Transactions on Pattern Analysis and Machine Intel- ligence, 46(12):9139–9160, 2024. 1

    2024

  81. [89]

    Pointcontrast: Unsupervised pre- training for 3d point cloud understanding

    Saining Xie, Jiatao Gu, Demi Guo, Charles R Qi, Leonidas Guibas, and Or Litany. Pointcontrast: Unsupervised pre- training for 3d point cloud understanding. In European Conference on Computer Vision, pages 574–591, 2020. 2, 7, 14, 15

    2020

  82. [90]

    Are vlms ready for autonomous driving? an empirical study from the reliability, data, and metric per- spectives

    Shaoyuan Xie, Lingdong Kong, Yuhao Dong, Chonghao Sima, Wenwei Zhang, Qi Alfred Chen, Ziwei Liu, and Liang Pan. Are vlms ready for autonomous driving? an empirical study from the reliability, data, and metric per- spectives. arXiv preprint arXiv:2501.04003, 2025. 1

    2025 arXiv

  83. [91]

    Benchmarking and improving bird’s eye view perception robustness in au- tonomous driving

    Shaoyuan Xie, Lingdong Kong, Wenwei Zhang, Jiawei Ren, Liang Pan, Kai Chen, and Ziwei Liu. Benchmarking and improving bird’s eye view perception robustness in au- tonomous driving. IEEE Transactions on Pattern Analysis and Machine Intelligence, 47(5):3878–3894, 2025. 2

    2025

  84. [92]

    Squeeze- segv3: Spatially-adaptive convolution for efficient point- cloud segmentation

    Chenfeng Xu, Bichen Wu, Zining Wang, Wei Zhan, Peter Vajda, Kurt Keutzer, and Masayoshi Tomizuka. Squeeze- segv3: Spatially-adaptive convolution for efficient point- cloud segmentation. In European Conference on Computer Vision, pages 1–19, 2020. 2

    2020

  85. [93]

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

    Jianyun Xu, Ruixiang Zhang, Jian Dou, Yushi Zhu, Jie Sun, and Shiliang Pu. Rpvnet: A deep and efficient range-point- voxel fusion network for lidar point cloud segmentation. In IEEE/CVF International Conference on Computer Vision , pages 16024–16033, 2021. 2

    2021

  86. [94]

    Visual foun- dation models boost cross-modal unsupervised domain adaptation for 3d semantic segmentation

    Jingyi Xu, Weidong Yang, Lingdong Kong, Youquan Liu, Rui Zhang, Qingyuan Zhou, and Ben Fei. Visual foun- dation models boost cross-modal unsupervised domain adaptation for 3d semantic segmentation. arXiv preprint arXiv:2403.10001, 2024. 2

    2024 arXiv

  87. [95]

    Weakly supervised seman- tic point cloud segmentation: Towards 10x fewer labels

    Xun Xu and Gim Hee Lee. Weakly supervised seman- tic point cloud segmentation: Towards 10x fewer labels. 25 In IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 13706–13715, 2020. 2

    2020

  88. [96]

    4d contrastive superflows are dense 3d representation learners

    Xiang Xu, Lingdong Kong, Hui Shuai, Wenwei Zhang, Liang Pan, Kai Chen, Ziwei Liu, and Qingshan Liu. 4d contrastive superflows are dense 3d representation learners. In European Conference on Computer Vision, pages 58–80,

  89. [97]

    Frnet: Frustum-range networks for scalable lidar segmen- tation

    Xiang Xu, Lingdong Kong, Hui Shuai, and Qingshan Liu. Frnet: Frustum-range networks for scalable lidar segmen- tation. IEEE Transactions on Image Processing, 34:2173– 2186, 2025. 2, 8

    2025

  90. [98]

    Limoe: Mixture of lidar represen- tation learners from automotive scenes

    Xiang Xu, Lingdong Kong, Hui Shuai, Liang Pan, Ziwei Liu, and Qingshan Liu. Limoe: Mixture of lidar represen- tation learners from automotive scenes. In IEEE/CVF Con- ference on Computer Vision and Pattern Recognition, pages 27368–27379, 2025. 2

    2025

  91. [99]

    Su- perflow++: Enhanced spatiotemporal consistency for cross- modal data pretraining

    Xiang Xu, Lingdong Kong, Hui Shuai, Wenwei Zhang, Liang Pan, Kai Chen, Ziwei Liu, and Qingshan Liu. Su- perflow++: Enhanced spatiotemporal consistency for cross- modal data pretraining. arXiv preprint arXiv:2503.19912,

  92. [100]

    2dpass: 2d priors assisted semantic segmentation on lidar point clouds

    Xu Yan, Jiantao Gao, Chaoda Zheng, Chao Zheng, Ruimao Zhang, Shuguang Cui, and Zhen Li. 2dpass: 2d priors assisted semantic segmentation on lidar point clouds. In European Conference on Computer Vision, pages 677–695,

  93. [101]

    Second: Sparsely embedded convolutional detection

    Yan Yan, Yuxing Mao, and Bo Li. Second: Sparsely embedded convolutional detection. Sensors, 18(10):3337,

  94. [102]

    Center-based 3d object detection and tracking

    Tianwei Yin, Xingyi Zhou, and Philipp Krahenbuhl. Center-based 3d object detection and tracking. In IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 11784–11793, 2021. 6, 7

    2021

  95. [103]

    Point-bert: Pre-training 3d point cloud transformers with masked point modeling

    Xumin Yu, Lulu Tang, Yongming Rao, Tiejun Huang, Jie Zhou, and Jiwen Lu. Point-bert: Pre-training 3d point cloud transformers with masked point modeling. In IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 19313–19322, 2022. 2

    2022

  96. [104]

    Hvdistill: Transferring knowledge from images to point clouds via unsupervised hybrid-view distillation

    Sha Zhang, Jiajun Deng, Lei Bai, Houqiang Li, Wanli Ouyang, and Yanyong Zhang. Hvdistill: Transferring knowledge from images to point clouds via unsupervised hybrid-view distillation. International Journal of Computer Vision, pages 1–15, 2024. 1, 2, 6, 14, 15

    2024

  97. [105]

    Polarnet: An improved grid representation for online lidar point clouds semantic segmentation

    Yang Zhang, Zixiang Zhou, Philip David, Xiangyu Yue, Ze- rong Xi, Boqing Gong, and Hassan Foroosh. Polarnet: An improved grid representation for online lidar point clouds semantic segmentation. In IEEE/CVF Conference on Com- puter Vision and Pattern Recognition , pages 9601–9610,

  98. [106]

    Self-supervised pretraining of 3d features on any point-cloud

    Zaiwei Zhang, Rohit Girdhar, Armand Joulin, and Ishan Misra. Self-supervised pretraining of 3d features on any point-cloud. In IEEE/CVF International Conference on Computer Vision, pages 10252–10263, 2021. 14, 15

    2021

  99. [107]

    Point cloud pre-training with diffusion models

    Xiao Zheng, Xiaoshui Huang, Guofeng Mei, Yuenan Hou, Zhaoyang Lyu, Bo Dai, Wanli Ouyang, and Yongshun Gong. Point cloud pre-training with diffusion models. In IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 22935–22945, 2024. 2

    2024

  100. [108]

    V oxelnet: End-to-end learning for point cloud based 3d object detection

    Yin Zhou and Oncel Tuzel. V oxelnet: End-to-end learning for point cloud based 3d object detection. In IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 4490–4499, 2018. 5, 7

    2018

  101. [109]

    Panoptic- polarnet: Proposal-free lidar point cloud panoptic segmen- tation

    Zixiang Zhou, Yang Zhang, and Hassan Foroosh. Panoptic- polarnet: Proposal-free lidar point cloud panoptic segmen- tation. In IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 13194–13203, 2021. 2

    2021

  102. [110]

    Cylindrical and asymmetrical 3d convolution networks for lidar segmenta- tion

    Xinge Zhu, Hui Zhou, Tai Wang, Fangzhou Hong, Yuexin Ma, Wei Li, Hongsheng Li, and Dahua Lin. Cylindrical and asymmetrical 3d convolution networks for lidar segmenta- tion. In IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 9939–9948, 2021. 2

    2021

  103. [111]

    Perception-aware multi- sensor fusion for 3d lidar semantic segmentation

    Zhuangwei Zhuang, Rong Li, Kui Jia, Qicheng Wang, Yuanqing Li, and Mingkui Tan. Perception-aware multi- sensor fusion for 3d lidar semantic segmentation. In IEEE/CVF International Conference on Computer Vision , pages 16280–16290, 2021. 2 26

    2021

  104. [2024]

    1, 2, 3, 6, 7, 8, 14, 15

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.