pith. sign in

arxiv: 2605.01466 · v2 · pith:A7XYV4UZnew · submitted 2026-05-02 · 💻 cs.CV · cs.LG

SplAttN: Bridging 2D and 3D with Gaussian Soft Splatting and Attention for Point Cloud Completion

Zhaoyang Li , Zhichao You , Tianrui Li This is my paper

Pith reviewed 2026-05-22 09:51 UTC · model grok-4.3

classification 💻 cs.CV cs.LG
keywords point cloud completionmulti-modal learninggaussian splattingcross-modal connectiondifferentiable projectionattention mechanismshape completion
0
0 comments X

The pith

Differentiable Gaussian splatting replaces hard projection to prevent cross-modal entropy collapse and enable real use of visual cues in point cloud completion.

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

The paper shows that standard hard projection of sparse point clouds onto image planes produces extremely sparse support, blocking the flow of visual priors and creating a failure mode called Cross-Modal Entropy Collapse. SplAttN replaces this with differentiable Gaussian splatting to generate dense, continuous image-plane representations that support gradient flow and better cross-modal learning. Experiments establish state-of-the-art results on PCN and ShapeNet-55/34, and counter-factual tests on KITTI confirm that the model continues to rely on visual input while baselines fall back to unimodal template retrieval.

Core claim

SplAttN identifies Cross-Modal Entropy Collapse as the result of hard projection severing modality connections, then addresses it by reformulating projection as continuous density estimation with differentiable Gaussian splatting, which produces dense support, improves learnability of visual priors, and yields an effective cross-modal connection validated by maintained performance dependence on image cues under counterfactual removal on real-world data.

What carries the argument

Differentiable Gaussian Splatting reformulated as continuous density estimation to produce dense image-plane representations from sparse point clouds, enabling visual prior propagation through the attention and completion pipeline.

If this is right

  • State-of-the-art completion accuracy on the PCN and ShapeNet-55/34 benchmarks.
  • Robust reliance on visual cues shown by counter-factual evaluation on KITTI, where baselines degrade into unimodal retrievers.
  • Improved gradient flow and cross-modal connection learnability from the dense continuous representation.
  • Avoidance of collapsed sparse support that otherwise hinders visual prior propagation.

Where Pith is reading between the lines

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

  • The same soft-splatting replacement could be tested on other sparse-to-dense fusion tasks such as multi-view 3D reconstruction or sensor fusion for robotics.
  • If the dense support already supplies most of the connection benefit, the attention layers might be simplified without loss of performance.
  • Real-world deployment in settings with partial image occlusion would likely show larger gains for SplAttN than for hard-projection baselines.

Load-bearing premise

The main barrier to multi-modal benefits is the sparse support and entropy collapse from hard projection, and differentiable Gaussian splatting removes that barrier without introducing new confounding effects in attention or completion.

What would settle it

A controlled test that removes or masks the visual input on KITTI samples and checks whether SplAttN performance drops substantially more than baselines, or an ablation that swaps Gaussian splatting back to hard projection and measures the resulting drop in both accuracy and visual dependence.

Figures

Figures reproduced from arXiv: 2605.01466 by Tianrui Li, Zhaoyang Li, Zhichao You.

Figure 1
Figure 1. Figure 1: The overall architecture of our proposed SplAttN. The pipeline consists of two integral stages. (a) Dual-Branch Feature Extraction. The GS-Bridge branch extracts comprehensive global representations by using geometric tokens Fgeo to actively query visual features Fvis derived from Gaussian Soft Splatting. In parallel, the Local Encoder captures topology-aware local details Fl through an EdgeConv module fol… view at source ↗
Figure 2
Figure 2. Figure 2: Visualizing the Alignment Gap. Top (Hard Projec￾tion): Hard projection suffers from sparsity and overlap, leading to high divergence from the true manifold. Bottom (Splatting): Our method generates a continuous density field, effectively predicting local features for empty regions and smoothing out overlap noise. continuous spatial query variable v ∈ Ω within the visual domain. Standard methods typically m… view at source ↗
Figure 3
Figure 3. Figure 3: Detailed architecture of the Gaussian Splatting Bridge (GS-Bridge). It illustrates how the geometric stream interacts with the visual stream through Differentiable Gaussian Splatting to perform density estimation. This strictly expands the effective information support Ssof t = S p {v | ∥v − π(p)∥ < 3σ}. By the subadditiv￾ity of measures, we guarantee positive information capacity: µ(Ssof t) ≥ µ(Shard) +X … view at source ↗
Figure 4
Figure 4. Figure 4: Architecture of the Global-Local Decoder. The decoder combines global priors with local details. It employs structure-aware attention to query local geometric primitives from the Hybrid Tokenizer for coordinate refinement. feature resolution and regress a continuous displacement field ψ : Pk → Pk+1. The predicted coordinate offsets ∆P project the coarse approximation onto the high-fidelity manifold via res… view at source ↗
Figure 5
Figure 5. Figure 5: Visual comparison on the PCN dataset. Compared with state-of-the-art methods, SplAttN recovers more faithful global topology and finer local details, particularly in thin structures like chair legs, verifying the effectiveness of our Hybrid Local Encoder. 4. Experiment 4.1. Datasets and Metrics We evaluate SplAttN on three standard benchmarks: PCN, ShapeNet-55/34, and KITTI. PCN Dataset (Yuan et al., 2018)… view at source ↗
Figure 6
Figure 6. Figure 6: Qualitative comparison on ShapeNet-55. SplAttN generates more complete and detailed shapes compared to the former baselines across diverse categories. structurally precise reconstruction. Rather than viewing KITTI merely as a target for domain adaptation, we identify a unique opportunity within its distributional irregularities and intrinsic data imperfections. We argue that the intrin￾sic artifacts of rea… view at source ↗
Figure 8
Figure 8. Figure 8: Verification of Multi-Modal Dependency. We com￾pare SCS sensitivity against Cross-Modal Information Throughput (CMIT). Unlike baselines with low CMIT showing negligible sen￾sitivity, SplAttN achieves a dominant CMIT of 200.5. This high throughput strictly correlates with a substantial consistency drop upon visual removal, confirming a valid cross-modal dependency rather than template retrieval. (−26.1%) wh… view at source ↗
Figure 7
Figure 7. Figure 7: Distributional Discrepancy. Visual comparison of (a) 3D density and (b) 2D projections between PCN and KITTI. The stark contrast reveals a fundamental topological gap, challenging the validity of standard normalization-based evaluation protocols. geometric memorization, we design a systematic counterfac￾tual evaluation protocol. We employ the Semantic Consis￾tency Score (SCS) as a measure of recognizabilit… view at source ↗
Figure 9
Figure 9. Figure 9: Qualitative Results on ShapeNet-55 (Easy Difficulty). Comparisons of reconstruction quality on representative samples. SplAttN faithfully recovers details that are blurred by baselines. 12 view at source ↗
Figure 10
Figure 10. Figure 10: Qualitative Results on ShapeNet-55 (Median Difficulty). Comparisons of reconstruction quality on representative samples. SplAttN faithfully recovers details that are blurred by baselines. 13 view at source ↗
Figure 11
Figure 11. Figure 11: Qualitative Results on ShapeNet-55 (Hard Difficulty). Comparisons on challenging samples with significant missing geometry. Our method maintains structural integrity and input fidelity better than competitors. 14 view at source ↗
Figure 12
Figure 12. Figure 12: Entropy Analysis - Sample 1. Our method produces dense feature maps compared to sparse baselines. 20 view at source ↗
Figure 13
Figure 13. Figure 13: Entropy Analysis - Sample 2. Histogram analysis demonstrates the broader value distribution of our method. 21 view at source ↗
Figure 14
Figure 14. Figure 14: Qualitative Results on KITTI. Comparisons of point cloud completion on real-world scans. The visual differences across methods are consistent with the rankings produced by our Semantic Consistency Score (SCS) metric. H. Additional KITTI Robustness Analysis To further investigate the performance trade-off discussed in the main text, we provide a detailed visualization of the intermediate feature representa… view at source ↗
Figure 15
Figure 15. Figure 15: KITTI Robustness - Sample 1. Three-view feature comparison under sim-to-real domain shift. 23 view at source ↗
Figure 16
Figure 16. Figure 16: KITTI Robustness - Sample 2. Visualization of point cloud projection and feature map coverage. 24 view at source ↗
read the original abstract

Although multi-modal learning has advanced point cloud completion, the theoretical mechanisms remain unclear. Recent works attribute success to the connection between modalities, yet we identify that standard hard projection severs this connection: projecting a sparse point cloud onto the image plane yields an extremely sparse support, which hinders visual prior propagation, a failure mode we term Cross-Modal Entropy Collapse. To address this practical limitation, we propose SplAttN, which replaces hard projection with Differentiable Gaussian Splatting to produce a dense, continuous image-plane representation. By reformulating projection as continuous density estimation, SplAttN avoids collapsed sparse support, facilitates gradient flow, and improves cross-modal connection learnability. Extensive experiments show that SplAttN achieves state-of-the-art performance on PCN and ShapeNet-55/34. Crucially, we utilize the real-world KITTI benchmark as a stress test for multi-modal reliance. Counter-factual evaluation reveals that while baselines degenerate into unimodal template retrievers insensitive to visual removal, SplAttN maintains a robust dependency on visual cues, validating that our method establishes an effective cross-modal connection. Code is available at https://github.com/zay002/SplAttN.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

1 major / 2 minor

Summary. The manuscript identifies Cross-Modal Entropy Collapse as a failure mode arising from hard projection of sparse point clouds onto image planes, which produces extremely sparse support and severs visual prior propagation. SplAttN replaces this with Differentiable Gaussian Splatting to generate dense continuous image-plane representations, combined with an attention pipeline, to improve cross-modal learnability and gradient flow. It reports state-of-the-art results on PCN and ShapeNet-55/34, and uses a counter-factual evaluation on the real-world KITTI benchmark showing that SplAttN retains visual dependency while baselines collapse to unimodal template retrieval.

Significance. If the central claims hold, the work offers a concrete engineering response to a practical barrier in multi-modal point cloud completion by reformulating projection as continuous density estimation. Code availability and the use of KITTI as a stress test for modality reliance are strengths. The significance hinges on whether the observed robustness is causally tied to the splatting change rather than ancillary architectural modifications.

major comments (1)
  1. [Experiments (KITTI counter-factual)] KITTI counter-factual evaluation: the reported robustness of SplAttN to visual cue removal is presented as evidence of an effective cross-modal connection established by differentiable Gaussian splatting. However, SplAttN also introduces a new attention pipeline over the dense splatted features. Without an ablation that holds the attention module and overall capacity fixed while swapping only hard projection versus Gaussian soft splatting, the causal attribution to the projection reformulation remains under-supported and the stress-test result cannot isolate the claimed mechanism.
minor comments (2)
  1. [Abstract] The abstract states that SplAttN achieves SOTA on PCN and ShapeNet-55/34 but supplies no quantitative metrics, dataset splits, or baseline comparisons; a brief summary of key numbers would improve readability.
  2. [Introduction / Method] The term 'Cross-Modal Entropy Collapse' is introduced as a new failure mode; a short formal definition or entropy calculation in the method section would clarify its relation to standard projection sparsity.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the constructive and insightful review. The feedback on isolating the contribution of differentiable Gaussian splatting in the KITTI counter-factual evaluation is well-taken, and we address it directly below.

read point-by-point responses

  1. Referee: [Experiments (KITTI counter-factual)] KITTI counter-factual evaluation: the reported robustness of SplAttN to visual cue removal is presented as evidence of an effective cross-modal connection established by differentiable Gaussian splatting. However, SplAttN also introduces a new attention pipeline over the dense splatted features. Without an ablation that holds the attention module and overall capacity fixed while swapping only hard projection versus Gaussian soft splatting, the causal attribution to the projection reformulation remains under-supported and the stress-test result cannot isolate the claimed mechanism.

    Authors: We agree that a controlled ablation isolating only the projection reformulation—while holding the attention module, overall capacity, and other architectural elements fixed—would provide stronger causal evidence for the role of differentiable Gaussian splatting in the observed robustness on KITTI. The attention pipeline is designed to operate on the dense continuous features produced by splatting, so the components are interdependent by design; however, this does not obviate the need for the requested isolation experiment. In the revised manuscript we will add this specific ablation to the KITTI counter-factual section, directly comparing hard projection versus Gaussian soft splatting under an otherwise identical attention-equipped architecture. This addition will clarify the mechanism and address the referee’s concern about ancillary modifications. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical engineering response with independent counter-factual validation

full rationale

The paper identifies Cross-Modal Entropy Collapse as a practical failure mode of hard projection and proposes Differentiable Gaussian Splatting plus attention as a direct engineering fix to produce dense continuous representations and better gradient flow. No equations, derivations, or fitted parameters are presented that reduce the claimed cross-modal benefit to a self-referential definition or input by construction. The KITTI counter-factual evaluation (performance drop under visual removal) constitutes independent empirical evidence rather than a statistical tautology or self-citation load-bearing step. The method is self-contained against external benchmarks (PCN, ShapeNet, KITTI) with no uniqueness theorems, ansatzes smuggled via prior self-work, or renaming of known results as new derivations.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

Review is limited to the abstract; no explicit free parameters, detailed axioms, or invented entities beyond the named failure mode are described.

axioms (1)
  • domain assumption Differentiable Gaussian splatting produces a dense continuous image-plane representation from sparse 3D points that facilitates gradient flow
    This is the core technical premise invoked to solve the identified projection problem.
invented entities (1)
  • Cross-Modal Entropy Collapse no independent evidence
    purpose: Term for the failure mode in which hard projection creates extremely sparse support that hinders visual prior propagation
    Newly introduced concept used to motivate the method

pith-pipeline@v0.9.0 · 5750 in / 1422 out tokens · 44806 ms · 2026-05-22T09:51:28.453688+00:00 · methodology

Review history (2 revisions) →

discussion (0)

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

Lean theorems connected to this paper

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

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

Reference graph

Works this paper leans on

63 extracted references · 63 canonical work pages · 3 internal anchors

  1. [1]

    2018 international conference on 3D vision (3DV) , pages=

    Pcn: Point completion network , author=. 2018 international conference on 3D vision (3DV) , pages=. 2018 , organization=

    work page 2018

  2. [2]

    Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

    Foldingnet: Point cloud auto-encoder via deep grid deformation , author=. Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

    work page

  3. [3]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Topnet: Structural point cloud decoder , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    work page

  4. [4]

    European conference on computer vision , pages=

    Grnet: Gridding residual network for dense point cloud completion , author=. European conference on computer vision , pages=. 2020 , organization=

    work page 2020

  5. [5]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Pmp-net: Point cloud completion by learning multi-step point moving paths , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    work page

  6. [6]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Lake-net: Topology-aware point cloud completion by localizing aligned keypoints , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    work page

  7. [7]

    Proceedings of the AAAI Conference on artificial intelligence , volume=

    Pointattn: You only need attention for point cloud completion , author=. Proceedings of the AAAI Conference on artificial intelligence , volume=

    work page

  8. [8]

    Wang, Xiaogang and , Marcelo H. Ang Jr. and Lee, Gim Hee , title =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , month =

    work page

  9. [9]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Cascaded refinement network for point cloud completion , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    work page

  10. [10]

    Proceedings of the Thirty-Fourth International Joint Conference on Artificial Intelligence , pages=

    Multi-modal point cloud completion with interleaved attention enhanced transformer , author=. Proceedings of the Thirty-Fourth International Joint Conference on Artificial Intelligence , pages=

    work page

  11. [11]

    European conference on computer vision , pages=

    Detail preserved point cloud completion via separated feature aggregation , author=. European conference on computer vision , pages=. 2020 , organization=

    work page 2020

  12. [12]

    Proceedings of the IEEE/CVF international conference on computer vision , pages=

    Pointr: Diverse point cloud completion with geometry-aware transformers , author=. Proceedings of the IEEE/CVF international conference on computer vision , pages=

    work page

  13. [13]

    Proceedings of the IEEE/CVF international conference on computer vision , pages=

    Snowflakenet: Point cloud completion by snowflake point deconvolution with skip-transformer , author=. Proceedings of the IEEE/CVF international conference on computer vision , pages=

    work page

  14. [14]

    IEEE Transactions on Pattern Analysis and Machine Intelligence , volume=

    PMP-Net++: Point cloud completion by transformer-enhanced multi-step point moving paths , author=. IEEE Transactions on Pattern Analysis and Machine Intelligence , volume=. 2022 , publisher=

    work page 2022

  15. [15]

    Proceedings of the IEEE/CVF international conference on computer vision , pages=

    Pointclip v2: Prompting clip and gpt for powerful 3d open-world learning , author=. Proceedings of the IEEE/CVF international conference on computer vision , pages=

    work page

  16. [16]

    European conference on computer vision , pages=

    Seedformer: Patch seeds based point cloud completion with upsample transformer , author=. European conference on computer vision , pages=. 2022 , organization=

    work page 2022

  17. [17]

    IEEE Trans

    Yu, Xumin and Rao, Yongming and Wang, Ziyi and Lu, Jiwen and Zhou, Jie , title =. IEEE Trans. Pattern Anal. Mach. Intell. , month = dec, pages =. 2023 , issue_date =. doi:10.1109/TPAMI.2023.3309253 , abstract =

    work page doi:10.1109/tpami.2023.3309253 2023

  18. [18]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Anchorformer: Point cloud completion from discriminative nodes , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    work page

  19. [19]

    Proceedings of the IEEE/CVF international conference on computer vision , pages=

    Hyperbolic chamfer distance for point cloud completion , author=. Proceedings of the IEEE/CVF international conference on computer vision , pages=

    work page

  20. [20]

    Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

    Svdformer: Complementing point cloud via self-view augmentation and self-structure dual-generator , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

    work page

  21. [21]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    SymmCompletion: High-Fidelity and High-Consistency Point Cloud Completion with Symmetry Guidance , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    work page

  22. [22]

    Forty-second International Conference on Machine Learning , year=

    Unpaired Point Cloud Completion via Unbalanced Optimal Transport , author=. Forty-second International Conference on Machine Learning , year=

    work page

  23. [23]

    Advances in Neural Information Processing Systems , volume=

    Cross-modal learning for image-guided point cloud shape completion , author=. Advances in Neural Information Processing Systems , volume=

    work page

  24. [24]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Pulsar: Efficient sphere-based neural rendering , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    work page

  25. [25]

    An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale

    An image is worth 16x16 words: Transformers for image recognition at scale , author=. arXiv preprint arXiv:2010.11929 , year=

    work page internal anchor Pith review Pith/arXiv arXiv 2010

  26. [26]

    Proceedings of the IEEE/CVF international conference on computer vision , pages=

    Swin transformer: Hierarchical vision transformer using shifted windows , author=. Proceedings of the IEEE/CVF international conference on computer vision , pages=

    work page

  27. [27]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Masked autoencoders are scalable vision learners , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    work page

  28. [28]

    Advances in neural information processing systems , volume=

    Pointnet++: Deep hierarchical feature learning on point sets in a metric space , author=. Advances in neural information processing systems , volume=

    work page

  29. [29]

    International conference on machine learning , pages=

    Mutual information neural estimation , author=. International conference on machine learning , pages=. 2018 , organization=

    work page 2018

  30. [30]

    Representation Learning with Contrastive Predictive Coding

    Representation learning with contrastive predictive coding , author=. arXiv preprint arXiv:1807.03748 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  31. [31]

    Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

    Deep residual learning for image recognition , author=. Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

    work page

  32. [32]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    DC-PCN: Point Cloud Completion Network with Dual-Codebook Guided Quantization , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    work page

  33. [33]

    Differentiable surface splatting for point-based geometry processing , year =

    Yifan, Wang and Serena, Felice and Wu, Shihao and \". Differentiable surface splatting for point-based geometry processing , year =. ACM Trans. Graph. , month = nov, articleno =. doi:10.1145/3355089.3356513 , abstract =

    work page doi:10.1145/3355089.3356513

  34. [34]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Softmax splatting for video frame interpolation , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    work page

  35. [35]

    Proceedings of the 32nd ACM International Conference on Multimedia , pages=

    Geoformer: Learning point cloud completion with tri-plane integrated transformer , author=. Proceedings of the 32nd ACM International Conference on Multimedia , pages=

    work page

  36. [36]

    and Yuille, Alan and Tan, Mingxing , title =

    Li, Yingwei and Yu, Adams Wei and Meng, Tianjian and Caine, Ben and Ngiam, Jiquan and Peng, Daiyi and Shen, Junyang and Lu, Yifeng and Zhou, Denny and Le, Quoc V. and Yuille, Alan and Tan, Mingxing , title =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , month =. 2022 , pages =

    work page 2022

  37. [37]

    European conference on computer vision , pages=

    Tinyvit: Fast pretraining distillation for small vision transformers , author=. European conference on computer vision , pages=. 2022 , organization=

    work page 2022

  38. [38]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , month =

    Zhang, Xuancheng and Feng, Yutong and Li, Siqi and Zou, Changqing and Wan, Hai and Zhao, Xibin and Guo, Yandong and Gao, Yue , title =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , month =. 2021 , pages =

    work page 2021

  39. [39]

    ACM Trans

    Li, Yixuan and Ma, Lipeng and Yang, Weidong and Fei, Ben , title =. ACM Trans. Multimedia Comput. Commun. Appl. , month = nov, keywords =. 2025 , publisher =. doi:10.1145/3774887 , abstract =

    work page doi:10.1145/3774887 2025

  40. [40]

    The international journal of robotics research , volume=

    Vision meets robotics: The kitti dataset , author=. The international journal of robotics research , volume=. 2013 , publisher=

    work page 2013

  41. [41]

    Decoupled Weight Decay Regularization

    Decoupled weight decay regularization , author=. arXiv preprint arXiv:1711.05101 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  42. [42]

    2017 , eprint=

    Cyclical Learning Rates for Training Neural Networks , author=. 2017 , eprint=

    work page 2017

  43. [43]

    , author=

    3D Gaussian splatting for real-time radiance field rendering. , author=. ACM Trans. Graph. , volume=

    work page

  44. [44]

    and Gui, Liang-Yan , title =

    Cheng, Yen-Chi and Lee, Hsin-Ying and Tulyakov, Sergey and Schwing, Alexander G. and Gui, Liang-Yan , title =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , month =. 2023 , pages =

    work page 2023

  45. [45]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , month =

    Melas-Kyriazi, Luke and Rupprecht, Christian and Vedaldi, Andrea , title =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , month =. 2023 , pages =

    work page 2023

  46. [46]

    Proceedings of the 29th ACM international conference on multimedia , pages=

    Asfm-net: Asymmetrical siamese feature matching network for point completion , author=. Proceedings of the 29th ACM international conference on multimedia , pages=

    work page

  47. [47]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    Cra-pcn: Point cloud completion with intra-and inter-level cross-resolution transformers , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    work page

  48. [48]

    Computational visual media , volume=

    Pct: Point cloud transformer , author=. Computational visual media , volume=. 2021 , publisher=

    work page 2021

  49. [49]

    ACM Transactions on Graphics (tog) , volume=

    Dynamic graph cnn for learning on point clouds , author=. ACM Transactions on Graphics (tog) , volume=. 2019 , publisher=

    work page 2019

  50. [50]

    Advances in neural information processing systems , volume=

    Attention is all you need , author=. Advances in neural information processing systems , volume=

    work page

  51. [51]

    Intelligence & Robotics , VOLUME =

    Dingchen Yang and Bowen Cao and Sanqing Qu and Fan Lu and Shangding Gu and Guang Chen , TITLE =. Intelligence & Robotics , VOLUME =. 2025 , NUMBER =

    work page 2025

  52. [52]

    Intelligence & Robotics , VOLUME =

    Zhengyi Lu and Yunhong Liao and Jia Li , TITLE =. Intelligence & Robotics , VOLUME =. 2025 , NUMBER =

    work page 2025

  53. [53]

    Advances in neural information processing systems , volume=

    Learning representations by maximizing mutual information across views , author=. Advances in neural information processing systems , volume=

    work page

  54. [54]

    Advances in Neural Information Processing Systems , volume=

    Point cloud completion with pretrained text-to-image diffusion models , author=. Advances in Neural Information Processing Systems , volume=

    work page

  55. [55]

    Advances in Neural Information Processing Systems , volume=

    A theory of multimodal learning , author=. Advances in Neural Information Processing Systems , volume=

    work page

  56. [56]

    The Thirteenth International Conference on Learning Representations , year=

    SplatFormer: Point Transformer for Robust 3D Gaussian Splatting , author=. The Thirteenth International Conference on Learning Representations , year=

    work page

  57. [57]

    ACM SIGGRAPH 2024 conference papers , pages=

    2d gaussian splatting for geometrically accurate radiance fields , author=. ACM SIGGRAPH 2024 conference papers , pages=

    work page 2024

  58. [58]

    Proceedings of the 29th ACM International Conference on Multimedia , pages =

    Xia, Yaqi and Xia, Yan and Li, Wei and Song, Rui and Cao, Kailang and Stilla, Uwe , title =. Proceedings of the 29th ACM International Conference on Multimedia , pages =. 2021 , isbn =. doi:10.1145/3474085.3475348 , abstract =

    work page doi:10.1145/3474085.3475348 2021

  59. [59]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    View-guided point cloud completion , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    work page

  60. [60]

    Proceedings of the IEEE/CVF international conference on computer vision , pages=

    Point transformer , author=. Proceedings of the IEEE/CVF international conference on computer vision , pages=

    work page

  61. [61]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Variational relational point completion network , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    work page

  62. [62]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Proxyformer: Proxy alignment assisted point cloud completion with missing part sensitive transformer , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    work page

  63. [63]

    European Conference on Computer Vision , pages=

    Fbnet: Feedback network for point cloud completion , author=. European Conference on Computer Vision , pages=

    work page