Pith. sign in

REVIEW 2 major objections 5 minor 61 references

RefComp: A Reference-guided Unified Framework for Unpaired Point Cloud Completion

T0 review · 2 major / 5 minor · reviewed 2026-08-16 · deepseek-v4-flash

Pith's one-line read RefComp claims to be the first unified framework for class-agnostic unpaired point cloud completion, reformulating completion as reference-guided shape translation in latent space.

desk verdict A solid reference-guided completion framework whose headline class-agnostic claim is undercut by test-time same-class retrieval; worth refereeing but needs a label-free evaluation or a revised claim. read the letter →

arxiv 2504.13788 v1 pith:U3FN6AZE submitted 2025-04-18 cs.CV

classification cs.CV
keywords unpairedpointcloudcompletionclass-agnosticreference-guidedshapetranslationlatentfusionretrieval-basedChamferdistance3D
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

RefComp tries to establish that unpaired point cloud completion—filling in partial 3D scans without paired ground-truth data—does not need one model per object class. For each partial input, the framework retrieves a complete shape of the same class from a multi-class reference database, degrades that complete shape into a partial-complete pair, and uses the pair to guide completion in a shared latent space. The paper reports that this design is the first unified class-agnostic unpaired completion method, achieves state-of-the-art results in the class-aware training setting, and stays competitive in the class-agnostic setting on virtual-scan and real-world benchmarks. If this holds, a single trained model could replace per-class completion systems and remove the need to identify an object's class before completing it.

What carries the argument

The load-bearing component is the Latent Shape Fusion Module (LSFM), a feature-fusion network with a reference branch and a target branch that share every parameter. It takes the encoded partial feature $z_{p_y}$ (or $z_{p_x}$) and a mask feature $z_{m_y}$, which is the embedding of the region missing from the reference partial (complete $c_y$ minus partial $p_y$), and produces completed features through residual blocks, concatenation, and element-wise additions. The mask is the carrier of the structural information: because the reference partial was corrupted using the target as a template, the mask tells the network where the target is likely missing points. A Wasserstein-distance loss on the reference branch pulls both completed feature codes into the complete-latent space, while Chamfer-Distance losses on the degraded reconstruction of the target and on the reconstruction of the reference supervise the decoder.

What would settle it

Hold out all complete shapes of one class, say cabinets, from the reference database while training the class-agnostic model, then test on partial cabinets from that class. If completion quality on the held-out class falls to the level of a random-reference baseline, the class-agnostic claim is falsified; if it stays high, the model is learning a generic shape prior rather than depending on same-class retrieval.

Watch

Extended reading notes

Core claim

The central claim is that unpaired completion can be recast as a shape translation problem: move the latent code of a partial point cloud into the latent space of complete shapes, using retrieved reference pairs as the only supervision. A reference pair is built by taking a complete cloud $c_y$ from a multi-class database, corrupting it using the target partial $p_x$ as a template to produce $p_y$, and selecting the top-N pairs by Chamfer distance. The Latent Shape Fusion Module then fuses the partial features with the missing-region mask embedding and, because the reference and target branches share parameters, the target branch inherits the reference branch's completion ability; a Wasserstein-distance loss pulls the completed features into the complete-latent space. The authors report state-of-the-art average Chamfer Distance and F1 in class-aware training over eight classes, and a class-agnostic version ('RefComp Unified') that outperforms several class-aware baselines, including on real-world KITTI, ScanNet, and MatterPort3D data.

Load-bearing premise

The framework works only if, for every partial scan to be completed, the reference database contains a complete shape of the same object class that is geometrically similar enough to be retrieved by Chamfer distance; if no such reference exists, or a wrong class is retrieved, the mask encodes the wrong missing region and the completion guidance fails.

Editorial extensions

If this is right

  • If the class-agnostic claim holds, one model trained on a pooled multi-class database can be deployed on arbitrary partial scans, removing per-class training, per-class storage, and inference-time class identification.
  • The reported KITTI inference times of 0.02 to 0.04 seconds per object with a 40.5-million-parameter model put the pipeline in a range usable for autonomous-driving perception.
  • On real-world scans from ScanNet and MatterPort3D, the class-aware versions report lower MMD without fine-tuning than supervised baselines that were fine-tuned to those datasets, indicating that the reference data transfers to depth-sensor noise.
  • Ablation shows that removing parameter sharing between the reference and target branches raises average Chamfer Distance on the CRN subset from 14.4 to 31.1, identifying the shared branches as the main carrier of the completion signal.

Reading between the lines

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

  • The retrieval step selects same-class references by Chamfer distance, but the architecture never consumes class labels; a direct extension would drop the class filter and retrieve purely by geometry, testing whether similarity alone can carry class-agnostic completion on unlabeled scans.
  • Because the mask feature is computed as complete-minus-partial for the reference pair, the same fusion machinery could be driven at inference time by a learned occlusion estimator, making the framework applicable to arbitrary partiality patterns instead of KNN-corrupted ones.
  • The paper's ablation shows that randomly selecting among the top-3 retrieved references improves CD by about 2.0 over using one fixed reference, so retrieval diversity acts as a regularizer; a harder-negative or adversarial retrieval scheme is a natural next step.
  • A stress test implicit in the design is to grow the reference database over time, e.g., from a scanner's accumulated observations; if completion quality scales with database size, the framework would double as a lifelong completion system.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 5 minor

Summary. The paper proposes RefComp, a reference-guided framework for unpaired point cloud completion. The method builds partial-complete reference pairs by degrading complete shapes from a database using the target partial cloud as a corruption template, then selects the top-N pairs by Chamfer distance. A reference branch and a target branch share parameters; a Latent Shape Fusion Module (LSFM) fuses missing-region mask features into the partial latent code, and the decoder produces the completed cloud. Training losses include Chamfer reconstruction on the reference branch, degradation-consistency on the target branch, and a Wasserstein loss that aligns the completed latent features with the complete latent space. The paper reports class-aware and class-agnostic training variants ('RefComp' and 'RefComp Unified') and evaluates them on CRN, 3D-EPN, PartNet, KITTI, ScanNet, and MatterPort3D, with ablations on the reference-selection strategy, the LSFM module, and parameter sharing.

Significance. If the class-agnostic claim were fully supported, RefComp would be a practically valuable single-model solution for unpaired completion across many object classes, and the reference-guided latent translation formulation is a sensible way to inject shape priors without paired ground truth. The paper gives credit for a clear formulation, extensive comparisons, and ablations that show the parameter-sharing mechanism and the LSFM are important for the reported performance. However, the evaluation protocol does not currently support the headline 'class-agnostic' claim, because reference retrieval at test time uses the same-class oracle. The central technical machinery is plausible, but the gap between the claimed contribution and the experimental protocol needs to be closed before the paper can be accepted as stated.

major comments (2)
  1. [Section III-A and Section IV.C] The central 'class-agnostic' claim is not supported by the evaluation protocol. Section III-A states that reference pairs are selected by Chamfer distance between px and all py 'of the same class,' and Section IV.C states that at test time 'we only use the reference pair that is the closest to px of the same class.' Consequently, every 'RefComp Unified' entry in Tables I, II, IV, and V is obtained with a same-class oracle for retrieval; the model is not evaluated under the label-free conditions that the introduction uses to motivate class-agnostic completion. To make the claim stand, the authors should either (a) evaluate with a class-label-free retrieval protocol over a pooled multi-class reference database, (b) report both oracle and label-free numbers, and (c) discuss the case where no same-class reference similar to px exists in the database. As written, the method is best described as a single model trained across classes with class-conditional retrieval, not a class-agnostic completion pipeline.
  2. [Section III-A and Section IV.C] The description of the 'class-agnostic training setting' (RefComp Unified) is incomplete. The paper does not state whether the reference pairs used during Unified training are also selected by same-class Chamfer distance (as in Section III-A) or by a label-free rule. If same-class selection is used during training, then class labels are required at train time for every reference pair, which further weakens the 'unified/class-agnostic' claim; if label-free selection is used, the retrieval rule and any changes to the Section III-A pipeline should be specified. This distinction is necessary because the loss and the pseudo-label role of the reference data depend on which pairs are admitted.
minor comments (5)
  1. [Equation (3)] In Eq. (3), the notation defines '+' as element-wise addition, but the prose reads 'denoted element-wise addition'; the grammar should be corrected, and the precedence of concatenation versus addition in the fused expression should be clarified.
  2. [Section IV.A] The text states 'we use a minimum CD = 1.0' when selecting reference pairs, but the CD values in the tables are scaled by 1e4; please specify whether the threshold is also scaled and give the threshold in the same units as the reported metrics.
  3. [Table VI] The entries 'Generalized + Random', 'Generalized + Matching', 'Fine-tuned + Random', and 'Fine-tuned + Matching' are not explicitly mapped to RefComp variants in the text; please add a sentence explaining what each configuration is and how it relates to RefComp Unified.
  4. [Section III-A and Section IV.A] The complete-shape database used for reference retrieval is described as ShapeNet with 55 classes, but it is not stated whether the same database is used for the KITTI, ScanNet, and MatterPort3D experiments; please specify the reference database used for each real-world benchmark.
  5. [References] Several references are duplicated under different numbers (e.g., [16]/[39], [20]/[40], and [22]/[63]); please consolidate the bibliography.

Circularity Check

2 steps flagged · score 6.0 of 10

Real-world UCD 'predictions' are the training objective itself, and the class-agnostic test protocol retrieves same-class (possibly same-instance) ShapeNet references, partially forcing the reported gains.

  1. fitted input called prediction [Section III-B (Eqs. 2 and 5), Section IV-B (Eq. 6), Section IV-E]
    "To backpropagate gradients through the target branch and maintain the overall shape of px, we use the Deg(·) module in [35] to degrade cˆx back to pˆx. We then compute the CD-based loss between px and pˆx: LCD(pˆx,px) = ... (2) ... L = α(Lref_CD + Lr_CD) + β(Ltar_CD + Lp_CD) + γLW, (5) ... UCD (px, ˆcx) = 1/|px| Σ_{a∈px} min_{b∈ˆcx} ∥a−b∥^2_2, (6)"

    By construction pˆx = Deg(cˆx) is a subset of cˆx, so the second summand of Ltar_CD = LCD(Deg(cˆx),px) is an upper bound on UCD(px,cˆx): for every px point its nearest neighbor in cˆx is no farther than its nearest neighbor in the degraded subset pˆx. Since Eq. (5) explicitly minimizes β·Ltar_CD, the training objective directly minimizes the same quantity that Table IV reports as the real-world UCD performance metric. The 'SOTA' UCD numbers are therefore a restatement of the optimized loss rather than an independent measure of completion quality. This is a fitted input renamed as a prediction.

  2. other [Section III-A, Section IV-A, Section IV-C]
    "After creating the reference pairs, we use the Chamfer Distance (CD) between px and all {py} of the same class to select the top N pairs ... The reference pairs are created from ShapeNet [62] ... We use for evaluation the virtual-scan datasets EPN-3D [39], CRN [40], PartNet [41] with ground truth. These three datasets are derived from ShapeNet [62] ... During testing, we only use the reference pair that is the closest to px of the same class."

    The class-agnostic claim and the virtual-scan numbers are produced by a test protocol that injects class labels: the reference complete cloud cy is selected only from the same class as px. Because the reference database is drawn from the same ShapeNet source that generated the CRN/EPN/PartNet test partials and no exclusion of test instances is stated, cy can be the exact complete model from which px's ground truth cx was rendered. In that case the 'unpaired' completion is conditioned on the target's ground truth, so the CD/F1/MMD results are partially forced by retrieval of the answer, not by completion from the partial alone. The headline generalization claim is thus self-definitional: class-agnostic performance is defined through same-class retrieval.

full rationale

The framework's internal training scheme is not circular in the usual sense: the reference branch is supervised by self-generated partial-complete pairs, the target branch inherits via shared weights, and the ablation studies support the LSFM and parameter-sharing design. I found no load-bearing self-citation chain. However, two evaluation-time reductions are real. First, the real-world UCD metric is essentially the second term of the target-branch CD loss minimized in Eq. (5), so Table IV's UCD 'SOTA' is a restatement of the training objective. Second, the protocol for RefComp Unified selects the closest reference pair 'of the same class' from a ShapeNet database while the virtual-scan test sets are themselves derived from ShapeNet, so the reference complete cloud can be the target's ground-truth object; the class-agnostic, unpaired claim is then supported by same-class, potentially same-instance retrieval. These issues make the headline predictions partially circular, though the architecture and loss design retain independent content.

Assumptions & free parameters 7 free parameters · 5 assumptions · 0 invented entities

The central claim rests on the transferability of missing-region masks between retrieved references and the target, and on the availability of same-class references at test time. These are domain assumptions, not standard math; no new physical entities are introduced.

free parameters (7)
  • Loss weights alpha, beta, gamma = 0.35, 0.65, 0.001
    Weights in Eq. 5 chosen by hand or validation; they balance reference, target, and Wasserstein losses.
  • Degradation top-k for reference generation = 15
    Number of nearest points in cy matched to each point in px when corrupting cy; set empirically in Section IV.C.
  • Degradation top-k for target reconstruction = 5
    Used in Deg(c_hat_x) to recover p_hat_x; empirically chosen.
  • Number of retrieved reference pairs N = 3
    Top-3 same-class reference pairs selected; one randomly chosen per iteration; empirically set.
  • Minimum Chamfer distance threshold = 1.0
    Avoids retrieving reference partials too similar to target; fixed in Section IV.A.
  • LSFM channel count = 256->512->256
    Feature dimensions in fusion module; architecture choice.
  • Training hyperparameters = lr=5e-4, wd=5e-4, batch=50, epochs=600
    Optimization settings; not central to the method but needed to reproduce.
assumptions (5)
  • domain assumption A mask my = cy - py computed from a reference pair represents missing structure transferable to the target partial px.
    Invoked in Section III-C to fuse zmy with zpx; if real corruption differs, target completion degrades.
  • domain assumption The degradation module from [35] produces partial point clouds whose corruption pattern resembles the target partial's missing regions.
    Reference pairs are generated by corrupting cy using px as template (Section III-A); this assumes template-based degradation transfers.
  • domain assumption Same-class reference retrieval is available at test time.
    Section III-A selects top-N pairs of the same class; Section IV.C states the closest same-class pair is used at test, requiring class labels.
  • domain assumption Wasserstein distance on latent features is a sufficient objective to align the target completion features with the complete latent space.
    Section III-C, Eq. 4; if imperfect, the target branch may not produce complete shapes.
  • standard math Standard deep learning and optimization assumptions (PointNet encoders, MLP decoders, AdamW) hold.
    Background architecture choices, not proven in this paper.

how reviews work

0 comments
Cite this review

Pith. "Pith review of RefComp: A Reference-guided Unified Framework for Unpaired Point Cloud Completion." pith.science (2026). https://pith.science/paper/U3FN6AZE

@misc{pith2026250413788,
  author       = {Pith},
  title        = {Pith review of: RefComp: A Reference-guided Unified Framework for Unpaired Point Cloud Completion},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/U3FN6AZE}},
  note         = {Machine review of arXiv:2504.13788}
}
read the original abstract

The unpaired point cloud completion task aims to complete a partial point cloud by using models trained with no ground truth. Existing unpaired point cloud completion methods are class-aware, i.e., a separate model is needed for each object class. Since they have limited generalization capabilities, these methods perform poorly in real-world scenarios when confronted with a wide range of point clouds of generic 3D objects. In this paper, we propose a novel unpaired point cloud completion framework, namely the Reference-guided Completion (RefComp) framework, which attains strong performance in both the class-aware and class-agnostic training settings. The RefComp framework transforms the unpaired completion problem into a shape translation problem, which is solved in the latent feature space of the partial point clouds. To this end, we introduce the use of partial-complete point cloud pairs, which are retrieved by using the partial point cloud to be completed as a template. These point cloud pairs are used as reference data to guide the completion process. Our RefComp framework uses a reference branch and a target branch with shared parameters for shape fusion and shape translation via a Latent Shape Fusion Module (LSFM) to enhance the structural features along the completion pipeline. Extensive experiments demonstrate that the RefComp framework achieves not only state-of-the-art performance in the class-aware training setting but also competitive results in the class-agnostic training setting on both virtual scans and real-world datasets.

Figures

Figures reproduced from arXiv: 2504.13788 by the authors.

Figure 1
Figure 1. Unpaired point cloud completion by shape translation. cy is corrupted to create py by using px as the corruption template. zpy and zpx are mapped into features representing complete point clouds by the Latent Shape Fusion Module (LSFM), T. zcyˆ is forced to be in the ZC space by a Wasserstein Distance loss, which pulls zcxˆ into the complete latent space. Finally, the completed features are decoded back to the compl… view at source ↗
Figure 2
Figure 2. Our RefComp framework. The reference branch shares the parameters with the target branch to assist in the unpaired point cloud completion. px, py are encoded by the same encoder Ep. Completion is achieved by the Latent Shape Fusion Module (LSFM) with the mask point cloud my in the latent feature space. cy provides the alignment objective, ensuring the completed features zcyˆ align with the complete features zcy . A … view at source ↗
Figure 3
Figure 3. Generation of the reference data. For each point in px, the degradation module uses K-nearest neighbour (KNN) to find the top K closest points in cy, where py is the union of these points. A CD loss between each py and px is used to find the top-N reference pairs. Specifically, we use the non-parameter degradation module proposed in [35] to corrupt cy by using px as the corruption template, where px is the target pa… view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: The Latent Shape Fusion Module (LSFM). The parameters between the reference and the target branch are shared to compute zcˆx and zcˆy . To complete the partial point clouds in the latent feature space, the LSFM fuses the features of the partial point clouds, i.e., zpx …
Figure 5
Figure 5. Figure 5: Sample completion results on the CRN dataset. All versions of our framework produce plausible results. The version RefComp w/dis further enhances edges. The reference data used by the three versions of RefComp is shown in yellow. sample in set Sc and samples in the gro…
Figure 6
Figure 6. Figure 6: Sample completion results on real-world datasets. Our framework, with or without discriminators, produces more plausible results than other methods. The reference data used by three versions of RefComp is shown in yellow. that RefComp and RefComp w/dis produce more sta…
Figure 7
Figure 7. Figure 7: Additional experiments. (a) Completion results with different reference data on real-world datasets (left to right) KITTI, ScanNet, and MatterPort. (b) Overlapping the partial input on the completed results on the CRN dataset to demonstrate the framework’s capability o…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

61 extracted references · 42 canonical work pages

  1. [1]

    Transformation-Equivariant 3D Object Detection for Autonomous Driving

    H. Wu, C. Wen, W. Li, X. Li, R. Yang, and C. Wang, “Transformation- equivariant 3d object detection for autonomous driving,” arXiv preprint arXiv:2211.11962, 2022

  2. [2]

    Sparse fuse dense: Towards high quality 3d detection with depth completion,

    X. Wu, L. Peng, H. Yang, L. Xie, C. Huang, C. Deng, H. Liu, and D. Cai, “Sparse fuse dense: Towards high quality 3d detection with depth completion,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , 2022, pp. 5418–5427

  3. [3]

    Casa: A cascade attention network for 3-d object detection from lidar point clouds,

    H. Wu, J. Deng, C. Wen, X. Li, C. Wang, and J. Li, “Casa: A cascade attention network for 3-d object detection from lidar point clouds,” IEEE Transactions on Geoscience and Remote Sensing , vol. 60, pp. 1–11, 2022

  4. [4]

    Sparse2Dense: Learning to Densify 3D Features for 3D Object Detection

    T. Wang, X. Hu, Z. Liu, and C.-W. Fu, “Sparse2dense: Learn- ing to densify 3d features for 3d object detection,” arXiv preprint arXiv:2211.13067, 2022

  5. [5]

    Virpnet: A multimodal virtual point generation network for 3d object detection,

    L. Wang, S. Sun, and J. Zhao, “Virpnet: A multimodal virtual point generation network for 3d object detection,” IEEE Transactions on Multimedia, 2024

  6. [6]

    Esc-net: Alleviating triple sparsity on 3d lidar point clouds for extreme sparse scene completion,

    P. An, D. Zhu, S. Quan, J. Ding, J. Ma, Y . Yang, and Q. Liu, “Esc-net: Alleviating triple sparsity on 3d lidar point clouds for extreme sparse scene completion,” IEEE Transactions on Multimedia , 2024

  7. [7]

    Multi-view 3d reconstruction with transformers,

    D. Wang, X. Cui, X. Chen, Z. Zou, T. Shi, S. Salcudean, Z. J. Wang, and R. Ward, “Multi-view 3d reconstruction with transformers,” in Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), October 2021, pp. 5722–5731

  8. [8]

    H3d-net: Few-shot high-fidelity 3d head reconstruction,

    E. Ramon, G. Triginer, J. Escur, A. Pumarola, J. Garcia, X. Gir ´o- i Nieto, and F. Moreno-Noguer, “H3d-net: Few-shot high-fidelity 3d head reconstruction,” in Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) , October 2021, pp. 5620–5629

Show all 61 references
  1. [9]

    Pon- der: Point cloud pre-training via neural rendering,

    D. Huang, S. Peng, T. He, X. Zhou, and W. Ouyang, “Pon- der: Point cloud pre-training via neural rendering,” arXiv preprint arXiv:2301.00157, 2022

  2. [10]

    Cp-net: contour- perturbed reconstruction network for self-supervised point cloud learn- ing,

    M. Xu, Z. Zhou, H. Xu, Y . Qiao, and Y . Wang, “Cp-net: contour- perturbed reconstruction network for self-supervised point cloud learn- ing,” IEEE Transactions on Multimedia , 2024

  3. [11]

    3d reconstruction based on hierarchical reinforcement learning with transferability,

    L. Li, F. He, R. Fan, B. Fan, and X. Yan, “3d reconstruction based on hierarchical reinforcement learning with transferability,” Integrated Computer-Aided Engineering, vol. 30, no. 4, pp. 327–339, 2023

  4. [12]

    From the semantic point cloud to heritage-building information mod- eling: A semiautomatic approach exploiting machine learning,

    V . Croce, G. Caroti, L. De Luca, K. Jacquot, A. Piemonte, and P. V ´eron, “From the semantic point cloud to heritage-building information mod- eling: A semiautomatic approach exploiting machine learning,” Remote Sensing, vol. 13, no. 3, p. 461, 2021

  5. [13]

    Classification of 3d digital heritage,

    E. Grilli and F. Remondino, “Classification of 3d digital heritage,” Remote Sensing, vol. 11, no. 7, p. 847, 2019

  6. [14]

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

    C. R. Qi, H. Su, K. Mo, and L. J. Guibas, “Pointnet: Deep learning on point sets for 3d classification and segmentation,” in Proceedings of the IEEE conference on computer vision and pattern recognition , 2017, pp. 652–660

  7. [15]

    Pointgt: A method for point-cloud classification and segmentation based on local geometric transformation,

    H. Zhang, C. Wang, L. Yu, S. Tian, X. Ning, and J. Rodrigues, “Pointgt: A method for point-cloud classification and segmentation based on local geometric transformation,” IEEE Transactions on Multimedia , 2024

  8. [17]

    Pf-net: Point fractal network for 3d point cloud completion,

    Z. Huang, Y . Yu, J. Xu, F. Ni, and X. Le, “Pf-net: Point fractal network for 3d point cloud completion,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2020, pp. 7662– 7670

  9. [18]

    Morphing and sampling network for dense point cloud completion,

    M. Liu, L. Sheng, S. Yang, J. Shao, and S.-M. Hu, “Morphing and sampling network for dense point cloud completion,” in Proceedings of the AAAI conference on artificial intelligence , 2020, pp. 11 596–11 603

  10. [19]

    Topnet: Structural point cloud decoder,

    L. P. Tchapmi, V . Kosaraju, H. Rezatofighi, I. Reid, and S. Savarese, “Topnet: Structural point cloud decoder,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , 2019, pp. 383–392

  11. [21]

    Point cloud completion by skip-attention network with hierarchical folding,

    X. Wen, T. Li, Z. Han, and Y .-S. Liu, “Point cloud completion by skip-attention network with hierarchical folding,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , 2020, pp. 1939–1948

  12. [23]

    Detail preserved point cloud com- pletion via separated feature aggregation,

    W. Zhang, Q. Yan, and C. Xiao, “Detail preserved point cloud com- pletion via separated feature aggregation,” in European Conference on Computer Vision. Springer, 2020, pp. 512–528

  13. [24]

    Pointr: Diverse point cloud completion with geometry-aware transformers,

    X. Yu, Y . Rao, Z. Wang, Z. Liu, J. Lu, and J. Zhou, “Pointr: Diverse point cloud completion with geometry-aware transformers,” in Proceedings of the IEEE/CVF international conference on computer vision , 2021, pp. 12 498–12 507

  14. [25]

    Learning 3d shape latent for point cloud completion,

    Z. Chen, F. Long, Z. Qiu, T. Yao, W. Zhou, J. Luo, and T. Mei, “Learning 3d shape latent for point cloud completion,” IEEE Transactions on Multimedia, 2024

  15. [26]

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

    X. Yu, L. Tang, Y . Rao, T. Huang, J. Zhou, and J. Lu, “Point-bert: Pre-training 3d point cloud transformers with masked point modeling,”

  16. [27]

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

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

  17. [28]

    Foldingnet: Point cloud auto-encoder via deep grid deformation,

    Y . Yang, C. Feng, Y . Shen, and D. Tian, “Foldingnet: Point cloud auto-encoder via deep grid deformation,” in Proceedings of the IEEE conference on computer vision and pattern recognition , 2018, pp. 206– 215

  18. [29]

    Psnet: Fast data structuring for hierarchical deep learning on point cloud,

    L. Li, L. He, J. Gao, and X. Han, “Psnet: Fast data structuring for hierarchical deep learning on point cloud,” IEEE Transactions on Circuits and Systems for Video Technology , vol. 32, no. 10, pp. 6835– 6849, 2022

  19. [30]

    Unpaired point cloud completion on real scans using adversarial training,

    X. Chen, B. Chen, and N. J. Mitra, “Unpaired point cloud completion on real scans using adversarial training,” openreview.net, 2019

  20. [31]

    Cy- cle4completion: Unpaired point cloud completion using cycle transfor- mation with missing region coding

    X. Wen, Z. Han, Y .-P. Cao, P. Wan, W. Zheng, and Y .-S. Liu, “Cy- cle4completion: Unpaired point cloud completion using cycle transfor- mation with missing region coding.” in Computer Vision and Pattern Recognition (CVPR), 2021, pp. 13 080–13 089

  21. [32]

    Learning a structured latent space for unsupervised point cloud completion,

    Y . Cai, K.-Y . Lin, C. Zhang, Q. Wang, X. Wang, and H. Li, “Learning a structured latent space for unsupervised point cloud completion,” in 13 Computer Vision and Pattern Recognition (CVPR) , 2022, pp. 5543– 5553

  22. [33]

    Multimodal shape com- pletion via conditional generative adversarial networks,

    R. Wu, X. Chen, Y . Zhuang, and B. Chen, “Multimodal shape com- pletion via conditional generative adversarial networks,” in Computer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Proceedings, Part IV 16 . Springer, 2020, pp. 281–296

  23. [34]

    Reverse2complete: Unpaired multimodal point cloud completion via guided diffusion,

    W. Zhang, H. Rahmani, X. Yang, and J. Liu, “Reverse2complete: Unpaired multimodal point cloud completion via guided diffusion,” in Proceedings of the 32nd ACM International Conference on Multimedia , 2024, pp. 5892–5901

  24. [35]

    Unsupervised 3d shape completion through gan inversion

    J. Zhang, X. Chen, Z. Cai, L. Pan, H. Zhao, S. Yi, C. K. Yeo, B. Dai, and C. C. Loy, “Unsupervised 3d shape completion through gan inversion.” in Computer Vision and Pattern Recognition (CVPR) , 2021, pp. 1768– 1777

  25. [36]

    Unpaired image-to-image translation using cycle-consistent adversarial networks,

    J.-Y . Zhu, T. Park, P. Isola, and A. A. Efros, “Unpaired image-to-image translation using cycle-consistent adversarial networks,” in Proceedings of the IEEE international conference on computer vision , 2017, pp. 2223–2232

  26. [37]

    Energy-based residual latent transport for unsupervised point cloud completion,

    R. Cui, S. Qiu, S. Anwar, J. Zhang, and N. Barnes, “Energy-based residual latent transport for unsupervised point cloud completion,” arXiv preprint arXiv:2211.06820, 2022

  27. [38]

    P2c: Self-supervised point cloud completion from single partial clouds,

    R. Cui, S. Qiu, S. Anwar, J. Liu, C. Xing, J. Zhang, and N. Barnes, “P2c: Self-supervised point cloud completion from single partial clouds,” in Proceedings of the IEEE/CVF International Conference on Computer Vision, 2023, pp. 14 351–14 360

  28. [40]

    Cascaded refinement network for point cloud completion,

    X. Wang, M. H. Ang Jr, and G. H. Lee, “Cascaded refinement network for point cloud completion,” in Proceedings of the IEEE/CVF Confer- ence on Computer Vision and Pattern Recognition , 2020, pp. 790–799

  29. [41]

    Partnet: A large-scale benchmark for fine-grained and hierarchical part-level 3d object understanding,

    K. Mo, S. Zhu, A. X. Chang, L. Yi, S. Tripathi, L. J. Guibas, and H. Su, “Partnet: A large-scale benchmark for fine-grained and hierarchical part-level 3d object understanding,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , 2019, pp. 909– 918

  30. [42]

    Vision meets robotics: The kitti dataset,

    A. Geiger, P. Lenz, C. Stiller, and R. Urtasun, “Vision meets robotics: The kitti dataset,” International Journal of Robotics Research (IJRR) , 2013

  31. [43]

    Scannet: Richly-annotated 3d reconstructions of indoor scenes,

    A. Dai, A. X. Chang, M. Savva, M. Halber, T. Funkhouser, and M. Nießner, “Scannet: Richly-annotated 3d reconstructions of indoor scenes,” in Proceedings of the IEEE conference on computer vision and pattern recognition, 2017, pp. 5828–5839

  32. [44]

    Matterport3d: Learning from rgb-d data in indoor environments,

    A. Chang, A. Dai, T. Funkhouser, M. Halber, M. Niessner, M. Savva, S. Song, A. Zeng, and Y . Zhang, “Matterport3d: Learning from rgb-d data in indoor environments,” arXiv preprint arXiv:1709.06158 , 2017

  33. [45]

    Shape completion using 3d-encoder-predictor cnns and shape synthesis,

    A. Dai, C. Ruizhongtai Qi, and M. Nießner, “Shape completion using 3d-encoder-predictor cnns and shape synthesis,” in Proceedings of the IEEE conference on computer vision and pattern recognition , 2017, pp. 5868–5877

  34. [46]

    Shape inpaint- ing using 3d generative adversarial network and recurrent convolutional networks,

    W. Wang, Q. Huang, S. You, C. Yang, and U. Neumann, “Shape inpaint- ing using 3d generative adversarial network and recurrent convolutional networks,” in Proceedings of the IEEE international conference on computer vision, 2017, pp. 2298–2306

  35. [47]

    3d object reconstruction from a single depth view with adversarial learning,

    B. Yang, H. Wen, S. Wang, R. Clark, A. Markham, and N. Trigoni, “3d object reconstruction from a single depth view with adversarial learning,” in Proceedings of the IEEE international conference on computer vision workshops, 2017, pp. 679–688

  36. [48]

    Pcn: Point completion network,

    W. Yuan, T. Khot, D. Held, C. Mertz, and M. Hebert, “Pcn: Point completion network,” in 2018 international conference on 3D vision (3DV). IEEE, 2018, pp. 728–737

  37. [49]

    Reference-based image super-resolution with deformable atten- tion transformer,

    J. Cao, J. Liang, K. Zhang, Y . Li, Y . Zhang, W. Wang, and L. V . Gool, “Reference-based image super-resolution with deformable atten- tion transformer,” in Computer Vision–ECCV 2022: 17th European Conference, Tel Aviv, Israel, October 23–27, 2022, Proceedings, Part XVIII. Sp...

  38. [50]

    Masa-sr: Matching acceleration and spatial adaptation for reference-based image super-resolution,

    L. Lu, W. Li, X. Tao, J. Lu, and J. Jia, “Masa-sr: Matching acceleration and spatial adaptation for reference-based image super-resolution,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2021, pp. 6368–6377

  39. [51]

    Crossnet: An end-to-end reference-based super resolution network using cross-scale warping,

    H. Zheng, M. Ji, H. Wang, Y . Liu, and L. Fang, “Crossnet: An end-to-end reference-based super resolution network using cross-scale warping,” in Proceedings of the European conference on computer vision (ECCV) , 2018, pp. 88–104

  40. [52]

    Bringing old photos back to life,

    Z. Wan, B. Zhang, D. Chen, P. Zhang, D. Chen, J. Liao, and F. Wen, “Bringing old photos back to life,” in proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2020, pp. 2747– 2757

  41. [53]

    Hairclip: Design your hair by text and reference image,

    T. Wei, D. Chen, W. Zhou, J. Liao, Z. Tan, L. Yuan, W. Zhang, and N. Yu, “Hairclip: Design your hair by text and reference image,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2022, pp. 18 072–18 081

  42. [54]

    Unpaired deep image deraining using dual contrastive learning,

    X. Chen, J. Pan, K. Jiang, Y . Li, Y . Huang, C. Kong, L. Dai, and Z. Fan, “Unpaired deep image deraining using dual contrastive learning,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2022, pp. 2017–2026

  43. [55]

    End-to-end referring video object segmentation with multimodal transformers,

    A. Botach, E. Zheltonozhskii, and C. Baskin, “End-to-end referring video object segmentation with multimodal transformers,” in Proceed- ings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2022, pp. 4985–4995

  44. [56]

    Gres: Generalized referring expression segmentation,

    C. Liu, H. Ding, and X. Jiang, “Gres: Generalized referring expression segmentation,” in Proceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition , 2023, pp. 23 592–23 601

  45. [57]

    Vi- sual prompting via image inpainting,

    A. Bar, Y . Gandelsman, T. Darrell, A. Globerson, and A. Efros, “Vi- sual prompting via image inpainting,” Advances in Neural Information Processing Systems, vol. 35, pp. 25 005–25 017, 2022

  46. [58]

    Images speak in images: A generalist painter for in-context visual learning,

    X. Wang, W. Wang, Y . Cao, C. Shen, and T. Huang, “Images speak in images: A generalist painter for in-context visual learning,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023, pp. 6830–6839

  47. [59]

    Explore in-context learning for 3d point cloud understanding,

    Z. Fang, X. Li, X. Li, J. M. Buhmann, C. C. Loy, and M. Liu, “Explore in-context learning for 3d point cloud understanding,” arXiv preprint arXiv:2306.08659, 2023

  48. [60]

    Deep residual learning for image recognition,

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

  49. [61]

    Statistical aspects of wasserstein distances,

    V . M. Panaretos and Y . Zemel, “Statistical aspects of wasserstein distances,” Annual review of statistics and its application , vol. 6, pp. 405–431, 2019

  50. [62]

    Shapenet: An information- rich 3d model repository,

    A. X. Chang, T. Funkhouser, L. Guibas, P. Hanrahan, Q. Huang, Z. Li, S. Savarese, M. Savva, S. Song, H. Su et al., “Shapenet: An information- rich 3d model repository,” arXiv preprint arXiv:1512.03012 , 2015

  51. [63]

    Grnet: Grid- ding residual network for dense point cloud completion,

    H. Xie, H. Yao, S. Zhou, J. Mao, S. Zhang, and W. Sun, “Grnet: Grid- ding residual network for dense point cloud completion,” in European Conference on Computer Vision . Springer, 2020, pp. 365–381

  52. [64]

    Crossvit: Cross-attention multi- scale vision transformer for image classification,

    C.-F. R. Chen, Q. Fan, and R. Panda, “Crossvit: Cross-attention multi- scale vision transformer for image classification,” in Proceedings of the IEEE/CVF international conference on computer vision , 2021, pp. 357– 366. Yixuan Yang is pursuing a Ph.D. degree in Com- puter Scie...

  53. [2021]

    Available: https://arxiv.org/abs/2111.14819

    [Online]. Available: https://arxiv.org/abs/2111.14819

Pith tools

Reviewed August 16, 2026 · model on record in the stance chip above.