Beyond the Final Layer: Hierarchical Query Fusion Transformer with Agent-Interpolation Initialization for 3D Instance Segmentation

Jiacheng Deng; Jiahao Lu; Tianzhu Zhang

REVIEW 3 major objections 5 minor 18 references

Beyond the Final Layer: Hierarchical Query Fusion Transformer with Agent-Interpolation Initialization for 3D Instance Segmentation

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

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

Pith's one-line read The paper claims that transformer-based 3D instance segmentation can be improved by interpolating content queries from learnable agents at farthest-point-sampled positions and by carrying low-overlap queries forward across decoder layers…

desk verdict Solid plug-and-play modules for 3D instance segmentation, but the headline SOTA margins are inflated by an NMS mismatch that the authors should fix. read the letter →

arxiv 2502.04139 v1 pith:2OYBEW6I submitted 2025-02-06 cs.CV

Jiahao Lu , Jiacheng Deng , Tianzhu Zhang This is my paper · ORCID

classification cs.CV

keywords 3Dinstancesegmentationtransformerobjectqueriesqueryinitializationfarthestpointsamplinglearnableagentsinterpolationrecall

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

Transformer-based 3D instance segmentation methods initialize a fixed set of object queries and supervise each decoder layer, but two failure modes remain: query initialization cannot keep both strong position and strong content information, and objects disappear as layers deepen because difficult instances get poorly updated. The paper proposes BFL, with an Agent-Interpolation Initialization Module that runs farthest point sampling on the scene and then interpolates a small set of learnable agent content embeddings at those sampled positions, and a Hierarchical Query Fusion Decoder that retains queries from the previous layer whose masks have low overlap with the current layer's masks. On ScanNetV2, ScanNet200, ScanNet++, and S3DIS, BFL reports higher mAP and AP than prior transformer-based methods, including 61.7 mAP on ScanNetV2 validation versus 58.4 for the Maft baseline. The paper's claim is that the query lifecycle—how queries are born and how they survive across layers—carries much of the remaining accuracy gap, and can be fixed without changing the backbone or the loss.

What carries the argument

The machinery is the agent-interpolation identity: for each of S farthest-point-sampled positions, find its K nearest learnable agents, weight their content vectors by inverse distance ($W_{i,j}=dis^{-1}_{i,j}/\sum_j dis^{-1}_{i,j}$), and sum to form the content query $Q^c_i$. A scene-specific refinement rescales agent positions into the scene's bounding box, and a stop-gradient trick ($\tilde{Q}^p = SG(Q^p - \Phi(W,Q^p_0,idx)) + \Phi(W,Q^p_0,idx)$) keeps agent positions trainable while leaving the actual query positions unchanged. The second mechanism is the fusion gate: between adjacent decoder layers, compute pairwise mask IoU, take each previous-layer mask's maximum IoU against all current-layer masks, keep the D1 previous-layer queries with the smallest such maxima, and concatenate them into the next layer. The first mechanism imports cross-scene content without sacrificing foreground coverage; the second prevents difficult instances from being silently dropped.

What would settle it

Run the ScanNetV2 validation experiment with Eq. 3's weights replaced by uniform weights over the K nearest agents, keeping all other settings (S=400, L=400, K=3) fixed. If mAP does not drop outside the noise of Table 8 (or if the K=8 row behaves the same as K=3), then the inverse-distance interpolation is not carrying the AI2M gain. A second check is to keep HQFD but initialize all content queries from a single shared learned vector with no agent interpolation, and observe whether first-layer recall and final mAP fall back toward the FPS-based baseline.

Watch

Extended reading notes

Core claim

The central claim is that the remaining accuracy gap in transformer-based 3D instance segmentation lies in the query lifecycle, not the backbone or the loss. The Agent-Interpolation Initialization Module treats query initialization as a balance between foreground coverage and content learning: farthest-point sampling supplies positions that cover nearly all foreground objects, while learnable agent content vectors supply dataset-level inductive bias, and inverse-distance interpolation fuses the two. The Hierarchical Query Fusion Decoder treats recall decline as a survival problem: when a query's mask after a decoder update has low maximum IoU against all masks from the previous layer, the pre-update query is copied forward and re-updated, preventing hard instances from disappearing. The paper reports that this combination makes per-layer recall improve monotonically, records the highest mAP/AP among compared methods on ScanNetV2 (61.7 validation and 60.6 test mAP), ScanNet200 (30.5), ScanNet++ (25.3 validation and 22.2 test), and S3DIS (71.9 AP@50), and shows that the fusion module also improves SPFormer and Maft when added on top.

Load-bearing premise

The load-bearing assumption is that a content query formed by inverse-distance-weighted interpolation of a few learned agent vectors at farthest-point-sampled positions is genuinely more informative than a pure positional embedding—that is, that useful content varies smoothly enough in 3D within the agent set—and the reported AP gains would not materialize if that smoothness assumption failed.

Editorial extensions

If this is right

If the central claim holds, query initialization in 3D instance segmentation can be improved without redesigning the backbone, the decoder layer, or the loss function—only the way content is attached to sampled positions needs to change.
The Hierarchical Query Fusion Decoder is claimed to be plug-and-play: adding it to SPFormer and Maft raises their mAP, so other per-layer-supervised query methods may get a similar recall lift by retaining low-overlap queries.
The reported convergence curve implies that training budgets can be cut: BFL at 128 epochs beats the baseline at 512, so the initialization and fusion mechanisms shorten training time as well as improving accuracy.
The cost of the recall fix is small: query count grows by at most D1 per fused layer, runtime rises by 7.8%, and parameters grow by 0.2M relative to Maft, suggesting the gains are not bought with a larger model.
The same architecture, loss, and decoder are used on ScanNetV2, ScanNet200, ScanNet++, and S3DIS, so the reported gains extend across annotation density (198 classes in ScanNet200, sub-millimeter resolution in ScanNet++) without dataset-specific changes.

Reading between the lines

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

Editorial inference: the inverse-distance weighting in Eq. 3 is an untested modeling choice; replacing it with uniform weights among the K nearest agents would show whether the smoothness prior itself, rather than having learnable content at FPS positions, drives the AI2M gain. Table 8's weak sensitivity to K hints the answer may be the latter, but the paper does not run this control.
Editorial inference: the low-overlap fusion mechanism is not inherently three-dimensional; 2D transformer-based instance segmentation also supervises every decoder layer and suffers from object disappearance, so HQFD-style retention is a transferable idea the paper does not test.
Editorial inference: the method's reliance on FPS assumes foreground objects are spread through Euclidean space; in scenes with strongly uneven instance density, such as street scenes, the fixed number of agents L may need to scale with the scene, a regime the four indoor benchmarks do not cover.

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.

Desk Editor's Note

read the letter

Two things to know. First, the core idea is good: combining FPS position sampling with learnable content agents through inverse-distance interpolation gives better initial queries than either approach alone, and the hierarchical query fusion decoder prevents the recall drop that plagues per-layer supervised transformers. The ablation evidence supports both modules, and the plug-and-play test on SPFormer and Maft is a nice touch. Second, the headline numbers are not to be trusted as reported. The main comparison table gives Maft without NMS while the proposed method uses NMS; the paper's own appendix shows Maft+NMS is 59.0 val and 59.6 test, versus 58.4 and 57.8 in the main table. That shrinks the advertised gains from 3.3/2.8 to about 2.7/1.0. The module-level ablation in Table 7, which does not use NMS for either side, still shows a real 2.7 mAP gain (61.1 vs 58.4), so the method survives, but the SOTA claim needs correcting.

The paper is clearly written and the experiments span four datasets. The AI2M design is genuinely new as a fusion of FPS and learnable queries; the smoothness assumption behind interpolation is reasonable and, more importantly, supported by the empirical gains. HQFD is also a simple and sensible fix to a known problem. The main weaknesses are the comparison inconsistency, the lack of code, and the single-run evaluation with no error bars. The appendix also has a few typos and a mislabeled table, but those are minor.

This paper is for researchers working on transformer-based 3D instance segmentation. It deserves a serious referee: the modules are promising and the experiments are broad, but the authors must fix the NMS mismatch and release code before I would trust the SOTA numbers. If the revision addresses that, it is a solid contribution.

Referee Report

3 major / 5 minor

Summary. The paper proposes BFL, a transformer-based 3D instance segmentation method built on the Maft baseline. It introduces two main components: the Agent-Interpolation Initialization Module (AI2M), which combines FPS-sampled position queries with content queries interpolated from learnable agent position-content pairs (Eqs. 1-5), and the Hierarchical Query Fusion Decoder (HQFD), which carries forward low-overlap queries from earlier decoder layers (Eq. 6) to counteract the observed inter-layer recall decline. Experiments on ScanNetV2, ScanNet200, ScanNet++, and S3DIS report gains over Maft and other transformer-based methods, supported by ablations for each module, a plug-and-play generalization study, and a runtime analysis.

Significance. If the comparison protocol is corrected, the contribution is a solid incremental improvement. The AI2M idea is simple and well motivated, and HQFD is shown to be plug-and-play on SPFormer and Maft (Table 10) and to produce monotonic recall improvement (Table 18). The paper provides extensive ablations (Tables 5, 7-10, 18-21) and clearly describes the method. However, the headline state-of-the-art claim currently rests on an inconsistent evaluation: BFL is inferred with NMS while the main-table Maft baseline appears to be without NMS, and the paper's own appendix contradicts the test-set baseline number. These issues are fixable without changing the method, but they directly affect the advertised performance margins.

major comments (3)

[Section 3.5, Tables 2, 7, 12] The headline comparison in Table 2 is not post-processing-matched. Section 3.5 states that BFL applies NMS at inference, and Table 7 shows that NMS adds 0.6 mAP to the full method (61.1 to 61.7). The Maft baseline in Table 2 is reported at 58.4 mAP on ScanNetV2 validation, which matches the no-NMS Maft in Table 12, while the same appendix table reports Maft+NMS at 59.0. With NMS applied to both, the validation gain would be 61.7 - 59.0 = 2.7 mAP rather than the advertised 3.3. Please re-run all baselines under the identical post-processing, or report both with-NMS and without-NMS columns in every comparison table and explicitly state which protocol each number uses. I note that the module ablation in Table 7 (row 4) still shows a 2.7 mAP gain without NMS, so the qualitative conclusion may survive, but the exact SOTA margin is currently overstated.
[Appendix Table 14 vs Table 2] The ScanNetV2 test numbers are internally inconsistent. Table 2 reports Maft at 57.8 mAP on the hidden test set, but Table 14 reports Maft at 59.6 mAP on the same test set, and Table 15 reports Maft AP@50 at 78.6 versus 77.4 in Table 2. If the appendix numbers are the NMS-applied ones, the advertised test margin shrinks from 60.6 - 57.8 = 2.8 mAP to 60.6 - 59.6 = 1.0 mAP. Please reconcile the two tables, state which evaluation protocol each number uses, and ensure the main-table baseline is the stronger of the two if NMS is part of the method's inference.
[Tables 3, 4, 6; Section 4.1] The multi-benchmark superiority claim rests on small margins over the Maft baseline (ScanNet++ test 22.2 vs 20.9; ScanNet200 validation 30.5 vs 29.2; S3DIS Area 5 AP@50 71.9 vs 69.1), and no error bars or repeated-seed results are reported. Given the post-processing mismatch documented above, it is currently unclear whether these differences are robust. Please report mean and standard deviation over at least three training runs for the main comparisons, or otherwise provide a stability analysis such as evaluation under different NMS thresholds or across checkpoints.

minor comments (5)

[Section 3.3.1, Table 1] The reported mean distances between FPS-sampled points and instance centers are not compared with the analogous distances for learnable-based or random initialization, so the table does not by itself establish that FPS provides a stronger position-content correspondence.
[Section 4.3] The caption of Figure 5 is preceded by a stray duplicate caption line beginning with 'Figure 1: The convergence curve...'; the figure is referenced as Figure 5, so the numbering should be corrected.
[Appendix Table 17] Appendix Table 17 mislabels the PointGroup row as Wu et al. (2022) and the HAIS row as Schult et al. (2022); the correct references are Jiang et al. (2020b) for PointGroup and Chen et al. (2021b) for HAIS.
[Throughout] There are several typos: 'promose' in Section 5, 'teh' in Appendix A.3, 'moudle' in Appendix A.6, and the Maft row in Table 13 contains run-together numbers ('55.144.355.0').
[Section 3.4] The roles of D1 and D2 are only partially specified in the main text; please add a one-sentence definition that D1 is the number of retained low-overlap queries per fusion step and D2 is the number of final decoder layers in which fusion is performed.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the method is benchmark-validated and the modules are not defined in terms of their own outputs.

full rationale

No circular derivation is present. BFL's central claims are empirical: Tables 2-6 report performance against external benchmarks (ScanNetV2, ScanNet200, ScanNet++, S3DIS), and the two new modules are validated by ablations on a held-out validation split (Tables 5, 7-10, 18-19). Equations 1-5 define a learned interpolation from agent embeddings to FPS-sampled positions; the resulting content queries are trained and measured, not fitted to the target metric. Equation 6 and the Bottom-K fusion select low-overlap queries by mask IoU between adjacent decoder layers; this is an architectural mechanism, not an assumption equivalent to the claimed recall improvement. Self-citations to QueryFormer (Lu et al. 2023) and Maft (Lai et al. 2023) provide baseline context and are not load-bearing: the gains over these baselines are evaluated on independent test data. Hyperparameters such as S, L, K, D1, D2 are tuned on validation sets in the standard way (Tables 8, 19, 21) and are not renamed predictions. The one notable flaw is a reporting inconsistency for the Maft baseline on ScanNetV2 test (Table 2 gives 57.8 mAP while Appendix Table 14 gives 59.6); that affects the exact SOTA margin but is a baseline-comparison and correctness issue, not a circular-derivation issue, so it does not raise the circularity score.

Assumptions & free parameters 6 free parameters · 4 assumptions · 1 invented entities

The method rests on several tuned hyperparameters (S, L, K, D1, D2, loss weights) and on heuristics about smoothness of content embeddings and the usefulness of IoU-based query retention. These are reasonable but not derived from first principles.

free parameters (6)

S = 400 (ScanNetV2/S3DIS), 500 (ScanNet200/ScanNet++)
Number of FPS-sampled query positions; tuned in Table 8.
L = 400 (ScanNetV2/S3DIS), 500 (ScanNet200/ScanNet++)
Number of learnable agents; tuned in Table 8.
K = 3
Number of nearest agents used in interpolation; tuned in Table 8.
D1 = 40
Number of low-overlap queries retained per fusion layer; tuned in Table 19.
D2 = 3
Number of decoder layers where fusion is applied; tuned in Table 19.
lambda_1..5 = 0.5, 1, 1, 0.5, 0.5 (ScanNetV2)
Weights in the loss function; tuned in Table 21.

assumptions (4)

ad hoc to paper Inverse distance weighting is a valid interpolation for content embeddings
Equations 2-4 assume that a weighted average of nearby agents' content gives a meaningful content query; no theoretical justification is provided.
domain assumption FPS positions correlate with instance centers
Section 3.3.1 and Table 1 provide empirical evidence on ScanNetV2, but the assumption is carried to all datasets.
ad hoc to paper Low-overlap masks between adjacent layers identify worth-retaining queries
Section 3.4 uses IoU to decide which queries to retain; this heuristic is motivated by examples but not proven optimal.
domain assumption Standard transformer and loss components from Maft are used as-is
The decoder layer and matching are taken from prior work (Lai et al. 2023), so their correctness is assumed.

invented entities (1)

Agents (learnable position-content pairs)
purpose: Used to generate content queries for FPS-sampled positions via interpolation
The agents are internal parameters of the model trained on the target datasets; they have no falsifiable handle outside the paper.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Beyond the Final Layer: Hierarchical Query Fusion Transformer with Agent-Interpolation Initialization for 3D Instance Segmentation." pith.science (2026). https://pith.science/paper/2OYBEW6I

@misc{pith2026250204139,
  author       = {Pith},
  title        = {Pith review of: Beyond the Final Layer: Hierarchical Query Fusion Transformer with Agent-Interpolation Initialization for 3D Instance Segmentation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/2OYBEW6I}},
  note         = {Machine review of arXiv:2502.04139}
}

read the original abstract

3D instance segmentation aims to predict a set of object instances in a scene and represent them as binary foreground masks with corresponding semantic labels. Currently, transformer-based methods are gaining increasing attention due to their elegant pipelines, reduced manual selection of geometric properties, and superior performance. However, transformer-based methods fail to simultaneously maintain strong position and content information during query initialization. Additionally, due to supervision at each decoder layer, there exists a phenomenon of object disappearance with the deepening of layers. To overcome these hurdles, we introduce Beyond the Final Layer: Hierarchical Query Fusion Transformer with Agent-Interpolation Initialization for 3D Instance Segmentation (BFL). Specifically, an Agent-Interpolation Initialization Module is designed to generate resilient queries capable of achieving a balance between foreground coverage and content learning. Additionally, a Hierarchical Query Fusion Decoder is designed to retain low overlap queries, mitigating the decrease in recall with the deepening of layers. Extensive experiments on ScanNetV2, ScanNet200, ScanNet++ and S3DIS datasets demonstrate the superior performance of BFL.

Figures

Figures reproduced from arXiv: 2502.04139 by the authors.

**Figure 1.** The phenomenon of Object Disappearance with the deepening of layers. Scene 1 Scene K Prediction 1 Prediction K Scene 1 Scene K Object Query Object Agent Copy Prediction 1 Prediction K Strong Position Information FPS Based Learnable Based Ours Weak Content Information Weak Position Information Strong Content Information Strong Content Information ... Strong Position Information (a) Agent-Interpolation Initialization … view at source ↗

**Figure 2.** (a) The comparison of different query initialization methods. The FPS-based methods conduct farthest point sampling separately for each scene, placing more emphasis on positional information but lacking in aggregating content information. The learnable-based methods initialize a fixed number of queries for aggregating content information across all scenes, which is prone to empty sampling, thereby compromising foreg… view at source ↗

**Figure 3.** The overall framework of our method BFL. The Agent-Interpolation Initialization Module is meticulously crafted to synergize the strengths of FPS and learnable queries, producing object queries better suited for complex and dynamic environments. The Hierarchical Query Fusion Decoder is utilized to retain low overlap queries that aid in recall rate. 3.3 AGENT-INTERPOLATION INITIALIZATION MODULE 3.3.1 DISCUSSION X Y Z … view at source ↗

Figures from the paper (7 more)

**Figure 4.** Figure 4: Visualization of instance segmentation results on ScanNetV2 validation set. The red boxes highlight the key regions. Method AP@50 AP@25 PointGroup Jiang et al. (2020b) 57.8 / MaskGroup Zhong et al. (2022) 65.0 / SoftGroup Vu et al. (2022) 66.1 / SSTNet Liang et al. (2…

**Figure 1.** Figure 1: The convergence curve under different settings on ScanNet-v2 validation set [PITH_FULL_IMAGE:figures/full_fig_p010_1.png]

**Figure 6.** Figure 6: The architecture of the transformer decoder layer. The figure is taken from the main text of Maft. where p and p denotes a predicted and ground-truth instance, C represents the matching cost matrix, and λcls, λdice, λbce, λcenter are the hyperparameters. Here, λcls, λd…

**Figure 7.** Figure 7: Additional Visual Comparison on ScanNetV2 validation set. The red boxes highlight the key regions. 21 [PITH_FULL_IMAGE:figures/full_fig_p021_7.png]

**Figure 8.** Figure 8: Visual comparisons between the baseline and our method across different decoder layers on ScanNetV2 validation set. The red boxes highlight the key regions. 22 [PITH_FULL_IMAGE:figures/full_fig_p022_8.png]

**Figure 9.** Figure 9: Visual comparisons between the baseline and our method across different decoder layers on ScanNetV2 validation set. The red boxes highlight the key regions. input layer 4 layer 5 layer 6 ground truth Baseline Ours Baseline Ours [PITH_FULL_IMAGE:figures/full_fig_p023_9.png]

**Figure 10.** Figure 10: Visual comparisons between the baseline and our method across different decoder layers on ScanNetV2 validation set. The red boxes highlight the key regions. 23 [PITH_FULL_IMAGE:figures/full_fig_p023_10.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

18 extracted references · 10 canonical work pages

[7]

BSNet: Box-Supervised Simulation-assisted Mean Teacher for 3D Instance Segmentation

Jiahao Lu, Jiacheng Deng, and Tianzhu Zhang. Bsnet: Box-supervised simulation-assisted mean teacher for 3d instance segmentation. arXiv preprint arXiv:2403.15019, 2024a. Jiahao Lu, Jiacheng Deng, Ruijie Zhu, Yanzhe Liang, Wenfei Yang, Tianzhu Zhang, and Xu Zhou. Dn-4dgs: Denoised deformable network with temporal-spatial aggregation for dynamic scene rende...

work page Pith review arXiv
[11]

Monst3r: A simple approach for estimating geometry in the presence of motion

Junyi Zhang, Charles Herrmann, Junhwa Hur, Varun Jampani, Trevor Darrell, Forrester Cole, Deqing Sun, and Ming-Hsuan Yang. Monst3r: A simple approach for estimating geometry in the presence of motion. arXiv preprint arXiv:2410.03825,

arXiv
[12]

Maskgroup: Hierarchical point grouping and masking for 3d instance segmentation

Min Zhong, Xinghao Chen, Xiaokang Chen, Gang Zeng, and Yunhe Wang. Maskgroup: Hierarchical point grouping and masking for 3d instance segmentation. In 2022 IEEE International Conference on Multimedia and Expo (ICME), pp. 1–6. IEEE,

work page 2022
[13]

Motiongs: Exploring explicit motion guidance for deformable 3d gaussian splatting

Ruijie Zhu, Yanzhe Liang, Hanzhi Chang, Jiacheng Deng, Jiahao Lu, Wenfei Yang, Tianzhu Zhang, and Yongdong Zhang. Motiongs: Exploring explicit motion guidance for deformable 3d gaussian splatting. arXiv preprint arXiv:2410.07707,

arXiv
[14]

A.1 O VERVIEW This supplementary material provides more model and experimental details to understand our pro- posed method

A A PPENDIX You may include other additional sections here. A.1 O VERVIEW This supplementary material provides more model and experimental details to understand our pro- posed method. After that, we present more experiments to demonstrate the effectiveness of our methods. Finally, we show a rich visualization of our modules. A.2 M ORE MODEL DETAILS Sparse...

work page 2017
[16]

The specific reasons for this performance decrease are twofold

However, if we apply NMS to the concatenated outputs, as described in Section 1 lines 63-65, a significant decrease in performance occur. The specific reasons for this performance decrease are twofold. Firstly, NMS heavily relies on confidence scores, retaining only the masks with the highest confidence among the duplicates. However, these confidence scor...

work page 2022
[17]

Method mAP bathtub bed bookshe

As indicated by the table, the significant performance improvement highlights the effectiveness of our method in managing denser point cloud scenes across a broader range of categories. Method mAP bathtub bed bookshe. cabinet chair counter curtain desk door other picture frige s. curtain sink sofa table toilet window SoftGroup Vu et al. (2022)45.8 66.6 48...

work page 2022
[21]

A.7 A SSETS AVAILABILITY The datasets that support the findings of this study are available in the following repositories: ScanNetV2 Dai et al

Based on the results, we find that the combination 0.5, 1, 1, 0.5, 0.5 yields the best performance. A.7 A SSETS AVAILABILITY The datasets that support the findings of this study are available in the following repositories: ScanNetV2 Dai et al. (2017) at http://www.scan-net.org/changelog# scannet-v2-2018-06-11 under the ScanNet Terms of Use. ScanNet200 Roz...

work page 2017

Show all 18 references

[32]

Our method incorporates 6 layers of Transformer decoders, with the head number set to 8, and the hidden and feed-forward dimensions set to 256 and 1024, respectively

Unless otherwise specified, we utilize coordinates, colors, and normals as input features. Our method incorporates 6 layers of Transformer decoders, with the head number set to 8, and the hidden and feed-forward dimensions set to 256 and 1024, respectively. For S3DIS Armeni et...

2016
[2006]

Towards end-to-end lane detection: an instance segmentation approach

Davy Neven, Bert De Brabandere, Stamatios Georgoulis, Marc Proesmans, and Luc Van Gool. Towards end-to-end lane detection: an instance segmentation approach. In 2018 IEEE intelligent vehicles symposium (IV), pp. 286–291. IEEE,

2018
[2016]

End-to-end object detection with transformers

Nicolas Carion, Francisco Massa, Gabriel Synnaeve, Nicolas Usunier, Alexander Kirillov, and Sergey Zagoruyko. End-to-end object detection with transformers. In Computer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Proceedings, Part I 16, pp. 213...

2020
[2017]

Imagenet: A large-scale hierarchical image database

Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei-Fei. Imagenet: A large-scale hierarchical image database. In 2009 IEEE conference on computer vision and pattern recognition, pp. 248–255. Ieee,

2009
[2019]

An image is worth 16x16 words: Transformers for image recognition at scale

Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Sylvain Gelly, et al. An image is worth 16x16 words: Transformers for image recognition at scale. arXiv preprint arXiv...

2010 arXiv
[2020]

Decoupled weight decay regularization

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

arXiv
[2021]

Learning gaussian instance segmentation in point clouds

Shih-Hung Liu, Shang-Yi Yu, Shao-Chi Wu, Hwann-Tzong Chen, and Tyng-Luh Liu. Learning gaussian instance segmentation in point clouds. arXiv preprint arXiv:2007.09860,

2007 arXiv
[2022]

Mask3d for 3d semantic instance segmentation

Jonas Schult, Francis Engelmann, Alexander Hermans, Or Litany, Siyu Tang, and Bastian Leibe. Mask3d for 3d semantic instance segmentation. arXiv preprint arXiv:2210.03105,

arXiv
[2023]

Mamba24/8d: Enhancing global interaction in point clouds via state space model

Zhuoyuan Li, Yubo Ai, Jiahao Lu, ChuXin Wang, Jiacheng Deng, Hanzhi Chang, Yanzhe Liang, Wenfei Yang, Shifeng Zhang, and Tianzhu Zhang. Mamba24/8d: Enhancing global interaction in point clouds via state space model. arXiv preprint arXiv:2406.17442,

arXiv
[2024]

3d instances as 1d kernels

Yizheng Wu, Min Shi, Shuaiyuan Du, Hao Lu, Zhiguo Cao, and Weicai Zhong. 3d instances as 1d kernels. In Computer Vision–ECCV 2022: 17th European Conference, Tel Aviv, Israel, October 23–27, 2022, Proceedings, Part XXIX, pp. 235–252. Springer,

2022

Pith tools

Reviewed August 8, 2026 · model on record in the stance chip above.

[1] [7]

BSNet: Box-Supervised Simulation-assisted Mean Teacher for 3D Instance Segmentation

Jiahao Lu, Jiacheng Deng, and Tianzhu Zhang. Bsnet: Box-supervised simulation-assisted mean teacher for 3d instance segmentation. arXiv preprint arXiv:2403.15019, 2024a. Jiahao Lu, Jiacheng Deng, Ruijie Zhu, Yanzhe Liang, Wenfei Yang, Tianzhu Zhang, and Xu Zhou. Dn-4dgs: Denoised deformable network with temporal-spatial aggregation for dynamic scene rende...

work page Pith review arXiv

[2] [11]

Monst3r: A simple approach for estimating geometry in the presence of motion

Junyi Zhang, Charles Herrmann, Junhwa Hur, Varun Jampani, Trevor Darrell, Forrester Cole, Deqing Sun, and Ming-Hsuan Yang. Monst3r: A simple approach for estimating geometry in the presence of motion. arXiv preprint arXiv:2410.03825,

arXiv

[3] [12]

Maskgroup: Hierarchical point grouping and masking for 3d instance segmentation

Min Zhong, Xinghao Chen, Xiaokang Chen, Gang Zeng, and Yunhe Wang. Maskgroup: Hierarchical point grouping and masking for 3d instance segmentation. In 2022 IEEE International Conference on Multimedia and Expo (ICME), pp. 1–6. IEEE,

work page 2022

[4] [13]

Motiongs: Exploring explicit motion guidance for deformable 3d gaussian splatting

Ruijie Zhu, Yanzhe Liang, Hanzhi Chang, Jiacheng Deng, Jiahao Lu, Wenfei Yang, Tianzhu Zhang, and Yongdong Zhang. Motiongs: Exploring explicit motion guidance for deformable 3d gaussian splatting. arXiv preprint arXiv:2410.07707,

arXiv

[5] [14]

A.1 O VERVIEW This supplementary material provides more model and experimental details to understand our pro- posed method

A A PPENDIX You may include other additional sections here. A.1 O VERVIEW This supplementary material provides more model and experimental details to understand our pro- posed method. After that, we present more experiments to demonstrate the effectiveness of our methods. Finally, we show a rich visualization of our modules. A.2 M ORE MODEL DETAILS Sparse...

work page 2017

[6] [16]

The specific reasons for this performance decrease are twofold

However, if we apply NMS to the concatenated outputs, as described in Section 1 lines 63-65, a significant decrease in performance occur. The specific reasons for this performance decrease are twofold. Firstly, NMS heavily relies on confidence scores, retaining only the masks with the highest confidence among the duplicates. However, these confidence scor...

work page 2022

[7] [17]

Method mAP bathtub bed bookshe

As indicated by the table, the significant performance improvement highlights the effectiveness of our method in managing denser point cloud scenes across a broader range of categories. Method mAP bathtub bed bookshe. cabinet chair counter curtain desk door other picture frige s. curtain sink sofa table toilet window SoftGroup Vu et al. (2022)45.8 66.6 48...

work page 2022

[8] [21]

A.7 A SSETS AVAILABILITY The datasets that support the findings of this study are available in the following repositories: ScanNetV2 Dai et al

Based on the results, we find that the combination 0.5, 1, 1, 0.5, 0.5 yields the best performance. A.7 A SSETS AVAILABILITY The datasets that support the findings of this study are available in the following repositories: ScanNetV2 Dai et al. (2017) at http://www.scan-net.org/changelog# scannet-v2-2018-06-11 under the ScanNet Terms of Use. ScanNet200 Roz...

work page 2017