Pith. sign in

REVIEW 3 major objections 5 minor 43 references

VOIC: Visible-Occluded Integrated Guidance for 3D Semantic Scene Completion

T0 review · 3 major / 5 minor · reviewed 2026-08-03 · deepseek-v4-flash

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

Pith's one-line read A visible-first, occlusion-second decoder design improves single-image 3D semantic scene completion, reporting new state of the art on SemanticKITTI and KITTI-360.

desk verdict VRLE dual-decoder is a genuine engineering recipe, but the paper's 'monocular' claim leans on MobileStereo's stereo depth, and the VD↔OD reverse path credited in the ablation never appears in the architecture description. read the letter →

arxiv 2512.18954 v6 pith:EI3SJA4Q submitted 2025-12-22 cs.CV

Zaidao Han , Risa Higashita , Jiang Liu This is my paper
classification cs.CV
keywords 3Dsemanticscenecompletionmonocularvisible-occludeddecouplingvisibilitylabelextractiondual-decodernetworkautonomousdrivingvoxelsupervision
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

Monocular 3D semantic scene completion must reconstruct both what a single image shows and what it hides. This paper argues that decoding all voxels uniformly dilutes features and propagates errors, so it separates the task: a Visible Decoder is trained with offline visibility labels extracted from ground truth, and an Occlusion Decoder uses the visible predictions as priors to complete the full scene. The visible labels come from VRLE, a z-buffered projection that marks which voxels the camera actually sees. On SemanticKITTI and SSCBench-KITTI-360, the method reports higher geometric IoU and semantic mIoU than prior monocular methods, with the clearest gains on small, long-tailed objects.

What carries the argument

The load-bearing object is the VRLE visibility mask: each occupied ground-truth voxel's eight vertices are projected into the image, rasterized with a z-buffer, and a voxel is marked visible if it is the nearest surface at any covered pixel. This mask splits supervision into visible-only labels for the Visible Decoder and full labels for the Occlusion Decoder. The network couples a Visible Embedding Feature Constructor (VEFC), which lifts 2D image features into 3D via deformable attention on depth-derived occupancy, with two interacting decoders: the Visible Decoder builds instance-aware priors under VRLE supervision, and the Occlusion Decoder reuses VEFC with visible features as content que

What would settle it

Replace the stereo-derived depth input with a purely monocular depth estimator at inference and re-evaluate on SemanticKITTI: if mIoU falls toward the level of prior single-frame methods, the reported gains come from stereo geometry rather than from the visible/occluded decoupling; if it holds, the decoupling claim is supported.

Watch

Extended reading notes

Core claim

The central claim is that explicitly decoupling visible-region perception from occluded-region completion—rather than decoding all voxels uniformly—produces better geometry and semantics in monocular semantic scene completion. The authors introduce Visible Region Label Extraction (VRLE), which projects the dense ground-truth voxel grid through the camera model with z-buffering to produce a per-voxel visibility mask. The Visible Decoder is supervised only on visible voxels; the Occlusion Decoder then uses the visible decoder's refined features and instance queries as priors, and bidirectional interaction lets global context flow back to refine visible predictions. In ablations, adding VRLE su

Load-bearing premise

The load-bearing premise is that the input is truly monocular; the paper obtains its depth prior from a stereo-matching network, so if the right image is used at inference the system is not single-image and comparisons to monocular baselines are weakened.

Editorial extensions

If this is right

  • If the visible-first separation is correct, any semantic scene completion method that treats all voxels uniformly is leaving accuracy on the table, and VRLE-style supervision is a cheap way to recover it.
  • The dual-decoder interaction makes the final prediction explicitly built from visible priors, so occluded-region errors can be traced back to visible perception, making the model more interpretable.
  • The reported gains are strongest on long-tail classes such as cars, trucks, and cyclists, suggesting visibility decoupling particularly helps small, easily confused objects.
  • Because the method is single-frame, it reaches or exceeds the accuracy of some multi-frame baselines without needing temporal history, which matters for real-time autonomous driving systems.
  • VRLE labels are generated offline from existing 3D ground truth, so the supervision strategy can be applied to any dataset that already has dense voxel annotations.

Reading between the lines

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

  • I infer that VRLE offers a cheap, general supervision signal: any camera-based 3D occupancy or depth model with dense ground truth could add a visible/occluded split, potentially improving small-object accuracy without architectural changes.
  • I infer the visible-first principle generalizes beyond the single-frame setting: multi-frame systems could first establish a confident visible estimate per frame and then complete the union of occluded areas, rather than fusing frames blindly.
  • I infer that the monocular claim is testable independently of the decoupling design: if the released model at inference uses the right stereo image for depth, its comparison to single-image baselines should be re-run with a true single-image depth estimator.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The paper proposes VOIC, a dual-decoder network for monocular 3D semantic scene completion. It introduces an offline Visible Region Label Extraction (VRLE) strategy that derives visible-voxel supervision from dense 3D ground truth, and a Visible Decoder / Occlusion Decoder architecture that first predicts visible regions and then completes the full scene. Additional components include VEFC, a depth-informed 2D-to-3D feature constructor, and cross-attention modules. Experiments on SemanticKITTI and SSCBench-KITTI360 report state-of-the-art IoU and mIoU. The central claim is that the visible-first, occlusion-second decoupling with bidirectional VD-OD interaction, enabled by VRLE, is the cause of the improvement over monocular baselines.

Significance. If the central claim were fully supported, the paper would make a useful contribution: the visible/occluded decoupling is a plausible and technically interesting decomposition of SSC, and the ablations suggest that both VEFC and VRLE contribute. The paper also provides public code and follows standard benchmark protocols. However, two load-bearing issues undermine the main claim: (1) the method is described as monocular but uses depth from MobileStereo, a stereo-matching network, so the comparison to monocular baselines in Tables I–II is not apples-to-apples; and (2) the claimed bidirectional VD-to-OD interaction in Table V is not described in the architecture, leaving the largest ablation gain unsubstantiated. These are not presentation issues but concerns about what the experiments actually demonstrate.

major comments (3)
  1. [Abstract; Sec. IV-B; Tables I–II] The paper claims monocular SSC from a single RGB image, but Sec. IV-B states that depth is obtained from MobileStereo [39] predictions. MobileStereo is a stereo-matching network that requires a left/right image pair to produce depth. If the right image is used at inference, the system is not single-image and the comparison to MonoScene, CGFormer, and other monocular baselines in Tables I–II is unequal. This is not a minor labeling issue: Table IV shows that VEFC, the depth-consuming module, contributes an 8.02 IoU gain (36.86 to 44.88), larger than the remaining visible/occluded contributions, so the stereo-derived depth is a plausible major confound. The abstract and Sec. III-A formulation Θ(I_rgb) must be reconciled with the actual input modality, or the comparisons must be redone against stereo-based SSC methods.
  2. [Sec. III-D, III-E; Table V] Table V claims that the best configuration uses a bidirectional interaction flow (VD↔OD), with the reverse OD-to-VD link raising mIoU from 17.02 to 17.64. However, the architecture description in Secs. III-D and III-E only specifies a forward flow: VD produces features/priors, and OD consumes them. No module or equation is provided for propagating OD's global context back into VD. Without a concrete description of the reverse path, the ablation result cannot be attributed to a bidirectional interaction, and the central 'interactive' claim is unsupported. The authors should either add the missing component and its equations or re-interpret Table V as a forward-only design.
  3. [Sec. IV-D, Table IV] The ablation progression in Table IV does not cleanly isolate the contributions claimed in the text. Method 2 adds VD+OD while omitting VRLE and uses 'labels produced by OD' to supervise VD, which is a different supervision scheme from the VRLE-trained VD in the final model. The text does not define how OD labels are produced or why this is a fair intermediate step. This makes it difficult to attribute the final +1.55 mIoU to VRLE alone, as opposed to the change in supervision target. Please clarify the supervision setup for Method 2 and consider an ablation that changes only the VRLE label.
minor comments (5)
  1. [Sec. IV-B] The statement 'In our camera-only setting, RGB images serve as the sole input modality' is inconsistent with the use of MobileStereo depth predictions. Please define precisely what inputs are fed at inference.
  2. [Eq. (2)] The visibility condition uses d_(u,v)^(i) = min over j of d_(u,v)^(j). This is a standard z-buffer test, but the notation is under-specified: d_(u,v)^(j) is only defined for voxels whose projection covers (u,v). State that the minimum is over the set of voxels covering that pixel.
  3. [Sec. III-F] The loss notation L_geo_scal and L_ce are cited from [1] and [7], but the precise definitions are not given. Since λ_miou=10 is a scaled weight, include the loss definitions or point to the exact equations in the cited works.
  4. [Table III] The method name 'Symphonize' appears in Table III and in Sec. IV-C, while the reference and Table I use 'Symphonies'. Please unify the spelling and the citation.
  5. [Sec. IV-D, Table IV] In Table IV, the baseline and Method 1 use 'OD' as the decoder, but the architecture section describes OD as the occlusion decoder taking VD features. Clarify what the OD-only configuration does without a VD.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: VOIC's VRLE supervision and dual-decoder gains are empirically grounded on held-out benchmarks; the only self-citation is non-load-bearing.

full rationale

VOIC's claimed derivation chain is self-contained with respect to the benchmark evaluations. VRLE is an offline preprocessing step that derives a binary visibility mask M_vis from the ground-truth voxel annotations using standard camera projection and Z-buffering (Eqs. 1-2); it does not use the network's predictions or the test labels to define the model output. The visible decoder is supervised on Y_vis = Y ⊙ M_vis and the occlusion decoder on the full Y, with losses (Eqs. 10-12) that are standard geometric/CE/mIoU losses; no benchmark constant or metric is fitted through these losses. The state-of-the-art claims in Tables I-II are measured on held-out SemanticKITTI hidden test and SSCBench-KITTI360 test sets, so the final numbers are not forced by construction. Ablations (Tables IV-V) compare architecture variants under the same evaluation protocol, supporting the visible-first/occlusion-second attribution empirically rather than by definition. The only self-citation (ref. [3]) is a general applications citation and is not load-bearing. The MobileStereo depth prior noted in Sec. IV-B and the documentation gap between the described VD->OD flow and the claimed VD<->OD 'reverse link' in Sec. IV-D are correctness/reproducibility concerns (potential input-modality confound and under-specification), not circular derivations: they do not equate any predicted quantity to an input by construction.

Assumptions & free parameters 2 free parameters · 5 assumptions · 0 invented entities

The central empirical claim rests on standard dataset assumptions, on the correctness of the VRLE projection/visibility test, and on the availability of reliable depth from MobileStereo. The only hand-set scalars that directly affect training are the loss weights and VRLE sampling stride; network weights are learned. No new physical or ontological entities are introduced; all components are learned modules.

free parameters (2)
  • Loss weights (lambda_scal, lambda_ce, lambda_miou) = 1, 1, 10
    Set by hand; mIoU loss weighted ten times the others; not swept in the paper.
  • VRLE sampling stride delta = 4
    Controls candidate-pixel density in visibility rasterization; chosen for efficiency, not validated.
assumptions (5)
  • domain assumption VRLE pinhole projection with Z-buffering (Eq. 1-2) correctly determines which ground-truth voxels are visible.
    If false, VD receives corrupt visible-region labels and the decoupling argument collapses; no independent validation of the visibility mask is provided.
  • domain assumption MobileStereo depth is an acceptable geometric prior in a 'monocular' pipeline.
    VEFC constructs the occupancy mask from D_pre; if D_pre comes from a stereo pair, the single-image premise is violated.
  • domain assumption SemanticKITTI and SSCBench-KITTI360 dense voxel annotations are complete enough to produce reliable visible/occluded supervision and benchmark labels.
    VRLE derives visible targets from these labels; incompleteness would propagate into both training and metrics.
  • ad hoc to paper The VD<->OD bidirectional interaction exists in the implemented model.
    Table V reports its gain, but Section III-E only describes VD-to-OD flow; no reverse path is specified in the methods.
  • domain assumption MaskDINO-initialized backbone and DETR-style attention transfer from 2D segmentation to SSC.
    The method relies on pretrained MaskDINO weights and deformable attention; no analysis of domain shift is given.

how reviews work

0 comments
Cite this review

Pith. "Pith review of VOIC: Visible-Occluded Integrated Guidance for 3D Semantic Scene Completion." pith.science (2026). https://pith.science/paper/EI3SJA4Q

@misc{pith2026251218954,
  author       = {Pith},
  title        = {Pith review of: VOIC: Visible-Occluded Integrated Guidance for 3D Semantic Scene Completion},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/EI3SJA4Q}},
  note         = {Machine review of arXiv:2512.18954}
}
read the original abstract

Camera-based 3D Semantic Scene Completion (SSC) is a critical task for autonomous driving and robotic scene understanding. It aims to infer a complete 3D volumetric representation of both semantics and geometry from a single image. Existing methods typically focus on end-to-end 2D-to-3D feature lifting and voxel completion. However, they often overlook the interference between high-confidence visible-region perception and low-confidence occluded-region reasoning caused by single-image input, which can lead to feature dilution and error propagation. To address these challenges, we introduce an offline Visible Region Label Extraction (VRLE) strategy that explicitly separates and extracts voxel-level supervision for visible regions from dense 3D ground truth. This strategy purifies the supervisory space for two complementary sub-tasks: visible-region perception and occluded-region reasoning. Building on this idea, we propose the Visible-Occluded Interactive Completion Network (VOIC), a novel dual-decoder framework that explicitly decouples SSC into visible-region semantic perception and occluded-region scene completion. VOIC first constructs a base 3D voxel representation by fusing image features with depth-derived occupancy. The visible decoder focuses on generating high-fidelity geometric and semantic priors, while the occlusion decoder leverages these priors together with cross-modal interaction to perform coherent global scene reasoning. Extensive experiments on the SemanticKITTI and SSCBench-KITTI360 benchmarks demonstrate that VOIC outperforms existing monocular SSC methods in both geometric completion and semantic segmentation accuracy, achieving state-of-the-art performance.

Figures

Figures reproduced from arXiv: 2512.18954 by the authors.

Figure 1
Figure 1. Overview of the proposed VOIC framework. Unlike conventional [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Overall architecture of the VOIC framework. (a) The model follows a progressive visible–occluded paradigm that decouples the monocular [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Sparse Voxel Feature Initialization. The VEFC module creates a [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Qualitative results on the SemanticKITTI validation set. VOIC enhances overall scene classification through high-quality visible-range semantic priors [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]

Review history (2 revisions) →

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

43 extracted references · 1 linked inside Pith

  1. [39]

    Mobilestereonet: Towards lightweight deep networks for stereo matching,

    F. Shamsafar, S. Woerz, R. Rahim, and A. Zell, “Mobilestereonet: Towards lightweight deep networks for stereo matching,” inProceedings of the Ieee/Cvf Winter Conference on Applications of Computer Vision, 2022, pp. 2417–2426

    2022

  2. [1]

    Monoscene: Monocular 3d semantic scene completion,

    A.-Q. Cao and R. De Charette, “Monoscene: Monocular 3d semantic scene completion,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2022, pp. 3991–4001

    2022

  3. [2]

    S3cnet: A sparse semantic scene completion network for lidar point clouds,

    R. Cheng, C. Agia, Y . Ren, X. Li, and L. Bingbing, “S3cnet: A sparse semantic scene completion network for lidar point clouds,” in Conference on Robot Learning, 2021, pp. 2148–2161

    2021

  4. [3]

    Multi-path sensory substitution device navigates the blind and visually impaired individuals,

    Z. Han, S. Li, X. Wang, X. Hu, R. Higashita, and J. Liu, “Multi-path sensory substitution device navigates the blind and visually impaired individuals,”Displays, p. 103200, 2025

    2025

  5. [4]

    LODE: Locally Conditioned Eikonal Implicit Scene Completion from Sparse LiDAR,

    P. Li, R. Zhao, Y . Shi, H. Zhao, J. Yuan, G. Zhou, and Y .-Q. Zhang, “LODE: Locally Conditioned Eikonal Implicit Scene Completion from Sparse LiDAR,” in2023 IEEE International Conference on Robotics and Automation (ICRA), 2023, pp. 8269–8276

    2023

  6. [5]

    Semcity: Semantic scene generation with triplane diffusion,

    J. Lee, S. Lee, C. Jo, W. Im, J. Seon, and S.-E. Yoon, “Semcity: Semantic scene generation with triplane diffusion,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 28 337–28 347

    2024

  7. [6]

    V oxformer: Sparse voxel transformer for camera- based 3d semantic scene completion,

    Y . Li, Z. Yu, C. Choy, C. Xiao, J. M. Alvarez, S. Fidler, C. Feng, and A. Anandkumar, “V oxformer: Sparse voxel transformer for camera- based 3d semantic scene completion,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023, pp. 9087–9098

    2023

  8. [7]

    Symphonize 3d semantic scene completion with contextual instance queries,

    H. Jiang, T. Cheng, N. Gao, H. Zhang, T. Lin, W. Liu, and X. Wang, “Symphonize 3d semantic scene completion with contextual instance queries,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 20 258–20 267

    2024

Show all 43 references
  1. [8]

    Semantickitti: A dataset for semantic scene understanding of lidar sequences,

    J. Behley, M. Garbade, A. Milioto, J. Quenzel, S. Behnke, C. Stachniss, and J. Gall, “Semantickitti: A dataset for semantic scene understanding of lidar sequences,” inProceedings of the IEEE/CVF International Conference on Computer Vision, 2019, pp. 9297–9307

    2019

  2. [9]

    Sscbench: A large-scale 3d semantic scene completion benchmark for autonomous driving,

    Y . Li, S. Li, X. Liu, M. Gong, K. Li, N. Chen, Z. Wang, Z. Li, T. Jiang, and F. Yu, “Sscbench: A large-scale 3d semantic scene completion benchmark for autonomous driving,” in2024 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2024, pp. 13 333– 13 340

    2024

  3. [10]

    Semantic scene completion from a single depth image,

    S. Song, F. Yu, A. Zeng, A. X. Chang, M. Savva, and T. Funkhouser, “Semantic scene completion from a single depth image,” inProceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2017, pp. 1746–1754

    2017

  4. [11]

    3d sketch-aware semantic scene completion via semi-supervised structure prior,

    X. Chen, K.-Y . Lin, C. Qian, G. Zeng, and H. Li, “3d sketch-aware semantic scene completion via semi-supervised structure prior,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2020, pp. 4193–4202

    2020

  5. [12]

    Rgbd based dimensional decomposition residual network for 3d semantic scene completion,

    J. Li, Y . Liu, D. Gong, Q. Shi, X. Yuan, C. Zhao, and I. Reid, “Rgbd based dimensional decomposition residual network for 3d semantic scene completion,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2019, pp. 7693–7702

    2019

  6. [13]

    Cascaded context pyra- mid for full-resolution 3d semantic scene completion,

    P. Zhang, W. Liu, Y . Lei, H. Lu, and X. Yang, “Cascaded context pyra- mid for full-resolution 3d semantic scene completion,” inProceedings of the IEEE/CVF International Conference on Computer Vision, 2019, pp. 7801–7810

    2019

  7. [14]

    Openoccupancy: A large scale benchmark for surrounding semantic occupancy perception,

    X. Wang, Z. Zhu, W. Xu, Y . Zhang, Y . Wei, X. Chi, Y . Ye, D. Du, J. Lu, and X. Wang, “Openoccupancy: A large scale benchmark for surrounding semantic occupancy perception,” inProceedings of the IEEE/CVF International Conference on Computer Vision, 2023, pp. 17 850–17 859

    2023

  8. [15]

    Lmscnet: Lightweight multiscale 3d semantic completion,

    L. Roldao, R. De Charette, and A. Verroust-Blondet, “Lmscnet: Lightweight multiscale 3d semantic completion,” in2020 International Conference on 3D Vision (3DV), 2020, pp. 111–119

    2020

  9. [16]

    Sparse single sweep lidar point cloud segmentation via learning contextual shape priors from scene completion,

    X. Yan, J. Gao, J. Li, R. Zhang, Z. Li, R. Huang, and S. Cui, “Sparse single sweep lidar point cloud segmentation via learning contextual shape priors from scene completion,” inProceedings of the AAAI Conference on Artificial Intelligence, vol. 35, 2021, pp. 3101–3109

    2021

  10. [17]

    A multi-phase camera-LiDAR fusion network for 3D semantic segmentation with weak supervision,

    X. Chang, H. Pan, W. Sun, and H. Gao, “A multi-phase camera-LiDAR fusion network for 3D semantic segmentation with weak supervision,” IEEE Transactions on Circuits and Systems for Video Technology, vol. 33, no. 8, pp. 3737–3746, 2023

    2023

  11. [18]

    LiDAR-camera continuous fusion in voxelized grid for semantic scene completion,

    Z. Lu, B. Cao, and Q. Hu, “LiDAR-camera continuous fusion in voxelized grid for semantic scene completion,”IEEE Transactions on Circuits and Systems for Video Technology, 2024

    2024

  12. [19]

    Occformer: Dual-path transformer for vision-based 3d semantic occupancy prediction,

    Y . Zhang, Z. Zhu, and D. Du, “Occformer: Dual-path transformer for vision-based 3d semantic occupancy prediction,” inProceedings of the IEEE/CVF International Conference on Computer Vision, 2023, pp. 9433–9443. 10

    2023

  13. [20]

    Ndc- scene: Boost monocular 3d semantic scene completion in normalized device coordinates space,

    J. Yao, C. Li, K. Sun, Y . Cai, H. Li, W. Ouyang, and H. Li, “Ndc- scene: Boost monocular 3d semantic scene completion in normalized device coordinates space,” in2023 IEEE/CVF International Conference on Computer Vision (ICCV), 2023, pp. 9421–9431

    2023

  14. [21]

    Tri-perspective view for vision-based 3d semantic occupancy prediction,

    Y . Huang, W. Zheng, Y . Zhang, J. Zhou, and J. Lu, “Tri-perspective view for vision-based 3d semantic occupancy prediction,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023, pp. 9223–9232

    2023

  15. [22]

    Not all voxels are equal: Hardness-aware semantic scene completion with self- distillation,

    S. Wang, J. Yu, W. Li, W. Liu, X. Liu, J. Chen, and J. Zhu, “Not all voxels are equal: Hardness-aware semantic scene completion with self- distillation,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 14 792–14 801

    2024

  16. [23]

    Instance-aware monocular 3D semantic scene completion,

    H. Xiao, H. Xu, W. Kang, and Y . Li, “Instance-aware monocular 3D semantic scene completion,”IEEE Transactions on Intelligent Trans- portation Systems, vol. 25, no. 7, pp. 6543–6554, 2024

    2024

  17. [24]

    Mixssc: Forward- backward mixture for vision-based 3d semantic scene completion,

    M. Wang, Y . Ding, Y . Liu, Y . Qin, R. Li, and Z. Tang, “Mixssc: Forward- backward mixture for vision-based 3d semantic scene completion,”IEEE Transactions on Circuits and Systems for Video Technology, 2025

    2025

  18. [25]

    Hierarchical Temporal Context Learning for Camera-Based Semantic Scene Completion,

    B. Li, J. Deng, W. Zhang, Z. Liang, D. Du, X. Jin, and W. Zeng, “Hierarchical Temporal Context Learning for Camera-Based Semantic Scene Completion,” inComputer Vision – ECCV 2024, A. Leonardis, E. Ricci, S. Roth, O. Russakovsky, T. Sattler, and G. Varol, Eds., Cham, 2025, vol....

    2024

  19. [26]

    CurriFlow: Curriculum-Guided Depth Fusion with Optical Flow-Based Temporal Alignment for 3D Semantic Scene Completion,

    J. Lin, J. Zhou, W. Xu, R. Xu, C. Wang, S. Chen, K. Fu, Y . Shao, L. Guo, and S. Xu, “CurriFlow: Curriculum-Guided Depth Fusion with Optical Flow-Based Temporal Alignment for 3D Semantic Scene Completion,” Oct. 2025

    2025

  20. [27]

    One Step Closer: Creating the Future to Boost Monocular Semantic Scene Completion,

    H. Lu, Y . Su, X. Zhang, and H. Hu, “One Step Closer: Creating the Future to Boost Monocular Semantic Scene Completion,” Jul. 2025

    2025

  21. [28]

    Unleashing Semantic and Geometric Priors for 3D Scene Completion,

    S. Chen, W. Sui, B. Zhang, Z. Boukhers, J. See, and C. Yang, “Unleashing Semantic and Geometric Priors for 3D Scene Completion,” Aug. 2025

    2025

  22. [29]

    MVFormer: UNet-like Transformer with Mix-V oxel Attention for Camera-Based 3D Semantic Scene Completion,

    F. Gao, Y . Chen, K. Wang, P. Zhou, and J. Lu, “MVFormer: UNet-like Transformer with Mix-V oxel Attention for Camera-Based 3D Semantic Scene Completion,”IEEE Transactions on Circuits and Systems for Video Technology, 2025

    2025

  23. [30]

    Semi-supervised 3D Semantic Scene Completion with 2D Vision Foundation Model Guidance,

    D.-H. Pham, D.-D. Nguyen, A. Pham, T. Ho, P. Nguyen, K. Nguyen, and R. Nguyen, “Semi-supervised 3D Semantic Scene Completion with 2D Vision Foundation Model Guidance,” inProceedings of the AAAI Conference on Artificial Intelligence, vol. 39, 2025, pp. 6514–6522

    2025

  24. [31]

    SPHERE: Semantic-PHysical Engaged REpre- sentation for 3D Semantic Scene Completion,

    Z. Yang and Y . Peng, “SPHERE: Semantic-PHysical Engaged REpre- sentation for 3D Semantic Scene Completion,” inProceedings of the 33rd ACM International Conference on Multimedia, Dublin Ireland, Oct. 2025, pp. 7681–7690

    2025

  25. [32]

    Memory-Augmented Re-Completion for 3D Semantic Scene Completion,

    Y .-W. Tseng, S.-P. Yang, J.-C. Wu, I.-B. Liao, Y .-H. Li, H.-H. Shuai, and W.-H. Cheng, “Memory-Augmented Re-Completion for 3D Semantic Scene Completion,” inProceedings of the AAAI Conference on Artificial Intelligence, vol. 39, 2025, pp. 7446–7454

    2025

  26. [33]

    Mask dino: Towards a unified transformer-based framework for object detection and segmentation,

    F. Li, H. Zhang, H. Xu, S. Liu, L. Zhang, L. M. Ni, and H.-Y . Shum, “Mask dino: Towards a unified transformer-based framework for object detection and segmentation,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023, pp. 3041–3050

    2023

  27. [34]

    Deformable DETR: Deformable Transformers for End-to-End Object Detection,

    X. Zhu, W. Su, L. Lu, B. Li, X. Wang, and J. Dai, “Deformable DETR: Deformable Transformers for End-to-End Object Detection,” Mar. 2021

    2021

  28. [35]

    Deeplab: Semantic image segmentation with deep convolutional nets, atrous convolution, and fully connected crfs,

    L.-C. Chen, G. Papandreou, I. Kokkinos, K. Murphy, and A. L. Yuille, “Deeplab: Semantic image segmentation with deep convolutional nets, atrous convolution, and fully connected crfs,”IEEE transactions on pattern analysis and machine intelligence, vol. 40, no. 4, pp. 834–848, 2017

    2017

  29. [36]

    Beverse: Unified perception and prediction in birds-eye-view for vision- centric autonomous driving,

    Y . Zhang, Z. Zhu, W. Zheng, J. Huang, G. Huang, J. Zhou, and J. Lu, “Beverse: Unified perception and prediction in birds-eye-view for vision- centric autonomous driving,”arXiv preprint arXiv:2205.09743, 2022

    2022 arXiv

  30. [37]

    Bevdepth: Acquisition of reliable depth for multi-view 3d object detec- tion,

    Y . Li, Z. Ge, G. Yu, J. Yang, Z. Wang, Y . Shi, J. Sun, and Z. Li, “Bevdepth: Acquisition of reliable depth for multi-view 3d object detec- tion,” inProceedings of the AAAI Conference on Artificial Intelligence, vol. 37, 2023, pp. 1477–1485

    2023

  31. [38]

    BEVDet: High- performance Multi-camera 3D Object Detection in Bird-Eye-View,

    J. Huang, G. Huang, Z. Zhu, Y . Ye, and D. Du, “BEVDet: High- performance Multi-camera 3D Object Detection in Bird-Eye-View,” Jun. 2022

    2022

  32. [40]

    Decoupled Weight Decay Regularization,

    I. Loshchilov and F. Hutter, “Decoupled Weight Decay Regularization,” Jan. 2019

    2019

  33. [41]

    Deep residual learning for image recognition,

    K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learning for image recognition,” inProceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2016, pp. 770–778

    2016

  34. [42]

    Camera-based 3d semantic scene completion with sparse guidance network,

    J. Mei, Y . Yang, M. Wang, J. Zhu, J. Ra, Y . Ma, L. Li, and Y . Liu, “Camera-based 3d semantic scene completion with sparse guidance network,”IEEE Transactions on Image Processing, 2024

    2024

  35. [43]

    Context and geometry aware voxel transformer for semantic scene completion,

    Z. Yu, R. Zhang, J. Ying, J. Yu, X. Hu, L. Luo, S.-Y . Cao, and H.- L. Shen, “Context and geometry aware voxel transformer for semantic scene completion,”Advances in Neural Information Processing Systems, vol. 37, pp. 1531–1555, 2024

    2024

Pith tools

Reviewed August 3, 2026 · model on record in the stance chip above.