Pith. sign in

REVIEW 4 major objections 5 minor 2 cited by

Frozen geometry-grounded VFM features, routed through a small adapter, raise Gaussian occupancy IoU from 30.6 to 34.4 on nuScenes.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-02 18:38 UTC pith:F5CRZHGY

load-bearing objection A useful empirical adapter for Gaussian occupancy, but the paper's central claim that geometric grounding is the cause is not supported by its own ablations. the 4 major comments →

arxiv 2603.06210 v2 pith:F5CRZHGY submitted 2026-03-06 cs.CV cs.RO

VG3S: Visual Geometry Grounded Gaussian Splatting for Semantic Occupancy Prediction

Xiaoyang Yan , Muleilan Pei , Shaojie Shen This is my paper
classification cs.CV cs.RO
keywords semantic occupancy prediction3D Gaussian splattingvisual foundation modelsfrozen backbonefeature adaptercross-view geometryautonomous drivingnuScenes
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The paper claims that frozen, geometry-grounded visual foundation models (VFMs) can materially improve Gaussian-based 3D semantic occupancy prediction when their token outputs are adapted through a small learned module. On the nuScenes benchmark, the proposed VG3S raises scene-completion IoU from 30.56 to 34.41 and semantic mIoU from 20.02 to 21.52 over the GaussianFormer-2 baseline. The gain is attributed to cross-view 3D geometric priors—relative depth, structural boundaries, multi-view correspondence—embedded in the VFM's frozen features. This matters because it suggests expensive VFM fine-tuning can be avoided while still harvesting the VFM's geometric understanding, and because the adapter is plug-and-play across multiple VFMs.

Core claim

The paper's central claim is that the geometric ceiling of Gaussian-based semantic occupancy prediction can be raised by injecting features from a frozen, geometry-grounded Vision Foundation Model (VFM) through a learned hierarchical adapter, without fine-tuning the VFM. Using the GaussianFormer-2 decoder as base, the proposed VG3S adapter (GATF, TATR, LSFP) transforms the VFM's multi-scale tokens into occupancy-specific spatial features, yielding nuScenes IoU of 34.41 vs 30.56 and mIoU of 21.52 vs 20.02. The authors attribute the gain to cross-view 3D geometric priors inherited from VFM pre-training, and show the adapter works across several VFMs, with the most geometry- and driving-specifi

What carries the argument

The Hierarchical Geometric Feature Adapter (HGFA) is the load-bearing component: it groups consecutive VFM layers, computes per-layer softmax fusion weights to aggregate them (GATF), refines the aggregated tokens with per-group feed-forward blocks of varying hidden sizes (TATR), and builds a multi-scale spatial pyramid via depthwise and pointwise convolutions plus positional embeddings (LSFP). These steps convert the frozen VFM's latent tokens, produced by alternating-attention blocks with camera, register, and image tokens, into features the Gaussian decoder can consume.

Load-bearing premise

The claim rests on the assumption that the performance gap over the baseline comes from the VFM's geometric priors; the comparison against a non-geometric DPT adapter is not parameter-matched, so the gain could come partly from the adapter's larger capacity rather than from geometric information.

What would settle it

Train VG3S with the same HGFA but a frozen backbone of comparable parameter count that was pre-trained without geometric supervision (e.g., an image-classification or masked-image-modeling model). If the IoU and mIoU gains over GaussianFormer-2 remain at roughly 34.4 and 21.5, the paper's attribution to geometry-grounded VFM features is falsified. Alternatively, remove the geometric attention blocks in a geometry VFM, keeping the parameter count fixed, and re-run the benchmark.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

Share X Bluesky LinkedIn Reddit HN

If this is right

  • Frozen VFMs can be used as drop-in geometric feature extractors for occupancy prediction, removing the need to train an image encoder from scratch.
  • The hierarchical adapter generalizes across VFMs (DINOv2, VGGT, DGGT, DINOv3, DVGT), so upgrades in foundation models can be swapped in without redesigning the occupancy head.
  • The largest per-class gains are in structural categories—drivable surface, manmade, vegetation—indicating that geometric priors specifically improve large planar and vertical structures.
  • Because the VFM is frozen, the adapter-training compute is modest relative to end-to-end VFM fine-tuning, which the paper contrasts with concurrent work that fine-tunes the backbone.

Where Pith is reading between the lines

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

  • The paper's control for the adapter (a single DPT layer) has far fewer trainable parameters than HGFA; a parameter-matched adapter with neutral features is needed to separate the geometric-prior effect from mere capacity. The published ablation does not isolate the source of the gain.
  • If the geometric-prior account is right, the same adapter should transfer to other dense 3D tasks (monocular depth, surface normals, BEV segmentation) where cross-view geometry helps, potentially reducing the need for dense volumetric labels.
  • A direct test: keep HGFA fixed, but replace the VFM with a same-sized backbone trained without multi-view geometric objectives; if the IoU gain over GaussianFormer-2 persists, the improvement is not specific to geometry-grounded pre-training.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. VG3S is a system for 3D semantic occupancy prediction on nuScenes that replaces the learned image encoder of GaussianFormer-2 with a frozen vision foundation model (VFM) and inserts a learnable Hierarchical Geometric Feature Adapter (HGFA) between the VFM tokens and the Gaussian decoder. HGFA consists of Grouped Adaptive Token Fusion (GATF), Task-Aligned Token Refinement (TATR), and a Latent Spatial Feature Pyramid (LSFP). The paper reports IoU 34.41 vs 30.56 and mIoU 21.52 vs 20.02 over GaussianFormer-2, claims that the gains come from the geometric priors embedded in the frozen VFM, and includes ablations over VFM choice, HGFA components, and grouping hyperparameter K.

Significance. If the central attribution claim is correct, the paper would show that frozen geometry-grounded VFMs can be injected into Gaussian-based occupancy prediction without fine-tuning, yielding substantial accuracy gains. The work is practically appealing: it evaluates several VFMs, keeps the backbone frozen, and provides a detailed component-level ablation. The paper also explicitly ships a plug-and-play adapter that could transfer to other Gaussian decoders. However, the evidence as presented does not isolate geometric grounding from model capacity or generic VFM features, so the significance of the specific mechanism is not yet established.

major comments (4)
  1. [IV-B2 / Table II] The claim that 'the primary performance gains stem directly from the injected 3D geometric priors' is not supported by the table. VG3S-DINOv3, which has no explicit geometric training objective, achieves 33.20 IoU / 21.36 mIoU, comparable to VG3S-VGGT (33.29 / 21.10) and VG3S-DGGT (33.37 / 20.81), and has the second-best mIoU overall. The only clear margin is against DINOv2, but DINOv2 differs in architecture, scale, and pretraining data. Please report multi-seed variance and add matched-capacity non-geometry VFM controls to separate geometric content from generic feature quality.
  2. [IV-C1 / Table III] The 'w/o HGFA' ablation replaces HGFA with a standard DPT layer, but the two are not parameter-matched. HGFA adds GATF MLPs, TATR FFNs, LSFP depthwise/pointwise convolutions and SE blocks; DPT is a single decoder block. The observed drop (33.29 to 30.59 IoU, 21.10 to 19.31 mIoU) could be explained by decreased trainable capacity rather than by loss of geometric adaptation. A parameter-matched control (e.g., an MLP or linear projection with the same parameter count) and a no-adapter frozen-token baseline are necessary.
  3. [IV-A3 / Tables I-V] All numerical results are single-run validation numbers. Hyperparameters K, ρ_k, D_H^k, and τ_k are selected on the same validation set used for final reporting, and several differences in Table II (e.g., 0.08 IoU between VGGT and DGGT) are small enough to be within random seed variation. Please report means and standard deviations over at least three seeds and, where possible, results on a held-out test split. Without this, the magnitude of the headline improvement over GaussianFormer-2 is credible but the finer attribution claims are not.
  4. [Abstract / IV-A3] The paper emphasizes computational efficiency of Gaussian-based methods, but no runtime, parameter count, or memory comparison is provided for VG3S versus GaussianFormer-2. Since HGFA inserts substantial parameters and multi-scale convolutions, it is possible that the framework is no longer 'substantially reduce computational overhead' as advertised. Please report FLOPs, latency, and parameter counts for the full pipeline.
minor comments (5)
  1. [Abstract] Typo: 'V isual Geometry' should be 'Visual Geometry'.
  2. [III-C, Eq. (2)] Camera tokens are attributed to [24], but [24] is the 'Vision Transformers need registers' paper, which does not introduce camera tokens. Please correct the citation (likely VGGSfM or VGGT-related work).
  3. [III-C] 'DINO backbone [19], [20]' is ambiguous; DINO, DINOv2, and DINOv3 are different models. Specify which model and which patch size is used for each VFM variant.
  4. [Fig. 3] The qualitative figure uses small text and thin boxes; please enlarge or provide zoomed insets so the claimed differences are legible.
  5. [II-C] The phrase 'training-free manner' is misleading: only the VFM is frozen; the adapter and decoder are trained. Suggest rewording to 'without fine-tuning the VFM backbone' throughout.

Circularity Check

0 steps flagged

No significant circularity; the derivation is empirical and the main weakness is an attribution/control concern, not a definitional reduction.

full rationale

VG3S is an empirical systems paper: the reported gains (Table I: IoU 34.41 vs 30.56; mIoU 21.52 vs 20.02) come from training the proposed adapter on nuScenes and evaluating on held-out validation frames. Nothing in the equations (1)-(13) defines the target metric in terms of a fitted parameter; the HGFA weights are optimized with CE and Lovasz-Softmax losses (Eq. 13), and the occupancy labels come from SurroundOcc, not from the VFM features or from the paper's own prior work. The method reuses existing components (frozen VGGT/DVGT/DINOv3 encoders, GaussianFormer-2 decoder, DPT layer, SE blocks, sinusoidal PE); citing [13], [21], [22], [27] for these is ordinary engineering and not a self-citation chain. The only self-citations ([2] in the intro, [29] in view-guided deformable attention) are contextual/building-block references and are not load-bearing. The legitimate concern is that Table II shows VG3S-DINOv3 (no geometric training objective) matching VGGT/DGGT on IoU and exceeding them on mIoU, while Table III's DPT control is not parameter-matched to HGFA; this weakens the attribution of gains specifically to geometric grounding, and the lack of error bars makes small gaps hard to interpret. But that is an experimental-control and statistical-robustness issue, not circularity: the paper does not assert a mathematical identity between the VFM's geometric pretraining and the occupancy improvement, nor does it fit a quantity and then rename it as a prediction. Thus the circularity pass is negative; I set score 1 only to acknowledge mild self-citations, not because any claim reduces to its inputs.

Axiom & Free-Parameter Ledger

7 free parameters · 5 axioms · 1 invented entities

The central claim rests on a substantial empirical fit: the adapter and decoder weights plus multiple hand-selected hyperparameters (K, ρ, D_H, τ) are all tuned on nuScenes validation. The paper assumes VFMs transfer geometric priors, that SurroundOcc labels are trustworthy, and that validation-set performance is representative. No new physical entities are postulated; the adapter modules are architectural inventions validated only internally.

free parameters (7)
  • HGFA and Gaussian decoder trainable weights
    All adapter and decoder parameters are learned on nuScenes; these constitute the empirical fit.
  • Loss coefficients λ and β (Eq. 13) = not reported
    Required for training but no values are given, so the exact objective is underspecified.
  • Token group count K and group size M = K=4, M=6
    Chosen by validation performance (Table V); K is explicitly tuned.
  • TATR expansion ratios ρ_k = {4, 3, 2, 1.5}
    Hand-selected hyperparameters affecting adapter capacity; no sensitivity analysis.
  • LSFP hidden dimensions D_H^k = {768, 512, 384, 256}
    Hand-selected hyperparameters for multi-scale feature pyramid.
  • LSFP spatial scale factors τ_k = {4, 2, 1, 0.5}
    Chosen to define the multi-resolution feature pyramid; not derived.
  • Gaussian primitive count J and channel D = J=25,600, D=128
    Follows prior GaussianFormer-2 design; a fixed architectural choice.
axioms (5)
  • domain assumption Frozen VFMs (DVGT, VGGT, DGGT, DINOv3) encode transferable cross-view 3D geometric priors in their intermediate tokens.
    The whole method depends on this; the paper cites pretraining results but does not independently establish that the priors transfer to occupancy labels.
  • domain assumption SurroundOcc dense voxel labels are accurate enough to serve as ground truth for training and evaluation.
    Adopted from the benchmark; if these pseudo-labels are noisy, reported improvements may reflect fitting label noise.
  • domain assumption The Gaussian-to-voxel splatting formulation from GaussianFormer-2 is a valid and sufficient occupancy decoder.
    Taken from prior work without re-derivation; the claim relies on this decoder being adequate.
  • ad hoc to paper The DPT-based adapter used in the 'w/o HGFA' ablation is a fair non-HGFA baseline.
    DPT has far fewer parameters than HGFA, so the ablation does not control for capacity.
  • domain assumption Validation-set performance on nuScenes generalizes to unseen scenes.
    All hyperparameters and final numbers come from the same validation set; no held-out test evaluation is reported.
invented entities (1)
  • HGFA (GATF, TATR, LSFP modules) no independent evidence
    purpose: Adapt frozen VFM tokens into occupancy-specific geometry-enhanced features for the Gaussian decoder.
    The efficacy of these modules is shown only in this paper's own nuScenes experiments; no third-party or external validation is provided.

pith-pipeline@v1.3.0-alltime-deepseek · 11747 in / 10533 out tokens · 92560 ms · 2026-08-02T18:38:51.079065+00:00 · methodology

0 comments
Cite this review

Pith. "Pith review of VG3S: Visual Geometry Grounded Gaussian Splatting for Semantic Occupancy Prediction." pith.science (2026). https://pith.science/paper/F5CRZHGY

@misc{pith2026260306210,
  author       = {Pith},
  title        = {Pith review of: VG3S: Visual Geometry Grounded Gaussian Splatting for Semantic Occupancy Prediction},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/F5CRZHGY}},
  note         = {Machine review of arXiv:2603.06210}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

3D semantic occupancy prediction has become a crucial perception task for comprehensive scene understanding in autonomous driving. While recent advances have explored 3D Gaussian splatting for occupancy modeling to substantially reduce computational overhead, the generation of high-quality 3D Gaussians relies heavily on accurate geometric cues, which are often insufficient in purely vision-centric paradigms. To bridge this gap, we advocate for injecting the strong geometric grounding capability from Vision Foundation Models (VFMs) into occupancy prediction. In this regard, we introduce Visual Geometry Grounded Gaussian Splatting (VG3S), a novel framework that empowers Gaussian-based occupancy prediction with cross-view 3D geometric grounding. Specifically, to fully exploit the rich 3D geometric priors from a frozen VFM, we propose a plug-and-play hierarchical geometric feature adapter, which can effectively transform generic VFM tokens via feature aggregation, task-specific alignment, and multi-scale restructuring. Extensive experiments on the nuScenes occupancy benchmark demonstrate that VG3S achieves remarkable improvements of 12.6% in IoU and 7.5% in mIoU over the baseline. Furthermore, we show that VG3S generalizes seamlessly across diverse VFMs, consistently enhancing occupancy prediction accuracy and firmly underscoring the immense value of integrating priors derived from powerful, pre-trained geometry-grounded VFMs.

Figures

Figures reproduced from arXiv: 2603.06210 by Muleilan Pei, Shaojie Shen, Xiaoyang Yan.

Figure 1
Figure 1. Figure 1: Comparison between existing Gaussian-based methods and our proposed VG3S. Existing approaches often produce semantic occupancy with incomplete object coverage due to the lack of accurate 3D geometric priors. In contrast, our VG3S incorporates rich 3D geometric priors embedded in a frozen VFM pre-trained on massive datasets, enabling the decoder to generate more geometrically accurate and consistent semanti… view at source ↗
Figure 2
Figure 2. Figure 2: Framework overview of VG3S. Our approach leverages a powerful, pre-trained frozen VFM to provide rich 3D geometric priors, empowering the downstream Gaussian-based decoder with cross-view 3D geometric grounding and thereby significantly improving 3D semantic occupancy prediction. for self-supervised representation learning, enabling the dis￾covery of semantic correspondences without human anno￾tations. Sub… view at source ↗
Figure 3
Figure 3. Figure 3: Qualitative comparison between the baseline GaussianFormer-2 [11] and our proposed VG3S. Our approach produces more geometrically accurate and consistent object structures across four challenging scenes compared to the baseline, demonstrating that leveraging strong 3D geometric priors embedded within VFMs significantly improves 3D semantic occupancy predictions. TABLE III EFFECT OF THE HGFA. Method HGFA Io… view at source ↗

discussion (0)

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

Forward citations

Cited by 2 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Learning Adaptive Semantic Gaussian Allocation for 3D Occupancy

    cs.CV 2026-07 conditional novelty 6.0

    A learned candidate-scoring and global Top-K selection module improves 3D semantic occupancy prediction and Gaussian budget utilization on nuScenes and KITTI-360.

  2. VGOcc: Learning Visual-Geometric Gaussians for Vision-Centric 3D Driving Occupancy Prediction

    cs.CV 2026-07 conditional novelty 5.0

    VGOcc fuses frozen VGGT/DINOv2 features with sparse 3D Gaussians to reach 34.07 SC IoU and 21.75 SSC mIoU on nuScenes, besting prior vision-only occupancy methods.

Reference graph

Works this paper leans on

33 extracted references · 3 linked inside Pith · cited by 2 Pith papers

  1. [1]

    Renderocc: Vision-centric 3d occupancy prediction with 2d rendering supervision,

    M. Pan, J. Liu, R. Zhang, P. Huang, X. Li, H. Xie, B. Wang, L. Liu, and S. Zhang, “Renderocc: Vision-centric 3d occupancy prediction with 2d rendering supervision,” inICRA, 2024

    2024

  2. [2]

    Sept: Standard-definition map enhanced scene perception and topology reasoning for autonomous driving,

    M. Pei, J. Shan, P. Li, J. Shi, J. Huo, Y . Gao, and S. Shen, “Sept: Standard-definition map enhanced scene perception and topology reasoning for autonomous driving,”IEEE Robotics and Automation Letters, 2025

    2025

  3. [3]

    Detr3d: 3d object detection from multi-view images via 3d-to-2d queries,

    Y . Wang, V . C. Guizilini, T. Zhang, Y . Wang, H. Zhao, and J. Solomon, “Detr3d: 3d object detection from multi-view images via 3d-to-2d queries,” inCoRL, 2022

    2022

  4. [4]

    Stereo neural vernier caliper,

    S. Li, Z. Liu, Z. Shen, and K.-T. Cheng, “Stereo neural vernier caliper,” inAAAI, 2022

    2022

  5. [5]

    Learning better represen- tations for crowded pedestrians in offboard lidar-camera 3d tracking- by-detection,

    S. Li, P. Li, Q. Lian, P. Yun, and X. Chen, “Learning better represen- tations for crowded pedestrians in offboard lidar-camera 3d tracking- by-detection,” inICRA, 2025

    2025

  6. [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,” inCVPR, 2023

    2023

  7. [7]

    Octnet: Learning deep 3d representations at high resolutions,

    G. Riegler, A. O. Ulusoy, and A. Geiger, “Octnet: Learning deep 3d representations at high resolutions,” inCVPR, 2017

    2017

  8. [8]

    Bevformer: Learning bird’s-eye-view representation from multi-camera images via spatiotemporal transformers,

    Z. Li, W. Wang, H. Li, E. Xie, C. Sima, T. Lu, Y . Qiao, and J. Dai, “Bevformer: Learning bird’s-eye-view representation from multi-camera images via spatiotemporal transformers,” inECCV, 2022

    2022

  9. [9]

    Goirl: Graph-oriented inverse reinforcement learning for multimodal trajectory prediction,

    M. Pei, S. Shi, L. Zhang, P. Li, and S. Shen, “Goirl: Graph-oriented inverse reinforcement learning for multimodal trajectory prediction,” inICML, 2025

    2025

  10. [10]

    Gaussianformer: Scene as gaussians for vision-based 3d semantic occupancy predic- tion,

    Y . Huang, W. Zheng, Y . Zhang, J. Zhou, and J. Lu, “Gaussianformer: Scene as gaussians for vision-based 3d semantic occupancy predic- tion,” inECCV, 2024

    2024

  11. [11]

    Gaussianformer-2: Probabilistic gaussian superposition for efficient 3d occupancy prediction,

    Y . Huang, A. Thammatadatrakoon, W. Zheng, Y . Zhang, D. Du, and J. Lu, “Gaussianformer-2: Probabilistic gaussian superposition for efficient 3d occupancy prediction,” inCVPR, 2025

    2025

  12. [12]

    3d gaussian splatting for real-time radiance field rendering,

    B. Kerbl, G. Kopanas, T. Leimk ¨uhler, and G. Drettakis, “3d gaussian splatting for real-time radiance field rendering,”ACM Transactions on Graphics, 2023

    2023

  13. [13]

    Vggt: Visual geometry grounded transformer,

    J. Wang, M. Chen, N. Karaev, A. Vedaldi, C. Rupprecht, and D. Novotny, “Vggt: Visual geometry grounded transformer,” inCVPR, 2025

    2025

  14. [14]

    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,” inICCV, 2023

    2023

  15. [15]

    Surroundocc: Multi-camera 3d occupancy prediction for autonomous driving,

    Y . Wei, L. Zhao, W. Zheng, Z. Zhu, J. Zhou, and J. Lu, “Surroundocc: Multi-camera 3d occupancy prediction for autonomous driving,” in ICCV, 2023

    2023

  16. [16]

    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,” inCVPR, 2023

    2023

  17. [17]

    Nerf: Representing scenes as neural radiance fields for view synthesis,

    B. Mildenhall, P. P. Srinivasan, M. Tancik, J. T. Barron, R. Ramamoor- thi, and R. Ng, “Nerf: Representing scenes as neural radiance fields for view synthesis,”Communications of the ACM, 2021

    2021

  18. [18]

    Emerging properties in self-supervised vision trans- formers,

    M. Caron, H. Touvron, I. Misra, H. J ´egou, J. Mairal, P. Bojanowski, and A. Joulin, “Emerging properties in self-supervised vision trans- formers,” inICCV, 2021

    2021

  19. [19]

    Dinov2: Learning robust visual features without supervision,

    M. Oquab, T. Darcet, T. Moutakanni, H. V o, M. Szafraniec, V . Khalidov, P. Fernandez, D. Haziza, F. Massa, A. El-Nouby, et al., “Dinov2: Learning robust visual features without supervision,” arXiv:2304.07193, 2023

    Pith/arXiv arXiv 2023

  20. [20]

    Sim ´eoni, H

    O. Sim ´eoni, H. V . V o, M. Seitzer, F. Baldassarre, M. Oquab, C. Jose, V . Khalidov, M. Szafraniec, S. Yi, M. Ramamonjisoa,et al., “Dinov3,” arXiv:2508.10104, 2025

    Pith/arXiv arXiv 2025

  21. [21]

    Dvgt: Driving visual geometry transformer,

    S. Zuo, Z. Xie, W. Zheng, S. Xu, F. Li, S. Jiang, L. Chen, Z.- X. Yang, and J. Lu, “Dvgt: Driving visual geometry transformer,” arXiv:2512.16919, 2025

    Pith/arXiv arXiv 2025

  22. [22]

    Dggt: Feedforward 4d reconstruction of dynamic driving scenes using unposed images,

    X. Chen, Z. Xiong, Y . Chen, G. Li, N. Wang, H. Luo, L. Chen, H. Sun, B. Wang, G. Chen,et al., “Dggt: Feedforward 4d reconstruction of dynamic driving scenes using unposed images,”arXiv:2512.03004, 2025

    arXiv 2025

  23. [23]

    Vg3t: Visual geometry grounded gaussian transformer,

    J. Kim and S. Lee, “Vg3t: Visual geometry grounded gaussian transformer,”arXiv:2512.05988, 2025

    arXiv 2025

  24. [24]

    Vision transform- ers need registers,

    T. Darcet, M. Oquab, J. Mairal, and P. Bojanowski, “Vision transform- ers need registers,” inICLR, 2024

    2024

  25. [25]

    Vggsfm: Visual geometry grounded deep structure from motion,

    J. Wang, N. Karaev, C. Rupprecht, and D. Novotny, “Vggsfm: Visual geometry grounded deep structure from motion,” inCVPR, 2024

    2024

  26. [26]

    Squeeze-and-excitation networks,

    J. Hu, L. Shen, and G. Sun, “Squeeze-and-excitation networks,” in CVPR, 2018

    2018

  27. [27]

    Vision transformers for dense prediction,

    R. Ranftl, A. Bochkovskiy, and V . Koltun, “Vision transformers for dense prediction,” inICCV, 2021

    2021

  28. [28]

    Viewformer: Exploring spatiotemporal modeling for multi-view 3d occupancy per- ception via view-guided transformers,

    J. Li, X. He, C. Zhou, X. Cheng, Y . Wen, and D. Zhang, “Viewformer: Exploring spatiotemporal modeling for multi-view 3d occupancy per- ception via view-guided transformers,” inECCV, 2024

    2024

  29. [29]

    St-gs: Vision-based 3d semantic occupancy prediction with spatial-temporal gaussian splatting,

    X. Yan, M. Pei, and S. Shen, “St-gs: Vision-based 3d semantic occupancy prediction with spatial-temporal gaussian splatting,” in ICRA, 2026

    2026

  30. [30]

    The lov ´asz- softmax loss: A tractable surrogate for the optimization of the intersection-over-union measure in neural networks,

    M. Berman, A. Rannen Triki, and M. B. Blaschko, “The lov ´asz- softmax loss: A tractable surrogate for the optimization of the intersection-over-union measure in neural networks,” inCVPR, 2018

    2018

  31. [31]

    Monoscene: Monocular 3d semantic scene completion,

    A.-Q. Cao and R. de Charette, “Monoscene: Monocular 3d semantic scene completion,” inCVPR, 2022

    2022

  32. [32]

    Atlas: End-to-end 3d scene reconstruction from posed images,

    Z. Murez, T. van As, J. Bartolozzi, A. Sinha, V . Badrinarayanan, and Rabinovich, “Atlas: End-to-end 3d scene reconstruction from posed images,” inECCV, 2020

    2020

  33. [33]

    nuscenes: A multimodal dataset for autonomous driving,

    H. Caesar, V . Bankiti, A. H. Lang, S. V ora, V . E. Liong, Q. Xu, A. Krishnan, Y . Pan, G. Baldan, and O. Beijbom, “nuscenes: A multimodal dataset for autonomous driving,” inCVPR, 2020

    2020