Pith. sign in

REVIEW 2 major objections 4 minor 48 references

Learning to select semantic Gaussians under a fixed budget improves 3D occupancy prediction.

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

2026-08-01 06:25 UTC pith:LCBOGMM2

load-bearing objection Solid fixed-budget selection method with clean ablations, but the 'better-utilized' claim needs opacity-weighted diagnostics before I'd trust the bottleneck narrative. the 2 major comments →

arxiv 2607.21896 v1 pith:LCBOGMM2 submitted 2026-07-24 cs.CV

Learning Adaptive Semantic Gaussian Allocation for 3D Occupancy

Kanglin Ning , Yiran Zhao , Wenrui Li , Houde Quan , Qifan Li , Xingtao Wang , Xiaopeng Fan This is my paper
classification cs.CV
keywords 3D occupancy predictionsemantic Gaussian splattingGaussian allocationcandidate selectionfixed budgetTransformerautonomous 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.

This paper argues that the main bottleneck in semantic Gaussian occupancy prediction is not how primitives are shaped, initialized, or added, but which primitives survive when the total count is capped for memory and compute. It proposes SAGFormer, a Transformer that scores keep, clone, split, and suppress candidates for each Gaussian and globally selects a fixed final set. On nuScenes-SurroundOcc and SSCBench-KITTI-360, this selection step improves occupancy IoU and mIoU over matched baselines without increasing the final Gaussian count. The paper also shows that the selected sets have fewer unused Gaussians, less semantic mixing, and stronger class-consistent voxel support, indicating that explicit capacity allocation complements Gaussian refinement.

Core claim

The central claim is that semantic Gaussian allocation should be treated as a candidate-selection problem under a fixed budget, rather than left implicit in representation, initialization, or densification choices. SAGFormer encodes each Gaussian with its intrinsic attributes and local geometric-semantic context, generates four types of candidates (keep, clone, split, suppress), and applies global Top-K_B selection so that every sample uses exactly the same number of Gaussians. Under raw single-frame input on nuScenes-SurroundOcc, it reaches 41.74 IoU and 28.47 mIoU, outperforming the matched learned-allocation baseline without semantic cues by 2.27 IoU and 0.79 mIoU, and beating a capacity-

What carries the argument

The Semantic Gaussian Allocation Transformer (SAGFormer) is a candidate-selection mechanism. For each initialized Gaussian it builds an allocation feature from geometric attributes, semantic logits/confidence/entropy, and four local-context cues (neighbor count, mean distance, scale-normalized geometric overlap, semantic consistency), then a Transformer encoder models relations among Gaussians. A small router predicts weights for keep, clone, split, and suppress generators; all generated candidates are pooled and globally ranked by a gate-weighted opacity score, and the top K_B = ⌊(1+B)N⌋ candidates are kept at inference. Training uses a differentiable soft expected count, so the hard select

Load-bearing premise

The paper's central claim depends on the hand-defined support metrics (Unused, Mix-G, Sem-Sup with a 3 m assignment radius and a 0.6 majority fraction threshold) faithfully measuring allocation quality, and on the gains being attributable to the allocation policy rather than to the semantic-entropy regularizer that feeds the router.

What would settle it

A reader could settle the claim by running the matched ablation with the entropy regularizer removed while keeping the router inputs identical and checking whether Mix-G and Sem-Sup still improve by the same margins; if they do, the allocation policy is the driver, and if they do not, the paper's bottleneck narrative is weakened. Alternatively, permuting the router scores (while preserving the candidate pool) and measuring the accuracy drop would show whether the learned ranking itself is what matters.

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

Share X Bluesky LinkedIn Reddit HN

If this is right

  • The allocation step can be bolted onto an existing semantic Gaussian occupancy model: adding SAGFormer to a camera-only GaussianFormer control raised IoU/mIoU from 29.60/19.10 to 33.04/23.67 with the same final count.
  • Semantic logits, confidence, and entropy are the main driver of the gains; removing them drops mIoU by 0.79 on nuScenes while keeping everything else matched.
  • Each candidate family has a distinct role: clone improves class-consistent support, split reduces semantic mixing, suppress improves overall accuracy and utilization; combining all four gives the best result.
  • The budget parameter B trades accuracy against Gaussian count: mIoU peaks at B=0.6 (~16.6K Gaussians) and the soft-to-hard selection gap is small (0.031 mIoU drop at the default budget).

Where Pith is reading between the lines

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

  • The same candidate-selection machinery could in principle be applied to other sparse scene representations (e.g., superquadrics or voxel grids) whenever a hard budget must be enforced, since the scoring uses only attributes and local context.
  • The paper's diagnosis that high raw coverage coexists with redundancy and semantic mixing suggests that coverage-based quality metrics alone are insufficient for occupancy representations; purity-weighted support measures may be more meaningful for downstream tasks that need class-consistent geometry.
  • Because the entropy regularizer sharpens the per-candidate confidence that feeds the router, part of the measured reduction in Mix-G and increase in Sem-Sup may come from the loss rather than the selection policy; a controlled ablation varying L_ent while fixing the router input features would separate the two.
  • The fixed-budget global Top-K selection is a form of differentiable sorting/routing; in streaming or online settings, a greedy or batched approximation might preserve most of the gain at lower latency.

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

2 major / 4 minor

Summary. The paper proposes SAGFormer, a learned allocation module for semantic 3D Gaussian occupancy prediction. Given an initialized Gaussian set, SAGFormer builds per-Gaussian features from geometry, semantics, and local context, uses a transformer to model inter-Gaussian relations, and generates keep/clone/split/suppress candidates. A global Top-K_B selection under a fixed budget produces the final Gaussian set used for voxel decoding. The method is evaluated on nuScenes-SurroundOcc and SSCBench-KITTI-360, with matched baselines and ablations. The central empirical claim is that the full SAGFormer policy improves occupancy mIoU over capacity-matched no-allocation and weaker allocation baselines, while also improving hand-defined allocation diagnostics such as Unused, Mix-G, and Sem-Sup.

Significance. If the result holds, the paper addresses a genuinely under-studied problem: not how to represent, initialize, or densify semantic Gaussians, but how to allocate a fixed final budget among candidate primitives. The internal evidence is solid on the accuracy side: Table 5 compares matched allocation policies under the same initial set, backbone, decoder, training objectives, and final budget, and the full SAGFormer beats the strongest control by 0.44 mIoU with seed standard deviations around 0.04-0.06. The paper is also honest about its limitations, including the small dynamic-object gain. The main risk is that the diagnostic metrics used to support the 'better-utilized' narrative are opacity-blind, which the stress-test note correctly identifies. The paper would be strengthened by opacity-aware metrics; the core accuracy claim is likely to survive such a check, but the allocation-quality conclusion needs verification.

major comments (2)
  1. [Evaluation Metrics; Eqs. (7)-(8); Tables 4-5] The support metrics Unused, Mix-G, Sem-Sup, P_valid, and P_pen. are defined purely geometrically: each Gaussian covers GT occupied voxels within tau_d=3 m, and no opacity or router weight enters the assignment. A suppress candidate has opacity gamma_i * alpha_i (Eq. 8) with gamma_i in (0,1), so it can be arbitrarily close to zero; once selected by Top-K_B, it is counted exactly like a full-opacity Gaussian. Clone/split children with negative opacity residuals have the same issue. Thus the reported reductions in Unused and Mix-G and increases in Sem-Sup may reflect geometric placement of near-zero-weight primitives rather than actual use in the decoded occupancy. This is load-bearing for the paper's central 'better-utilized Gaussian representations' claim. Please recompute the diagnostics with opacity- or weight-aware coverage (e.g., weight each covered voxel by the candidate's effective
  2. [Training Loss (Eq. 11) and Eq. (4); Tables 4-5] The semantic-entropy regularizer L_ent directly sharpens each candidate's softmax, and the same sharpened quantities (q_i, kappa_i, h_i) are the router's semantic input features in Eq. (4). Part of the observed reduction in Mix-G and increase in Sem-Sup could therefore be due to the loss rather than to the allocation policy itself. I do not regard this as invalidating the primary controlled comparison, because the variants in Table 5 share identical training objectives, so the relative policy gain is credible. However, the paper's absolute narrative that allocation produces more semantically consistent representations would be strengthened by an ablation that removes L_ent or feeds detached, pre-sharpening logits to the router.
minor comments (4)
  1. [Evaluation Metrics] The support metrics rely on manually chosen thresholds rho=0.6 and tau_d=3, with sensitivity deferred to Supplementary Sec. C. Since these metrics drive the allocation-bottleneck narrative, a compact sensitivity table should appear in the main text.
  2. [Table 4] Table 4 compares methods at very different final Gaussian counts (e.g., GaussianFormer 25.6K vs SAGFormer 16.6K). The paper should clarify that this table is descriptive, while the controlled comparison is Table 5.
  3. [Eq. (7)] The claim that hard Top-K_B enforces an 'exact physical budget' is true at inference, but the suppress candidate's opacity can be near zero. A brief clarification that the budget counts all selected candidates regardless of effective opacity would help readers connect the selection rule to the opacity-aware issue raised above.
  4. [Limitations] The limitations paragraph is commendable and appropriately notes the lack of demonstrated dynamic-object gains. Consider adding a sentence on whether the allocation module itself is robust to sensor degradation or distribution shift, beyond the general deployment caveat.

Circularity Check

0 steps flagged

No significant circularity: SAGFormer's occupancy gains are measured against external benchmarks with matched baselines, and the allocation diagnostics are post-hoc and not fed back into training.

full rationale

The paper's central claims are empirical and externally benchmarked. SAGFormer improves IoU/mIoU on nuScenes-SurroundOcc and SSCBench-KITTI-360 relative to matched controls (Tables 2, 3, 5), and the final Gaussian set is decoded and evaluated against held-out voxel labels. No equation defines the target metric in terms of the router output; the support measures are explicitly 'used only for post-hoc evaluation and are never provided to the allocation module.' The L_ent regularizer could sharpen per-candidate confidence and affect Mix-G/Sem-Sup, but the paper includes ablations (Table 7) and matched-objective controls (Table 5), and it does not present L_ent as a prediction of allocation quality. The 'soft count ratio stays close to 1 + B' observation follows from the budget loss L_bud by construction, but it is reported as a training sanity check, not as an independent result. The paper's references are ordinary related-work citations; no load-bearing premise depends on a self-citation or an imported uniqueness theorem. The Limitations section explicitly admits that dynamic-object gains are not established (0.06 mIoU improvement), which is a scoping caveat, not circularity. The skeptic's opacity gap—support diagnostics ignore opacity, so near-zero-opacity suppress candidates can count as 'used'—is a legitimate measurement-validity concern about hand-defined metrics, but it does not reduce a prediction to its inputs: the metrics are post-hoc and not optimized by the model. Overall, the derivation chain is self-contained and externally evaluated; no circular step is exhibited.

Axiom & Free-Parameter Ledger

3 free parameters · 4 axioms · 0 invented entities

The central claim is empirical; the ledger captures the tuned hyperparameters and task assumptions. The main pull-from-upstream items are the standard training losses, the candidate generators taken from 3DGS densification, and the benchmark datasets. No new physical or conceptual entity is postulated; the method's modules are directly tested. The most assumption-heavy part is the diagnostic framework: the whole 'allocation bottleneck' narrative depends on the hand-defined Mix-G/Sem-Sup/Unused metrics, whose thresholds are not exhaustively justified in the main text.

free parameters (3)
  • Budget B = 0.6 (default, tuned on nuScenes validation; Table 8 sweep)
    Controls final Gaussian count K_B = floor((1+B)N); mIoU peaks at B=0.6 in Table 8, so it is a validation-selected hyperparameter rather than an intrinsic constant.
  • Loss weights λent, λrec, λbud, λlov = not stated in main text (deferred to Supplementary Sec. A)
    Weights in Eqs. (10)–(14) are tuned on validation; values are not in the arXiv text, which prevents exact reproduction and makes the sensitivity of the results to these choices untestable.
  • Support-metric thresholds ρ=0.6, τd=3 = ρ=0.6, τd=3 m
    Hand-chosen thresholds in §Evaluation Metrics define Unused/Mix-G/Sem-Sup; the paper says threshold tests are in Supplementary Sec. C, so the centrality of these choices to the 'allocation quality' narrative is not fully testable here.
axioms (4)
  • domain assumption Multi-modal calibrated inputs (six surround cameras + LiDAR) and voxel-wise supervised occupancy labels are available for training and evaluation.
    The method's initialization and losses (Eqs. 10–14) presuppose LiDAR-based Gaussian initialization and dense 3D occupancy ground truth; this is the standard task setup of nuScenes-SurroundOcc and SSCBench-KITTI-360.
  • domain assumption The voxel-to-Gaussian assignment rule with cutoff τd=3 m is a faithful measure of a Gaussian's support of occupied voxels.
    All allocation diagnostics (Unused, Mix-G, Sem-Sup, Pvalid) in Table 4 rest on this assignment; if the assignment mischaracterizes 'support', the allocation-bottleneck claim is weakened.
  • ad hoc to paper Candidate operations keep/clone/split/suppress with continuous gate weights during training are differentiable approximations that transfer to hard Top-K_B selection at inference.
    Eqs. (6)–(9) define soft routing with an expected count; the conversion to hard selection at inference is assumed to be approximately lossless, which is empirically checked in Table 8 but not proven.
  • ad hoc to paper Local geometric-semantic context features (neighbor density, distance, overlap ηij, semantic agreement qi·qj) contain sufficient signal to rank candidate quality.
    Equations (1)–(5) encode the assumption that these four hand-designed context cues can predict which Gaussians to keep/clone/split/suppress; the Transformer encoder then relies on that premise.

pith-pipeline@v1.3.0-alltime-deepseek · 11971 in / 18700 out tokens · 178607 ms · 2026-08-01T06:25:19.427335+00:00 · methodology

0 comments
Cite this review

Pith. "Pith review of Learning Adaptive Semantic Gaussian Allocation for 3D Occupancy." pith.science (2026). https://pith.science/paper/LCBOGMM2

@misc{pith2026260721896,
  author       = {Pith},
  title        = {Pith review of: Learning Adaptive Semantic Gaussian Allocation for 3D Occupancy},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/LCBOGMM2}},
  note         = {Machine review of arXiv:2607.21896}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Semantic 3D Gaussians provide a compact representation for 3D semantic occupancy prediction by rendering semantic primitives into a voxel volume under voxel-wise supervision. Recent methods have improved the modeling ability and efficiency of this representation through more flexible primitive shapes, geometry-guided initialization, and progressive densification. However, these advances mainly determine how primitives are represented, initialized, or added, and do not explicitly address how to select the most useful Gaussians when their total number must be limited to control memory and computation. This imbalance creates an allocation bottleneck: redundant Gaussians remain in simple regions, while difficult regions receive insufficient semantic support. We propose the Semantic Gaussian Allocation Transformer (SAGFormer), which uses Gaussian attributes and local geometric-semantic features to score candidates and select a fixed final Gaussian set. Experiments on nuScenes-SurroundOcc and SSCBench-KITTI-360 show that SAGFormer improves occupancy prediction under the evaluated protocols and yields more semantically consistent and better-utilized Gaussian representations. Under similar final counts and raw coverage, it reduces semantic mixing, strengthens class-consistent voxel support, and produces fewer unused Gaussians. The results indicate that explicit capacity allocation is a useful complement to Gaussian refinement for semantic occupancy prediction.

Figures

Figures reproduced from arXiv: 2607.21896 by Houde Quan, Kanglin Ning, Qifan Li, Wenrui Li, Xiaopeng Fan, Xingtao Wang, Yiran Zhao.

Figure 1
Figure 1. Figure 1: Semantic Gaussian allocation. Under a fixed Gaus [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Overall framework of SAGFormer. Multi-modal features initialize semantic Gaussians. The allocation module uses [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Allocation diagnostics on nuScenes-SurroundOcc: [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗

discussion (0)

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

Reference graph

Works this paper leans on

48 extracted references · 1 linked inside Pith

  1. [1]

    Communication, Simulation, and Intelligent Agents: Implications of Personal Intelligent Machines for Medical Education

    Clancey, William J. Communication, Simulation, and Intelligent Agents: Implications of Personal Intelligent Machines for Medical Education. Proceedings of the Eighth International Joint Conference on Artificial Intelligence (IJCAI-83)

  2. [2]

    Classification Problem Solving

    Clancey, William J. Classification Problem Solving. Proceedings of the Fourth National Conference on Artificial Intelligence

  3. [3]

    , title =

    Robinson, Arthur L. , title =. 1980 , doi =. https://science.sciencemag.org/content/208/4447/1019.full.pdf , journal =

    1980

  4. [4]

    New Ways to Make Microcircuits Smaller---Duplicate Entry

    Robinson, Arthur L. New Ways to Make Microcircuits Smaller---Duplicate Entry. Science

  5. [5]

    Clancey and Glenn Rennels , abstract =

    Diane Warner Hasling and William J. Clancey and Glenn Rennels , abstract =. Strategic explanations for a diagnostic consultation system , journal =. 1984 , issn =. doi:https://doi.org/10.1016/S0020-7373(84)80003-6 , url =

    doi:10.1016/s0020-7373(84)80003-6 1984

  6. [6]

    and Rennels, Glenn R

    Hasling, Diane Warner and Clancey, William J. and Rennels, Glenn R. and Test, Thomas. Strategic Explanations in Consultation---Duplicate. The International Journal of Man-Machine Studies

  7. [7]

    Poligon: A System for Parallel Problem Solving

    Rice, James. Poligon: A System for Parallel Problem Solving

  8. [8]

    Transfer of Rule-Based Expertise through a Tutorial Dialogue

    Clancey, William J. Transfer of Rule-Based Expertise through a Tutorial Dialogue

  9. [9]

    The Engineering of Qualitative Models

    Clancey, William J. The Engineering of Qualitative Models

  10. [10]

    2023 , eprint=

    Attention Is All You Need , author=. 2023 , eprint=

    2023

  11. [11]

    Pluto: The 'Other' Red Planet

    NASA. Pluto: The 'Other' Red Planet

  12. [12]

    and Vora, Sourabh and Liong, Venice Erin and Xu, Qiang and Krishnan, Anush and Pan, Yu and Baldan, Giancarlo and Beijbom, Oscar , booktitle =

    Caesar, Holger and Bankiti, Varun and Lang, Alex H. and Vora, Sourabh and Liong, Venice Erin and Xu, Qiang and Krishnan, Anush and Pan, Yu and Baldan, Giancarlo and Beijbom, Oscar , booktitle =

  13. [13]

    Liao, Yiyi and Xie, Jun and Geiger, Andreas , journal =

  14. [14]

    Li, Yiming and Li, Sihang and Liu, Xinhao and Gong, Moonjun and Li, Kenan and Chen, Nuo and Wang, Zijun and Li, Zhiheng and Jiang, Tao and Yu, Fisher and Wang, Yue and Zhao, Hang and Yu, Zhiding and Feng, Chen , booktitle =

  15. [15]

    2021 , eprint =

    MonoScene: Monocular 3D Semantic Scene Completion , author =. 2021 , eprint =

    2021

  16. [16]

    2023 , eprint =

    VoxFormer: Sparse Voxel Transformer for Camera-based 3D Semantic Scene Completion , author =. 2023 , eprint =

    2023

  17. [17]

    2023 , eprint =

    SurroundOcc: Multi-Camera 3D Occupancy Prediction for Autonomous Driving , author =. 2023 , eprint =

    2023

  18. [18]

    2023 , eprint =

    Occ3D: A Large-Scale 3D Occupancy Prediction Benchmark for Autonomous Driving , author =. 2023 , eprint =

    2023

  19. [19]

    2024 , eprint =

    GaussianFormer: Scene as Gaussians for Vision-Based 3D Semantic Occupancy Prediction , author =. 2024 , eprint =

    2024

  20. [20]

    2025 , eprint =

    GaussianFormer3D: Multi-Modal Gaussian-based Semantic Occupancy Prediction with 3D Deformable Attention , author =. 2025 , eprint =

    2025

  21. [21]

    2025 , eprint =

    GraphGSOcc: Semantic and Geometric Graph Transformer for 3D Gaussian Splating-based Occupancy Prediction , author =. 2025 , eprint =

    2025

  22. [22]

    2025 , eprint =

    ST-GS: Vision-Based 3D Semantic Occupancy Prediction with Spatial-Temporal Gaussian Splatting , author =. 2025 , eprint =

    2025

  23. [23]

    2026 , eprint =

    GaussianOcc3D: A Gaussian-Based Adaptive Multi-modal 3D Occupancy Prediction , author =. 2026 , eprint =

    2026

  24. [24]

    2026 , eprint =

    Gau-Occ: Geometry-Completed Gaussians for Multi-Modal 3D Occupancy Prediction , author =. 2026 , eprint =

    2026

  25. [25]

    2024 , eprint =

    Generative Densification: Learning to Densify Gaussians for High-Fidelity Generalizable 3D Reconstruction , author =. 2024 , eprint =

    2024

  26. [26]

    2026 , eprint =

    F4Splat: Feed-Forward Predictive Densification for Feed-Forward 3D Gaussian Splatting , author =. 2026 , eprint =

    2026

  27. [27]

    2026 , eprint =

    SplatWeaver: Learning to Allocate Gaussian Primitives for Generalizable Novel View Synthesis , author =. 2026 , eprint =

    2026

  28. [28]

    2025 , eprint =

    AnySplat: Feed-forward 3D Gaussian Splatting from Unconstrained Views , author =. 2025 , eprint =

    2025

  29. [29]

    2024 , eprint =

    Pixel-GS: Density Control with Pixel-aware Gradient for 3D Gaussian Splatting , author =. 2024 , eprint =

    2024

  30. [30]

    2024 , eprint =

    AbsGS: Recovering Fine Details for 3D Gaussian Splatting , author =. 2024 , eprint =

    2024

  31. [31]

    2024 , eprint =

    DAOcc: 3D Object Detection Assisted Multi-Sensor Fusion for 3D Occupancy Prediction , author =. 2024 , eprint =. doi:10.1109/TCSVT.2025.3610634 , url =

    arXiv 2024

  32. [32]

    2025 , eprint =

    SDGOCC: Semantic and Depth-Guided Bird's-Eye View Transformation for 3D Multimodal Occupancy Prediction , author =. 2025 , eprint =

    2025

  33. [33]

    2020 International Conference on 3D Vision (3DV) , year =

    LMSCNet: Lightweight Multiscale 3D Semantic Completion , author =. 2020 International Conference on 3D Vision (3DV) , year =. doi:10.1109/3DV50981.2020.00021 , url =

    arXiv 2020

  34. [34]

    arXiv preprint arXiv:2603.06210 , year=

    VG3S: Visual Geometry Grounded Gaussian Splatting for Semantic Occupancy Prediction , author=. arXiv preprint arXiv:2603.06210 , year=

    Pith/arXiv arXiv

  35. [35]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Generalizing Visual Geometry Priors to Sparse Gaussian Occupancy Prediction , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  36. [36]

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

    Autoocc: Automatic open-ended semantic occupancy annotation via vision-language guided gaussian splatting , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

  37. [37]

    arXiv e-prints , pages=

    Occgs: zero-shot 3d occupancy reconstruction with semantic and geometric-aware gaussian splatting , author=. arXiv e-prints , pages=

  38. [38]

    Advances in Neural Information Processing Systems , volume=

    Odg: Occupancy prediction using dual gaussians , author=. Advances in Neural Information Processing Systems , volume=

  39. [39]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    Vision-Only Gaussian Splatting for Collaborative Semantic Occupancy Prediction , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

  40. [40]

    2024 , eprint =

    GaussianFormer-2: Probabilistic Gaussian Superposition for Efficient 3D Occupancy Prediction , author =. 2024 , eprint =

    2024

  41. [41]

    Advances in Neural Information Processing Systems , year =

    QuadricFormer: Scene as Superquadrics for 3D Semantic Occupancy Prediction , author =. Advances in Neural Information Processing Systems , year =

  42. [42]

    The Fourteenth International Conference on Learning Representations , year =

    Progressive Gaussian Transformer with Anisotropy-aware Sampling for Open Vocabulary Occupancy Prediction , author =. The Fourteenth International Conference on Learning Representations , year =

  43. [43]

    2026 , doi =

    Qian, Rui and Cao, Haozhi and Deng, Tianchen and Yuan, Shenghai and Xie, Lihua , booktitle =. 2026 , doi =

    2026

  44. [44]

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

    Joint Semantic and Rendering Enhancements in 3D Gaussian Modeling with Anisotropic Local Encoding , author =. Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) , pages =. 2025 , url =

    2025

  45. [45]

    2025 , url =

    Zhang, Jiaxin and Jiang, Junjun and Chen, Youyu and Jiang, Kui and Liu, Xianming , booktitle =. 2025 , url =

    2025

  46. [46]

    2025 , eprint =

    SpatialSplat: Efficient Semantic 3D from Sparse Unposed Images , author =. 2025 , eprint =

    2025

  47. [47]

    2025 , eprint =

    Off The Grid: Detection of Primitives for Feed-Forward 3D Gaussian Splatting , author =. 2025 , eprint =

    2025

  48. [48]

    2026 , eprint =

    On the Design of Mixture-of-Experts for Dynamic Gaussian Splatting , author =. 2026 , eprint =

    2026