Pith. sign in

REVIEW 4 major objections 5 minor 32 references

Learning Gaussian Structure: Intervention-Guided Density Control for Feed-Forward Driving Reconstruction

T0 review · 4 major / 5 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read This paper claims that feed-forward Gaussian reconstruction of driving scenes can learn when to prune and add Gaussian primitives from intervention-induced changes in local gradient responses, and that the resulting policy plus cross-time…

desk verdict A genuinely new densification-supervision idea, but the intervention target is confounded by per-branch normalization, and the empirical claims need code and variance to fully check. read the letter →

arxiv 2608.11077 v1 pith:73QQUQKM submitted 2026-08-11 cs.CV

Hang Li , Jiahe Li , Meiying Gu , Jin Zheng , Lina Yu , Xiao Bai This is my paper · ORCID
classification cs.CV
keywords 3DGaussianSplattingfeed-forwardreconstructiondrivingscenedensificationpruninggradientresponsecross-timefeatureaggregationnovelviewsynthesis
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

Feed-forward Gaussian reconstruction of driving scenes typically keeps the initial LiDAR point-to-primitive mapping fixed, so a model can update attributes but never decide whether a primitive should be removed or duplicated. This paper introduces Learning Gaussian Structure (LGS), whose Gaussian Densify Policy predicts per-primitive prune and addition scores in a single forward pass. The training signal comes from controlled interventions: temporarily pruning or duplicating a primitive and measuring how the rendering gradients of its spatial neighbors change, arguing that a reduction in local gradient response reveals a beneficial structural adjustment. A second component, Cross-Time Point Query, retrieves features from Gaussian primitives at other timestamps so attribute prediction uses explicit cross-time evidence. If correct, feed-forward reconstruction gains the density-control ability previously limited to per-scene optimization.

What carries the argument

The load-bearing machinery is the Gaussian Densify Policy supervised by intervention-induced gradient responses. Concretely, after normalizing each attribute gradient by its maximum across primitives, the scalar response $r_j^B = \frac{1}{D}\sum_c \left| \frac{(\nabla_{g_j} \mathcal{L}^B)_c}{\max_l |(\nabla_{g_l} \mathcal{L}^B)_c| + \epsilon} \right|$ measures how strongly a nearby primitive $j$ responds to the rendering loss under branch $B$. The intervention targets $\Delta^{\text{prune}}_i$ and $\Delta^{\text{add}}_i$ sum the response changes over the $K$ nearest neighbors, and a signed-log, z-score, sigmoid transform turns them into per-primitive supervision targets $y^a_i \in [0,1]$. A confidence-weighted regression loss with weights $w_i = (2|y_i - 0.5|)^\gamma$ trains a Point Transformer V3 encoder to predict the Densify Map; during inference the intervention branches are discarded and thresholds $\theta_p = \theta_a = 0.7$ are applied to prune and add. The second component, Cross-Time Point Query, mean-pools features from the $K$ nearest primitives at other timestamps and fuses them with the decoder feature via a residual projection, feeding both the attribute head and the densify policy.

What would settle it

Take a trained LGS model and apply its prune/add policy to an out-of-distribution driving clip, such as one with substantially sparser LiDAR or a different camera rig, while holding the Gaussian count fixed by matching the action rate with random decisions. If the learned policy does not outperform the random baseline under these conditions, the intervention-response proxy has failed to transfer and the central claim that local gradient changes reveal beneficial structural adjustments is called into question.

Watch

Extended reading notes

Core claim

The paper's central claim is that structural decisions in feed-forward Gaussian reconstruction can be learned without per-scene optimization by treating prune and add operations as interventions and reading their effects on local rendering gradients. For a sampled primitive, the authors compute a scalar neighbor-gradient response $r_j^B$ for each branch (original, prune, add), then aggregate the differences $\Delta^{\text{prune}}_i$ and $\Delta^{\text{add}}_i$ over the primitive's $K$ spatial neighbors. These differences become training targets for the Densify Policy: a lower local gradient magnitude after an intervention means the operation helps reconstruction. At inference, the policy outputs a Densify Map of Prune and Addition Scores, and the Gaussian set is updated by pruning primitives above a threshold and adding perturbed copies of primitives whose addition score is high. On Waymo and PandaSet the full model reports consistent gains over feed-forward baselines, including full-image PSNR of 28.04 dB versus 26.28 dB for UniSplat on Waymo.

Load-bearing premise

The load-bearing premise is that the scalar neighbor-gradient response $r_j^B$, defined as the average normalized gradient magnitude over the primitive's $K$ spatial neighbors, is a valid and transferable proxy for whether pruning or adding that primitive helps reconstruction, so the sign of $\Delta^{\text{prune}}_i$ and $\Delta^{\text{add}}_i$ correctly supervises the policy.

Editorial extensions

If this is right

  • Feed-forward Gaussian reconstruction gains a densification mechanism previously available only to per-scene optimization, without requiring test-time optimization or intervention rendering.
  • The Densify Map adapts primitive count and placement to scene content: on Waymo the full model uses 584.2K Gaussians versus 519.8K for the baseline, with additions concentrated on vehicle surfaces and lane markings.
  • Because the policy is applied in a single forward pass, it adds modest inference cost (about 0.05 s per frame for the policy) while improving rendering quality.
  • Cross-Time Point Query adds explicit cross-time evidence for attribute prediction and is reported to be complementary to density adjustment, improving PSNR further when combined.

Reading between the lines

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

  • The intervention-response signal could serve as a general critic for other feed-forward 3D representations: any representation in which adding or removing an element changes neighbor gradients could be supervised the same way, including pixel-aligned Gaussians or voxel grids.
  • The paper's ablations show that random additions already improve PSNR over the fixed initialization, suggesting that part of the gain is restoring under-represented capacity; a natural test is whether the addition score correlates with local reconstruction error or feature uncertainty on out-of-distribution scenes.
  • A stronger transferability check would apply the trained policy to a dataset with substantially different LiDAR density or camera configuration without fine-tuning; the paper's premise predicts the prune/add decisions should remain sensible because they are grounded in gradient responses rather than dataset-specific statistics.
  • The signed-log transform and fixed thresholds introduce hyperparameters; an implicit extension, which the paper itself flags as a limitation, is to make the density adjustment scene-dependent or fully learnable.
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

4 major / 5 minor

Summary. The paper proposes LGS, a feed-forward framework for driving-scene Gaussian reconstruction. The method has two main components: a Gaussian Densify Policy that learns to prune or add Gaussian primitives using supervision derived from controlled interventions, and a Cross-Time Point Query that aggregates features from other timestamps. The interventions are applied to sampled primitives, and the resulting changes in neighbor gradient responses (Eqs. 3-5) are converted into training targets for the policy. At inference, the policy predicts prune/add scores and adjusts the Gaussian set without per-scene optimization. Experiments on Waymo and PandaSet report consistent improvements over prior feed-forward methods (e.g., full-image PSNR 28.04 vs 26.28 for UniSplat on Waymo).

Significance. If the intervention-based supervision is valid, the work is a meaningful step toward enabling feed-forward Gaussian reconstruction to adjust primitive composition without per-scene optimization. The two-dataset evaluation, the component ablations in Table 3, and the threshold analysis in Table 4 are useful empirical contributions. The central idea of using controlled prune/add interventions to generate training targets is novel and potentially transferable. However, the significance is conditional on resolving the normalization confound in Eq. (3) and on validating the target signal against actual reconstruction quality, not just downstream rendering metrics.

major comments (4)
  1. [§3.2, Eq. (3)] The per-branch global-max normalization confounds the intervention deltas in Eqs. (4)-(5). Because the original, prune, and addition branches contain different primitive sets, the denominator max_l |(grad_{g_l} L_B)_c| is computed over different sets in each branch. Pruning a large-gradient primitive can lower the denominator and inflate the normalized responses of its neighbors; adding a perturbed copy with a large gradient can raise the denominator and suppress neighbor responses. Thus Delta_prune and Delta_add mix the intended local response change with a branch-level rescaling artifact. Since the sign of the delta is mapped to the training target y_i, the policy may be learning to predict this scale shift rather than the local benefit of the intervention. Please either use a shared normalization reference across all branches (e.g., the original branch's maxima) or provide an analysis showing that the rescaling does not dominate the sign of the deltas.
  2. [§4.4, Table 4] The thresholds theta_p and theta_a are set to 0.7 based on the same Table 4 that supports the final performance claim, and no validation split is described. The table shows PSNR varying between 26.94 and 28.09 dB across threshold choices, so selecting thresholds on the test set can inflate the reported 28.04 dB result. Please select thresholds on a held-out validation set and report final numbers on the test split only after freezing the thresholds, or provide a sensitivity analysis with confidence intervals.
  3. [§3.2] The intervention targets are not directly validated against reconstruction quality. The text states that a negative Delta_i indicates that the operation reduces the local gradient response and a positive value indicates the opposite, but it never checks whether these signs correlate with actual changes in rendered PSNR/SSIM when the intervention is applied. Please add an oracle-style analysis on a held-out validation set: compare the learned prune/add decisions with decisions that maximize real render quality, and report the agreement rate. Without this, the causal interpretation of the improved rendering metrics remains indirect.
  4. [§3.4] The three-stage training creates a potential distribution mismatch for the policy. In Stage 2 the policy is trained on targets generated by the Stage-1 backbone, but in Stage 3 the backbone is fine-tuned while the policy is frozen. At inference, the policy consumes features produced by the fine-tuned backbone, whose distribution differs from the Stage-1 features seen in training. Please add an ablation that compares (i) the current pipeline with (ii) a variant that re-trains the policy after Stage 3, to show that the mismatch does not degrade the policy's decisions.
minor comments (5)
  1. [§3.2] The notation Delta_i is introduced without a subscript for prune/add; please use Delta_prune_i and Delta_add_i consistently throughout the text and equations.
  2. [Table 2] LGSbase is defined in the Table 2 caption as 'the representation before iterative refinement,' but the term is also used in Table 1 without a definition. Please define it once in Section 4.1 or in the main text.
  3. [Figures 6 and 7] The score maps in Figures 6 and 7 would be much more interpretable if they included a shared color scale and a quantitative measure of the claimed spatial complementarity between Addition Scores and Prune Scores.
  4. [§4.1] The inference time of 1.90 s per frame is reported, but it is unclear whether this includes the three-stage training procedure or only the feed-forward inference; please clarify.
  5. [Various tables] No variance or repeated-seed results are reported in Tables 1-4. Given that some comparisons are close (e.g., Table 3 baseline vs CTPQ-only), please report standard deviations over at least three runs or a significance test.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the intervention-derived targets are a self-supervised training signal, while the core claim is validated on held-out novel views with a random-action control.

full rationale

The paper's load-bearing claim is that prune/add interventions change local rendering-gradient responses, and that a learned Densify Policy can predict these changes to restructure Gaussians at inference. This is not circular: the intervention targets in Eqs. 4-5 are an intermediate self-supervised signal, not the final evaluation metric. The policy is trained to imitate the sign of gradient-response changes, and its usefulness is then tested against held-out novel-view PSNR/SSIM/LPIPS on Waymo and PandaSet, which are external to the target-construction process. Table 4 further includes a random-action baseline matched for action rates and Gaussian count (27.52 dB vs. 28.04 dB), showing that the learned locations, not merely the added capacity, drive the improvement. The branch-dependent normalization in Eq. 3 is a possible confound because the per-branch maximum in the denominator can change when primitives are pruned or added, so Delta_prune and Delta_add may partly reflect a global rescaling artifact rather than purely local response changes. However, this is a correctness/robustness concern about the proxy, not a case where the prediction is equivalent to its input by construction: the paper does not define reconstruction benefit as the gradient response itself. The threshold values in Table 4 are tuned on the target benchmark, but this is hyperparameter selection rather than a fitted parameter being renamed as a prediction, and the learned policy still must generalize beyond the tuning signal. No load-bearing self-citation chain or imported uniqueness theorem appears; citations to EVolSplat, Flux4D, and Point Transformer V3 supply architectural components, not the paper's central justification. Overall, the derivation is self-contained against external benchmarks, so no significant circularity is found.

Assumptions & free parameters 9 free parameters · 5 assumptions · 0 invented entities

The paper introduces no new physical entities; the Densify Map and prune/add scores are learned model components. The central design rests on domain assumptions about the validity of gradient-response proxies, transferability of intervention targets, and Euclidean cross-time correspondence. The main hyperparameters are hand-chosen or selected from ablations on the target datasets.

free parameters (9)
  • theta_p = 0.7
    Prune threshold in Eq. 8; selected from Table 4 ablation on Waymo to balance quality and Gaussian count.
  • theta_a = 0.7
    Addition threshold in Eq. 8; selected from Table 4 ablation on Waymo to balance quality and Gaussian count.
  • gamma = 2
    Confidence weighting exponent in Eq. 7; chosen by hand; controls suppression of ambiguous targets near 0.5.
  • K_neighbors = 8
    Number of neighbors used for intervention targets and Cross-Time Point Query; set in implementation without sensitivity analysis.
  • intervention_sample_rate = 0.2 with 3 rounds
    Sampling probability and number of intervention rounds in Section 4.1; chosen by hand to control training cost and coverage.
  • add_perturbation_noise = 0.01 m
    Standard deviation of zero-mean Gaussian noise added to copied positions in the addition branch; chosen by hand.
  • loss_weights = L1:0.8, SSIM:0.2, depth:0.01, velocity:0.01
    Weights in Eq. 2 for rendering, SSIM, depth, and velocity losses; hand-chosen.
  • voxel_size = 0.1 m
    Sparse convolution backbone voxel size; a design choice that affects how many primitives are grouped and queried.
  • refinement_iterations = 3
    Number of iterative Gaussian state refinements in the backbone; a design choice affecting quality and speed.
assumptions (5)
  • domain assumption Local gradient response r_j^B (Eq. 3), the average normalized gradient magnitude over neighboring primitives, is a valid proxy for whether a prune or add intervention benefits reconstruction.
    Section 3.2 states this as the key observation; it is the training signal for the Densify Policy, but no proof or independent measurement is given.
  • domain assumption Intervention targets computed from gradient responses with a frozen backbone are transferable to unseen scenes, so the policy can restructure Gaussians at inference without intervention rendering.
    Assumed in Section 3.2 and 3.4; the policy is trained on targets from the training split and applied directly to test clips.
  • domain assumption The K-nearest neighbors in Euclidean space at other timestamps correspond to the same or related physical structures, so pooled cross-time features are relevant for attribute prediction.
    Section 3.3; the limitations section acknowledges that fast-moving objects can violate this and mix different structures.
  • domain assumption Adding a perturbed copy of an existing primitive is an effective densification operation that can supply missing representation capacity.
    Section 3.2 and Table 4; random additions improve PSNR, but the operation is not derived from any coverage or error criterion.
  • domain assumption LiDAR-initialized primitives plus a fixed monocular-depth background model provide a sufficient starting representation for the learned structure adjustments.
    Section 4.1 and Limitations; the paper relies on LiDAR-aligned monocular depth and sampled sky points for background.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Learning Gaussian Structure: Intervention-Guided Density Control for Feed-Forward Driving Reconstruction." pith.science (2026). https://pith.science/paper/73QQUQKM

@misc{pith2026260811077,
  author       = {Pith},
  title        = {Pith review of: Learning Gaussian Structure: Intervention-Guided Density Control for Feed-Forward Driving Reconstruction},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/73QQUQKM}},
  note         = {Machine review of arXiv:2608.11077}
}
read the original abstract

Feed-forward Gaussian reconstruction has recently emerged as an efficient approach for driving scene reconstruction. However, prevailing LiDAR-based methods preserve the initial correspondence between observed points and Gaussian primitives, treating the initialized primitive set as the final representation. Unlike optimization-based 3DGS, these methods cannot accumulate gradients during training to determine how the scenes representation should be densified. Meanwhile, the shared sparse backbone only fuses observations from different timestamps implicitly, without explicitly aggregating cross-time evidence for individual primitives. In this paper, we present Learning Gaussian Structure (LGS), a framework that enhances both Gaussian structure and primitive attributes. Our key observation is that changes in local gradient responses induced by a prune or add intervention reveal whether the corresponding structural adjustment benefits reconstruction. Based on this observation, our Gaussian Densify Policy learns a Densify Map comprising Prune and Addition Scores from controlled interventions, and directly adjusts the Gaussian structure during inference. We further develop a compact Cross-Time Point Query that explicitly retrieves and aggregates neighboring features from Gaussian primitives at other timestamps for reliable attribute prediction. Extensive experiments on the Waymo Open Dataset and PandaSet demonstrate that LGS consistently outperforms existing methods.

Figures

Figures reproduced from arXiv: 2608.11077 by the authors.

Figure 1
Figure 1. Learning Gaussian Structure for feed-forward driv [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Overview of LGS. Given multi-camera images and LiDAR points captured at different timestamps, the sparse 3D [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Intervention supervision for the Gaussian Den [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Qualitative comparison of novel-view synthesis on Waymo. Compared with STORM (Yang et al. 2025) and UniS [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]
Figure 6
Figure 6. Figure 6: Effect of the Addition Scores predicted by the [PITH_FULL_IMAGE:figures/full_fig_p006_6.png]
Figure 7
Figure 7. Figure 7: Relation between the Addition and Prune Scores. [PITH_FULL_IMAGE:figures/full_fig_p007_7.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

32 extracted references · 13 canonical work pages

  1. [1]

    Advances in Neural Information Processing Systems , volume=

    Flux4d: Flow-based unsupervised 4d reconstruction , author=. Advances in Neural Information Processing Systems , volume=

    work page

  2. [2]

    International Conference on Learning Representations , volume=

    Storm: Spatio-temporal reconstruction model for large-scale outdoor scenes , author=. International Conference on Learning Representations , volume=

    work page

  3. [3]

    arXiv preprint arXiv:2511.04595 , year=

    UniSplat: Unified Spatio-Temporal Fusion via 3D Latent Scaffolds for Dynamic Driving Scene Reconstruction , author=. arXiv preprint arXiv:2511.04595 , year=

    arXiv

  4. [4]

    Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

    Evolsplat: Efficient volume-based gaussian splatting for urban view synthesis , author=. Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

    work page

  5. [5]

    PointForward: Feedforward Driving Reconstruction through Point-Aligned Representations

    PointForward: Feedforward Driving Reconstruction through Point-Aligned Representations , author=. arXiv preprint arXiv:2605.11594 , year=

    work page Pith review arXiv

  6. [6]

    Ground4D: Spatially-Grounded Feedforward 4D Reconstruction for Unstructured Off-Road Scenes

    Ground4D: Spatially-Grounded Feedforward 4D Reconstruction for Unstructured Off-Road Scenes , author=. arXiv preprint arXiv:2605.04435 , year=

    work page Pith review arXiv

  7. [7]

    arXiv preprint arXiv:2603.08254 , year=

    Dynamicvggt: Learning dynamic point maps for 4d scene reconstruction in autonomous driving , author=. arXiv preprint arXiv:2603.08254 , year=

    arXiv

  8. [8]

    European Conference on Computer Vision , pages=

    Lgm: Large multi-view gaussian model for high-resolution 3d content creation , author=. European Conference on Computer Vision , pages=. 2024 , organization=

    2024

Show all 32 references
  1. [9]

    arXiv preprint arXiv:2312.12337 , year=

    pixelsplat: 3d gaussian splats from image pairs for scalable generalizable 3d reconstruction , author=. arXiv preprint arXiv:2312.12337 , year=

    arXiv

  2. [10]

    European conference on computer vision , pages=

    Mvsplat: Efficient 3d gaussian splatting from sparse multi-view images , author=. European conference on computer vision , pages=. 2024 , organization=

    2024

  3. [11]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Splatter image: Ultra-fast single-view 3d reconstruction , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  4. [12]

    Advances in Neural Information Processing Systems , volume=

    L4gm: Large 4d gaussian reconstruction model , author=. Advances in Neural Information Processing Systems , volume=

  5. [13]

    arXiv preprint arXiv:2412.09043 , year=

    DrivingRecon: Large 4D Gaussian reconstruction model for autonomous driving , author=. arXiv preprint arXiv:2412.09043 , year=

    arXiv

  6. [14]

    Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

    Depthsplat: Connecting gaussian splatting and depth , author=. Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

  7. [15]

    Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

    Generative densification: Learning to densify gaussians for high-fidelity generalizable 3d reconstruction , author=. Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

  8. [16]

    arXiv preprint arXiv:2603.21304 , year=

    F4Splat: Feed-Forward Predictive Densification for Feed-Forward 3D Gaussian Splatting , author=. arXiv preprint arXiv:2603.21304 , year=

  9. [17]

    arXiv preprint arXiv:2604.03069 , year=

    SparseSplat: Towards Applicable Feed-Forward 3D Gaussian Splatting with Pixel-Unaligned Prediction , author=. arXiv preprint arXiv:2604.03069 , year=

    arXiv

  10. [18]

    arXiv preprint arXiv:2602.19753 , year=

    RAP: Fast Feedforward Rendering-Free Attribute-Guided Primitive Importance Score Prediction for Efficient 3D Gaussian Splatting Processing , author=. arXiv preprint arXiv:2602.19753 , year=

  11. [19]

    and Tancik, Matthew and Barron, Jonathan T

    Mildenhall, Ben and Srinivasan, Pratul P. and Tancik, Matthew and Barron, Jonathan T. and Ramamoorthi, Ravi and Ng, Ren , booktitle=

  12. [20]

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

    Neurad: Neural rendering for autonomous driving , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  13. [21]

    International Conference on Learning Representations , volume=

    Emernerf: Emergent spatial-temporal scene decomposition via self-supervision , author=. International Conference on Learning Representations , volume=

  14. [22]

    arXiv preprint arXiv:2411.11921 , year=

    Desire-gs: 4d street gaussians for static-dynamic decomposition and surface reconstruction for urban driving scenes , author=. arXiv preprint arXiv:2411.11921 , year=

    arXiv

  15. [23]

    Proceedings of the ieee/cvf conference on computer vision and pattern recognition , pages=

    Suds: Scalable urban dynamic scenes , author=. Proceedings of the ieee/cvf conference on computer vision and pattern recognition , pages=

  16. [24]

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

    Unisim: A neural closed-loop sensor simulator , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  17. [25]

    arXiv preprint arXiv:2306.04988 , year=

    Streetsurf: Extending multi-view implicit surface reconstruction to street views , author=. arXiv preprint arXiv:2306.04988 , year=

    arXiv

  18. [26]

    CAAI International Conference on Artificial Intelligence , pages=

    Mars: An instance-aware, modular and realistic simulator for autonomous driving , author=. CAAI International Conference on Artificial Intelligence , pages=. 2023 , organization=

    2023

  19. [27]

    , author=

    3d gaussian splatting for real-time radiance field rendering. , author=. ACM Trans. Graph. , volume=

  20. [28]

    arXiv preprint arXiv:2312.10035 , year=

    Point transformer v3: Simpler, faster, stronger , author=. arXiv preprint arXiv:2312.10035 , year=

    arXiv

  21. [29]

    Proceedings of Machine Learning and Systems , volume=

    Torchsparse: Efficient point cloud inference engine , author=. Proceedings of Machine Learning and Systems , volume=

  22. [30]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Scalability in perception for autonomous driving: Waymo open dataset , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  23. [31]

    2021 IEEE international intelligent transportation systems conference (ITSC) , pages=

    Pandaset: Advanced sensor suite dataset for autonomous driving , author=. 2021 IEEE international intelligent transportation systems conference (ITSC) , pages=. 2021 , organization=

    2021

  24. [32]

    arXiv preprint arXiv:2511.10647 , year=

    Depth anything 3: Recovering the visual space from any views , author=. arXiv preprint arXiv:2511.10647 , year=

    arXiv

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.