Pith. sign in

REVIEW 3 major objections 5 minor 25 references

LocusGS: Spatially Grounded Tokens for Feed-Forward 3D Gaussian Splatting

T0 review · 3 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read LocusGS claims that equipping each query token with a learnable 3D anchor state—a center and a support radius—improves rendering quality and spatial coherence in feed-forward 3D Gaussian splatting under matched budgets.

desk verdict Genuine anchor-token advance for feed-forward 3DGS with consistent gains, but the spatial-grounding claim is not isolated from added capacity and supervision. read the letter →

arxiv 2608.12825 v1 pith:JBGPWFWP submitted 2026-08-13 cs.CV

Wenyu Li , Sidun Liu , Tongrui Hu , Peng Qiao , Yong Dou This is my paper · ORCID
classification cs.CV
keywords 3DGaussianSplattingfeed-forwardreconstructionquerytokensspatialanchorsnovelviewsynthesiscross-attentiongeometricbiasanchor-centereddecoding
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 3D Gaussian splatting systems that use learnable queries to decode groups of Gaussians suffer from a spatial decoupling: the Gaussians produced by one query scatter across the scene instead of covering one local region. LocusGS claims this happens because queries are purely latent, and that giving each query an explicit 3D anchor—a center plus a support radius—fixes it. The anchor is refined layer by layer, steers which image features the query attends to, and localizes the decoded Gaussians around the anchor. The paper reports that this single change improves novel-view rendering quality over the query-based baseline under identical token and Gaussian budgets, and that it sharply reduces token-level Gaussian dispersion. A sympathetic reader would take the paper as establishing that spatial grounding, not just representation capacity, is what organizes query-based scene tokens.

What carries the argument

The load-bearing object is the anchor state $a_i^l=(\mu_i^l,r_i^l)$ attached to each Gaussian token: a 3D center and a scalar support radius, both learned and progressively refined across decoder layers by residual updates. It is used in three places: anchor-aware self-attention injects a positional embedding from the center; anchor-to-ray cross-attention adds a geometric bias $b_{ij}^l=-\tfrac12\left(D(\mu_i^l,\ell_j)/(\sigma_0 r_i^l)\right)^2$ to the content logits, where $D$ is the point-to-ray distance, so queries prefer image tokens whose rays pass near the anchor; and anchor-centered decoding predicts Gaussian centers as $\mu_{i,k}^G=\mu_i^L+r_i^L\delta_{i,k}$, local offsets scaled by the radius. Together these convert a latent query into a spatial hypothesis that selects evidence and organizes primitives.

What would settle it

Train the baseline with the same total parameters (for example, enlarge its prediction heads to reach the same 241.5M count), the same multi-layer rendering supervision, and the same visibility losses on its predicted centers, while keeping its latent queries and free Gaussian decoding; if its PSNR and dispersion metrics match LocusGS, the anchor state is not the active ingredient.

Watch

Extended reading notes

Core claim

The central claim is that an explicit, learnable 3D anchor state—rather than any new loss or larger model—is what makes query-based feed-forward 3D Gaussian prediction spatially coherent. On RealEstate10K and DL3DV, the method outperforms the strongest query-based baseline with the same number of tokens and the same total Gaussian budget, for example 28.89 versus 28.41 PSNR on RealEstate10K at 4096 tokens and 24.80 versus 23.44 at four views on DL3DV. The learned anchors form a coarse spatial scaffold over the scene, with smaller radii near detailed regions and larger radii in weakly constrained areas. The paper also reports that anchor-aware cross-attention produces localized, view-consistent evidence aggregation, and that the decoded Gaussians from one token are far more compact, with the mean token-level dispersion score dropping from about 5.1 to 0.20 on RealEstate10K.

Load-bearing premise

The comparison assumes the extra anchor machinery is the cause of the gains, since the baseline has fewer parameters and lacks the intermediate rendering and anchor losses; if a matched-capacity baseline with the same supervision closes the gap, the spatial-grounding claim collapses.

Editorial extensions

If this is right

  • At equal token and Gaussian budgets, rendering quality improves on both benchmarks and across 2-, 4-, and 6-view test settings, so the spatial-grounding design transfers to unseen context lengths.
  • Token-level Gaussian dispersion falls by roughly an order of magnitude, meaning each query's Gaussians describe one local surface patch rather than many floating pieces.
  • Cross-attention becomes localized around the anchor, so the query gathers evidence from geometrically consistent image regions, which helps in texture-rich and cluttered scenes.
  • Middle-layer rendering supervision is helpful and dense supervision is not; anchoring plus supervision at the middle and final layers suffices.

Reading between the lines

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

  • The anchor-to-ray bias should tolerate pose noise reasonably well: replacing exact rays with a distribution over rays would turn the geometric bias into an expected-distance term, a natural extension the paper does not test.
  • The center-plus-scalar-radius state is isotropic; an anisotropic or covariance-based support would likely capture elongated structures such as wires and poles better than the current scalar radius.
  • If spatial grounding is the active mechanism, the same anchor state could transfer to other token-based 3D representations, such as tokens decoding neural radiance fields or signed distance functions, not only Gaussian splats.
  • A parameter-matched baseline with the same auxiliary losses but no anchor-conditioned attention would settle whether the improvement comes from the anchor or from added capacity; the paper does not run this control.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

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 LocusGS, a feed-forward 3D Gaussian Splatting method that augments each learnable query token with an explicit 3D anchor state consisting of a center and a support radius. The anchor is progressively refined across decoder layers and is used in self-attention, cross-attention through an anchor-to-ray geometric bias, and Gaussian center decoding via radius-scaled offsets. Experiments on RE10K and DL3DV compare LocusGS against TokenGS under matched token and Gaussian budgets and report consistent improvements in PSNR, SSIM, and LPIPS, together with ablations and qualitative analysis of anchor layouts, cross-attention patterns, and token-level Gaussian compactness.

Significance. If the central claim holds, this is a useful and timely contribution: it identifies a concrete limitation of query-based feed-forward 3DGS, proposes a simple and plausible remedy, and evaluates it on large-scale benchmarks with fixed token and Gaussian budgets. Strengths include the use of external novel-view-synthesis benchmarks, component-level ablations that are mostly consistent, and a clear comparison against a well-matched query-based baseline in terms of token and Gaussian counts. The significance is moderated by an attribution confound: LocusGS differs from TokenGS not only by the anchor states but also by a non-negligible parameter increase and an additional multi-layer supervision schedule, so the reported gains are not yet cleanly attributable to spatial grounding. The token-level dispersion metric is also partly enforced by the proposed center parameterization, which weakens the claim of independent evidence for spatial organization. With matched controls, the contribution could be solid.

major comments (3)
  1. [§4.1–§4.2, Tables 1–2, Table 8, Eq. (10)–(11)] The paper claims in Section 4.1 that comparing under matched token and Gaussian budgets 'allows us to isolate the effect of explicit spatial grounding,' but the comparison against TokenGS does not control for model capacity or training objective. Table 8 reports 241.5M parameters for LocusGS versus 222.0M for TokenGS, and §3.2.7/§A.6 add two training signals that TokenGS does not have: intermediate-layer rendering supervision (Eq. 10–11) and an additional visibility-regularization term on anchor centers. Table 5 shows that, within LocusGS, switching from final-layer-only supervision to {6,12} supervision changes PSNR by about 0.735 dB (23.549 to 24.284), which is a substantial fraction of the 1.36 dB gap over TokenGS in the 4-view DL3DV row of Table 1. A matched control that augments TokenGS with the same extra parameters and the same supervision schedule, but without anchor states, is needed before the abstract's claim that explicit anchor states improve rendering quality can be accepted. This is load-bearing because the central contribution is causal attribution of the gains to spatial grounding.
  2. [§4.3.3, Eq. (9), Eq. (12), Table 3] The token-level dispersion metric is partly mechanical. Eq. 9 sets the Gaussian center as µG_i,k = µL_i + rL_i δ_i,k, so the within-token spread of centers is explicitly scaled by the learned radius rL_i. A lower Ccentroid value for LocusGS is therefore expected from the decoding parameterization alone and is not independent evidence of a more structured learned representation. The comparison against TokenGS, which uses unconstrained centers, conflates the effect of the center parameterization with the effect of spatial grounding through attention and refinement. To support the claim, dispersion should be compared against a TokenGS variant that uses the same anchor-centered decoding but without anchor-aware attention or anchor refinement, or the dispersion should be normalized by the anchor radius. As written, Table 3's large reductions (e.g., 5.1164 to 0.1978 on RE10K) are consistent with a mechanical explanation.
  3. [§4.2, Tables 1–2] The main quantitative claims are made without any measure of uncertainty. Tables 1 and 2 report point estimates only, with no standard deviations, number of seeds, or statistical tests. Given that some of the reported differences are modest in absolute terms, such as 28.41 vs. 28.89 PSNR on RE10K with 4096 tokens, the reader cannot assess whether these differences are robust. I request means and standard deviations over at least three training runs, or a paired evaluation across the same held-out scenes, for the primary TokenGS comparisons.
minor comments (5)
  1. [§4.1] The phrase 'identical token and Gaussian budgets' should be qualified by noting that the models differ in parameter count and supervision schedule; otherwise the wording overstates the degree of control in the comparison.
  2. [§4.3.2 and Figure 5] The qualitative claims about anchors concentrating around informative regions and radii adapting to local detail are supported only by selected visualizations. Reporting an aggregate statistic, such as the correlation between anchor radius and local image gradient or depth discontinuity, would strengthen this analysis.
  3. [§A.5, Eq. (24)] The point-to-ray distance formula in Eq. (24) is written for a normalized direction, but the main text introduces D(µ_i^l, ℓ_j) in Eq. (3) without stating that normalization is applied first. Please add a forward reference to §A.5 at the first occurrence.
  4. [§B.3, Table 8] The forward-time comparison shows LocusGS is about 19% slower than TokenGS (407.3 ms vs. 341.0 ms per sample), but the main paper does not discuss this compute overhead. Since the paper emphasizes a controlled comparison, the latency and memory cost should be mentioned in the main text along with the rendering-quality gains.
  5. [References] Reference [8] cites a CVPR 2026 paper for TokenGS; please verify the publication venue and page numbers at the time of final submission, especially given the arXiv version date.

Circularity Check

1 steps flagged · score 3.0 of 10

Spatial-organization evidence is partly built into the anchor-centered decoding, while the rendering comparisons remain external and non-circular.

  1. self definitional [Section 3.2.6, Eq. (9); Section 4.3.3, Eq. (12); Table 3]
    "The final Gaussian center is then obtained by anchoring this local offset around the refined anchor center: µG_i,k = µL_i + rL_i δ_i,k (9). We further quantify this property using token-level Gaussian dispersion: Ccentroid = 1/(NK) Σ_i,k ||µG_i,k − ¯µG_i||^2 (12)."

    Equation (9) defines every Gaussian center in a token as that token's anchor center plus a radius-scaled offset, so the within-token variance computed in Eq. (12) is algebraically the variance of those offsets scaled by r_i^2. The Table 3 dispersion comparison therefore largely measures whether LocusGS obeys its own anchor-centered decoding parameterization, rather than independently confirming that explicit anchors create spatial structure. The conclusion that the anchor-based formulation yields more coherent local Gaussian groups is thus partly a restatement of the construction, especially since TokenGS predicts free centers while LocusGS does not.

full rationale

The core rendering claims are compared on external benchmarks (RE10K and DL3DV) against TokenGS and other published feed-forward 3DGS methods, so the PSNR/SSIM/LPIPS improvements are not circular. I found no load-bearing self-citation: the cited baselines are external prior work, and DAB-DETR and Scaffold-GS are used only as conceptual analogies. The one partial circularity is the spatial-organization evidence in Section 4.3.3: Eq. (9) constructs Gaussian centers from anchors and Eq. (12) measures exactly the resulting clustering, so the large dispersion reduction in Table 3 is substantially built into the model design rather than being an independent confirmation of the anchor hypothesis. Separately, the paper's claim in Section 4.1 that matching token and Gaussian budgets isolates the effect of spatial grounding is weakened by the additional 19.5M parameters shown in Table 8 and the added multi-layer supervision in Eqs. (10)-(11), but that is an attribution or experimental-control concern, not a circular derivation. Overall, the central rendering result stands on external evaluation, so the circularity score is modest.

Assumptions & free parameters 6 free parameters · 6 assumptions · 0 invented entities

The central claim rests mainly on model architecture and training choices: learnable anchor states, a hand-crafted geometric bias, and anchor-centered decoding. No new physical entities are introduced; the anchor is a learned latent state. The free parameters listed above are the hand-chosen or learned numbers on which the empirical gains depend.

free parameters (6)
  • Initial anchor centers mu^0_i = learned, not reported
    Learnable and shared across scenes, randomly initialized in normalized 3D space (Section 3.2.3). Anchor positions are the core spatial state and affect all downstream results.
  • Raw support radii rho^0_i and per-token radii = learned, not reported
    Initialized via inverse softplus to a predefined scale and refined per layer (Eqs 1, 6, 7). Controls attention bandwidth and the scale of Gaussian offsets.
  • Learnable bias scale gamma = learned, not reported
    gamma = softplus(gamma_tilde) >= 0 scales the geometric bias in cross-attention (Eq 4).
  • Base bandwidth sigma_0 = 0.1
    Fixed by hand (Appendix A.5). Controls how sharp the geometric bias is; no sensitivity study is reported.
  • Loss weights = lambda_SSIM=0.2, lambda_G=1.0, lambda_A=0.1
    Chosen without reported sensitivity analysis (Eq 10 and Appendix A.6).
  • Token budget and Gaussians per token = 1024/4096 tokens, 64 Gaussians/token
    Hand-chosen to match TokenGS; the claim of same Gaussian budget depends on these numbers.
assumptions (6)
  • domain assumption Input camera poses are calibrated and available
    Plucker rays and anchor-to-ray distances are computed from known intrinsics and extrinsics (Sections 3.2.2 and A.4); the paper lists pose-free inputs as a limitation (B.4).
  • domain assumption Patch-level averaged Plucker rays approximate per-pixel geometry well enough for the geometric bias
    The implementation pools Plucker tensors with the image patch size (Eq 21); the bias uses pooled rays rather than per-pixel rays.
  • domain assumption The fixed set of learnable anchors can cover arbitrary scene content
    Anchors are randomly initialized in normalized scene space and refined; no coverage guarantee is given (Section 3.2.3).
  • ad hoc to paper The Gaussian geometric bias form is appropriate
    The negative Gaussian penalty in Eq 3 is chosen by hand; no derivation or comparison to other distance kernels is provided.
  • ad hoc to paper Anchor-centered center decoding is a valid inductive bias
    Eq 9 (center equals anchor plus radius-scaled offset) is the core design; its benefit is shown empirically through ablations, not derived.
  • standard math Differentiable Gaussian splatting and transformer attention are standard tools
    The method relies on standard rendering and attention primitives from prior work, assumed correct and available.

how reviews work

0 comments
Cite this review

Pith. "Pith review of LocusGS: Spatially Grounded Tokens for Feed-Forward 3D Gaussian Splatting." pith.science (2026). https://pith.science/paper/JBGPWFWP

@misc{pith2026260812825,
  author       = {Pith},
  title        = {Pith review of: LocusGS: Spatially Grounded Tokens for Feed-Forward 3D Gaussian Splatting},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/JBGPWFWP}},
  note         = {Machine review of arXiv:2608.12825}
}
read the original abstract

Recent query-based feed-forward 3DGS methods represent a scene using learnable queries, each aggregating multi-view evidence and decoding a group of Gaussians. Ideally, different queries should specialize in coherent local regions of the scene. However, we observe that Gaussians decoded from the same query often scatter across distant scene regions, resulting in weak query-level spatial coherence and poor alignment with the scene structure. We attribute this behavior to the purely latent representation of existing Gaussian queries. To address this limitation, we introduce LocusGS, which augments each Gaussian query with a 3D anchor state consisting of a center and a support radius. The anchor state is progressively refined across decoder layers and is used throughout query interaction, multi-view feature aggregation, and Gaussian generation. Specifically, an anchor-to-ray geometric bias guides each query toward spatially relevant image observations, while anchor-centered decoding organizes its Gaussians within a local region. Experiments on novel view synthesis benchmarks show that LocusGS improves rendering quality over query-based Gaussian token baselines under the same Gaussian budget. Further analysis shows that the learned anchors form coherent spatial layouts and lead to more structured Gaussian distributions, demonstrating that explicit anchor states improve the spatial organization. Our project page: https://leo-frank.github.io/LocusGS_viewer.

Figures

Figures reproduced from arXiv: 2608.12825 by the authors.

Figure 2
Figure 2. However, removing the dense image-to-Gaussian correspondence also leaves the spatial role of each query [PITH_FULL_IMAGE:figures/full_fig_p001_2.png] view at source ↗
Figure 1
Figure 1. Spatial grounding improves both global Gaussian organization and token-level locality. The orange points highlight all 64 Gaussians decoded from a single representative token. Under the same input and Gaussian budget, TokenGS produces diffuse scene structures, with the Gaussians from one token scattered across a broad spatial region. LocusGS instead reconstructs Gaussians that better follow the scene geometry, and i… view at source ↗
Figure 2
Figure 2. Conceptual comparison of three feed-forward 3DGS paradigms. (a) Dense view-wise prediction decodes multi-view image features and uses a per-view head to produce view-dependent Gaussian splats. (b) Fixed-budget query-based generation introduces learnable queries to decode a set of Gaussians whose size is independent of image resolution and input view count. (c) Our spatially grounded query paradigm further equips que… view at source ↗
Figures from the paper (11 more)
Figure 3
Figure 3. Figure 3: From implicit Gaussian tokens to explicit anchor tokens. Left: standard query-based methods uses learnable 3DGS tokens as implicit embeddings, where self-attention and image-to-3DGS cross-attention are mainly driven by token and image features. Right: LocusGS augments …
Figure 4
Figure 4. Figure 4: Qualitative comparison on DL3DV novel-view synthesis. LocusGS produces sharper renderings and more coherent depth structures, especially in texture-rich regions and cluttered scenes. where kj is the projected key of the j-th image token and d is the feature dimension p…
Figure 5
Figure 5. Figure 5: Visualization of anchor supports and token-level Gaussian compactness. The first three rows show LocusGS: all projected anchors, selected projected radii, and Gaussian groups decoded from the selected anchor tokens. The last row shows Gaussian groups decoded from Token…
Figure 6
Figure 6. Figure 6: Qualitative comparisons of the reconstructed Gaussians produced by TokenGS and LocusGS. same dataset. As shown in [PITH_FULL_IMAGE:figures/full_fig_p010_6.png]
Figure 7
Figure 7. Figure 7: Spatial anchors yield more localized and view-consistent cross-view attention. For each method, we show the cross-attention map of a representative token. Our method produces more localized and view-consistent attention patterns around the projected anchor region. also…
Figure 8
Figure 8. Figure 8: Decomposition of anchor-guided cross-attention. We visualize the content-only attention, geometry-only attention, final attention, and projected Gaussians from the same token. The final attention combines visual similarity with geometric compatibility, leading to more …
Figure 9
Figure 9. Figure 9: Convergence comparison on DL3DV and RE10K. Our method exhibits faster convergence with respect to training epochs, reaches lower training loss, and achieves consistently higher PSNR on both training and validation sets. self-attention, indicating that explicit anchor p…
Figure 10
Figure 10. Figure 10: Visualization of anchor-aware self-attention. For each selected query anchor, we visualize its decoded Gaussian group and the top-k attended anchors. Edge thickness and opacity indicate attention strength [PITH_FULL_IMAGE:figures/full_fig_p013_10.png]
Figure 11
Figure 11. Figure 11: Qualitative comparison on DL3DV novel-view synthesis. LocusGS produces sharper renderings and more coherent depth structures, especially in texture-rich regions and cluttered scenes. References [1] Johannes Lutz Schönberger and Jan-Michael Frahm. Structure-from-motion…
Figure 12
Figure 12. Figure 12: Per-scene dispersion score comparison between TokenGS and our method on DL3DV. Each overlaid pair of bars corresponds to the same scene. Our method consistently yields substantially lower dispersion scores than TokenGS, indicating that the Gaussian groups associated w…
Figure 13
Figure 13. Figure 13: Per-scene dispersion score comparison between TokenGS and our method on RE10K. Each overlaid pair of bars corresponds to the same scene. Our method consistently yields substantially lower dispersion scores than TokenGS, indicating that the Gaussian groups associated w…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

25 extracted references · 11 canonical work pages

  1. [1]

    Structure-from-motion revisited

    Johannes Lutz Schönberger and Jan-Michael Frahm. Structure-from-motion revisited. InProc. CVPR, 2016

    work page 2016

  2. [2]

    Pixelwise view selection for unstructured multi-view stereo

    Johannes Lutz Schönberger, Enliang Zheng, Marc Pollefeys, and Jan-Michael Frahm. Pixelwise view selection for unstructured multi-view stereo. InProc. ECCV, 2016

    work page 2016

  3. [3]

    Nerf: Representing scenes as neural radiance fields for view synthesis.Communications of the ACM, 65(1):99–106, 2021

    Ben Mildenhall, Pratul P Srinivasan, Matthew Tancik, Jonathan T Barron, Ravi Ramamoorthi, and Ren Ng. Nerf: Representing scenes as neural radiance fields for view synthesis.Communications of the ACM, 65(1):99–106, 2021

    2021

  4. [4]

    3d gaussian splatting for real-time radiance field rendering.ACM Transactions on Graphics, 42(4), July 2023

    Bernhard Kerbl, Georgios Kopanas, Thomas Leimkühler, and George Drettakis. 3d gaussian splatting for real-time radiance field rendering.ACM Transactions on Graphics, 42(4), July 2023

    2023

  5. [5]

    pixelSplat: 3D Gaussian splats from image pairs for scalable generalizable 3D reconstruction

    David Charatan, Sizhe Li, Andrea Tagliasacchi, and Vincent Sitzmann. pixelSplat: 3D Gaussian splats from image pairs for scalable generalizable 3D reconstruction. InProc. CVPR, 2024

    work page 2024

  6. [6]

    MVSplat: efficient 3d gaussian splatting from sparse multi-view images.arXiv, 2403.14627, 2024

    Yuedong Chen, Haofei Xu, Chuanxia Zheng, Bohan Zhuang, Marc Pollefeys, Andreas Geiger, Tat-Jen Cham, and Jianfei Cai. MVSplat: efficient 3d gaussian splatting from sparse multi-view images.arXiv, 2403.14627, 2024

    arXiv 2024

  7. [7]

    Depthsplat: Connecting gaussian splatting and depth

    Haofei Xu, Songyou Peng, Fangjinhua Wang, Hermann Blum, Daniel Barath, Andreas Geiger, and Marc Pollefeys. Depthsplat: Connecting gaussian splatting and depth. InCVPR, 2025. 18 APREPRINT Figure 12:Per-scene dispersion score comparison between TokenGS and our method on DL3DV .Each overlaid pair of bars corresponds to the same scene. Our method consistently...

    work page 2025

  8. [8]

    Tokengs: Decoupling 3d gaussian prediction from pixels with learnable tokens.Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2026

    Jiawei Ren, Michal Tyszkiewicz, Jiahui Huang, and Zan Gojcic. Tokengs: Decoupling 3d gaussian prediction from pixels with learnable tokens.Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2026

    work page 2026

Show all 25 references
  1. [9]

    Globalsplat: Efficient feed-forward 3d gaussian splatting via global scene tokens.arXiv preprint arXiv:2604.15284, 2026

    Roni Itkin, Noam Issachar, Yehonatan Keypur, Xingyu Chen, Anpei Chen, and Sagie Benaim. Globalsplat: Efficient feed-forward 3d gaussian splatting via global scene tokens.arXiv preprint arXiv:2604.15284, 2026

    2026 arXiv

  2. [10]

    Selfsplat: Pose-free and 3d prior-free generalizable 3d gaussian splatting

    Gyeongjin Kang, Jisang Yoo, Jihyeon Park, Seungtae Nam, Hyeonsoo Im, Sangheon Shin, Sangpil Kim, and Eunbyung Park. Selfsplat: Pose-free and 3d prior-free generalizable 3d gaussian splatting. InProceedings of the Computer Vision and Pattern Recognition Conference, pages 22012–...

    2025

  3. [11]

    No pose, no problem: Surprisingly simple 3d gaussian splats from sparse unposed images.arXiv preprint arXiv:2410.24207, 2024

    Botao Ye, Sifei Liu, Haofei Xu, Li Xueting, Marc Pollefeys, Ming-Hsuan Yang, and Peng Songyou. No pose, no problem: Surprisingly simple 3d gaussian splats from sparse unposed images.arXiv preprint arXiv:2410.24207, 2024

    2024 arXiv

  4. [12]

    Freesplat: Generalizable 3d gaussian splatting towards free-view synthesis of indoor scenes.arXiv preprint arXiv:2405.17958, 2024

    Yunsong Wang, Tianxin Huang, Hanlin Chen, and Gim Hee Lee. Freesplat: Generalizable 3d gaussian splatting towards free-view synthesis of indoor scenes.arXiv preprint arXiv:2405.17958, 2024

    2024 arXiv

  5. [13]

    Longsplat: Robust unposed 3d gaussian splatting for casual long videos

    Chin-Yang Lin, Cheng Sun, Fu-En Yang, Min-Hung Chen, Yen-Yu Lin, and Yu-Lun Liu. Longsplat: Robust unposed 3d gaussian splatting for casual long videos. InICCV, 2025

    2025

  6. [14]

    Zpressor: Bottleneck- aware compression for scalable feed-forward 3dgs.arXiv preprint arXiv:2505.23734, 2025

    Weijie Wang, Donny Y Chen, Zeyu Zhang, Duochao Shi, Akide Liu, and Bohan Zhuang. Zpressor: Bottleneck- aware compression for scalable feed-forward 3dgs.arXiv preprint arXiv:2505.23734, 2025

    2025

  7. [15]

    Long-lrm: Long-sequence large reconstruction model for wide-coverage gaussian splats

    Chen Ziwen, Hao Tan, Kai Zhang, Sai Bi, Fujun Luan, Yicong Hong, Li Fuxin, and Zexiang Xu. Long-lrm: Long-sequence large reconstruction model for wide-coverage gaussian splats. InProceedings of the IEEE/CVF International Conference on Computer Vision, 2025

    2025

  8. [16]

    Tokensplat: Token-aligned 3d gaussian splatting for feed-forward pose-free reconstruction

    Yihui Li, Chengxin Lv, Zichen Tang, Hongyu Yang, and Di Huang. Tokensplat: Token-aligned 3d gaussian splatting for feed-forward pose-free reconstruction. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 40886–40895, June 2026. 1...

    2026

  9. [17]

    Anysplat: Feed-forward 3d gaussian splatting from unconstrained views.ACM Transactions on Graphics (TOG), 44(6):1–16, 2025

    Lihan Jiang, Yucheng Mao, Linning Xu, Tao Lu, Kerui Ren, Yichen Jin, Xudong Xu, Mulin Yu, Jiangmiao Pang, Feng Zhao, et al. Anysplat: Feed-forward 3d gaussian splatting from unconstrained views.ACM Transactions on Graphics (TOG), 44(6):1–16, 2025

    2025

  10. [18]

    Anchorsplat: Feed-forward 3d gaussian splatting with 3d geometric priors.arXiv preprint arXiv:2604.07053, 2026

    Xiaoxue Zhang, Xiaoxu Zheng, Yixuan Yin, Tiao Zhao, Kaihua Tang, Michael Bi Mi, Zhan Xu, and Dave Zhenyu Chen. Anchorsplat: Feed-forward 3d gaussian splatting with 3d geometric priors.arXiv preprint arXiv:2604.07053, 2026

    2026 arXiv

  11. [19]

    C3g: Learning compact 3d representations with 2k gaussians

    Honggyu An, Jaewoo Jung, Mungyeom Kim, Sunghwan Hong, Chaehyun Kim, Kazumi Fukuda, Minkyeong Jeon, Jisang Han, Takuya Narihira, Hyuna Ko, et al. C3g: Learning compact 3d representations with 2k gaussians. arXiv preprint arXiv:2512.04021, 2025

    2025 arXiv

  12. [20]

    DAB-DETR: Dynamic anchor boxes are better queries for DETR

    Shilong Liu, Feng Li, Hao Zhang, Xiao Yang, Xianbiao Qi, Hang Su, Jun Zhu, and Lei Zhang. DAB-DETR: Dynamic anchor boxes are better queries for DETR. InInternational Conference on Learning Representations, 2022

    2022

  13. [21]

    Scaffold-gs: Structured 3d gaussians for view-adaptive rendering

    Tao Lu, Mulin Yu, Linning Xu, Yuanbo Xiangli, Limin Wang, Dahua Lin, and Bo Dai. Scaffold-gs: Structured 3d gaussians for view-adaptive rendering. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 20654–20664, 2024

    2024

  14. [22]

    Stereo magnification: Learning view synthesis using multiplane images.arXiv preprint arXiv:1805.09817, 2018

    Tinghui Zhou, Richard Tucker, John Flynn, Graham Fyffe, and Noah Snavely. Stereo magnification: Learning view synthesis using multiplane images.arXiv preprint arXiv:1805.09817, 2018

    2018 arXiv

  15. [23]

    Dl3dv-10k: A large-scale scene dataset for deep learning-based 3d vision

    Lu Ling, Yichen Sheng, Zhi Tu, Wentian Zhao, Cheng Xin, Kun Wan, Lantao Yu, Qianyu Guo, Zixun Yu, Yawen Lu, et al. Dl3dv-10k: A large-scale scene dataset for deep learning-based 3d vision. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pag...

    2024

  16. [24]

    Gs-lrm: Large reconstruction model for 3d gaussian splatting.European Conference on Computer Vision, 2024

    Kai Zhang, Sai Bi, Hao Tan, Yuanbo Xiangli, Nanxuan Zhao, Kalyan Sunkavalli, and Zexiang Xu. Gs-lrm: Large reconstruction model for 3d gaussian splatting.European Conference on Computer Vision, 2024

    2024

  17. [25]

    Decoupled weight decay regularization.arXiv preprint arXiv:1711.05101, 2017

    Ilya Loshchilov and Frank Hutter. Decoupled weight decay regularization.arXiv preprint arXiv:1711.05101, 2017. 20

    2017 arXiv

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.