Pith. sign in

REVIEW 3 major objections 4 minor 71 references

UniqueSplat: View-conditioned 3D Gaussian Splatting for Generalizable 3D Reconstruction

T0 review · 3 major / 4 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 The paper claims that conditioning the Gaussian predictor on the target view via a two-branch hypernetwork gives sharper sparse-view renderings and stronger cross-dataset generalization than fixed-Gaussian feed-forward models.

desk verdict Useful incremental extension of MVSplat with a genuinely new two-branch view-conditioned hypernetwork; the capacity confound leaves the central attribution unproven. read the letter →

arxiv 2608.02145 v1 pith:EEZFTOFP submitted 2026-08-03 cs.CV cs.AI

Haixu Song , Xiaoke Yang , Shengjun Zhang , Jiwen Lu , Yueqi Duan This is my paper · ORCID
classification cs.CVcs.AI
keywords 3DGaussianSplattingview-conditionedhypernetworkgeneralizablenovelviewsynthesissparse-viewreconstructionfeed-forwardcross-datasetgeneralizationcostvolumedynamicprediction
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 methods currently predict one fixed set of Gaussians per scene, tuned to render every viewpoint about equally well. UniqueSplat claims that this leaves quality on the table, because the target viewpoint is known at inference time but never used when the Gaussians are predicted. The paper proposes conditioning the Gaussian predictor on the query view through a two-branch hypernetwork: one branch supplies stable, shareable weights, and the other injects target-view-specific adjustments. If it is right, a single forward pass renders sharper novel views than fixed-Gaussian baselines and transfers better to unseen datasets. The reported gains include PSNR increases from 26.39 to 27.28 on RealEstate10K and from 13.94 to 16.01 on the cross-dataset DTU test.

What carries the argument

The central object is the view-conditioned hypernetwork, a weight generator for the Gaussian predictor. A view-agnostic branch produces a base weight $W=H_a(e_a)$ from a learned embedding $e_a$ that abstracts knowledge shared across views, while a view-specific branch produces an adjustment $\Delta W=H_s(e_s)$ from an embedding $e_s$ obtained by projecting source features onto the target view along epipolar lines. The fused weight $W_H = W + \Delta W$ is reshaped and injected into the predictor's convolutional layers, so the predicted Gaussian parameters $\mu$, $\Sigma$, $\alpha$ and $c$ depend on the query view and are rendered by rasterization-based splatting.

What would settle it

Train two variants of the same primary network: one with fixed weights and the same added parameter count, and one that concatenates the target camera pose into the encoder features instead of generating weights; if either matches or exceeds the reported PSNR on RealEstate10K, the claim that view-customized weights cause the improvement is falsified. Alternatively, check whether the Gaussians predicted for two different target views of the same scene differ by more than floating-point noise; if they are effectively identical, the view-specific branch is inert.

Watch

Extended reading notes

Core claim

On its own terms, the paper's discovery is that the network which turns multi-view features into 3D Gaussian parameters does not have to be a fixed function shared across all target views. By generating the predictor's convolutional weights from a learned view-agnostic embedding plus a view-specific embedding derived from the target camera and the source features, the model produces a different set of Gaussians for each query view while keeping training single-stage and feed-forward. The authors argue that this dynamic-weight mechanism removes the cracks, blurs and artifacts visible in fixed-Gaussian baselines, and they support it with consistent PSNR, SSIM and LPIPS gains in both intra-dataset and cross-dataset settings. They also report that the added machinery is lighter than the transformer-based alternative: 38.1M parameters, 0.065s and 3.12GB versus 110.5M parameters, 0.083s and 4.98GB.

Load-bearing premise

The paper attributes the performance gains to the view-conditioning mechanism, but the ablations compare only the presence of the two branches and do not test a static predictor with matched capacity or a simpler injection of the target pose, so the gains could in principle come from the extra parameters rather than from conditioning.

Editorial extensions

If this is right

  • If the claim holds, feed-forward sparse-view reconstruction no longer has to compromise for all viewpoints at once: each target query can receive its own Gaussian field in a single forward pass.
  • The same two-branch weight-generation scheme could be attached to other cost-volume or epipolar predictors, not only the Gaussian head demonstrated here.
  • Cross-dataset transfer improves: training once on one distribution yields stronger zero-shot rendering on other datasets than fixed-Gaussian baselines, suggesting view conditioning acts as a partial domain-invariant prior.
  • The overhead stays practical: the reported parameter count, latency and memory sit between the lightweight baseline and the transformer-based alternative, so the specialization does not require per-scene optimization.

Reading between the lines

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

  • I infer that the view-agnostic branch probably acts as a learned initialization that anchors the fused weights near a stable average, while the view-specific branch supplies a correction; measuring the norm of $\Delta W$ across target views would show how much of the weight space is actually customized.
  • I infer the same conditioning principle could extend beyond Gaussian splatting to depth or cost-volume predictors, where the target camera would reshape the cost volume or the refinement head.
  • I infer the cross-dataset gain on object-centric scenes may come partly from the epipolar feature projection rather than from the hypernetwork itself, since that projection injects explicit geometric consistency; an ablation separating projection from weight modulation would settle it.
  • A testable extension is to condition on per-pixel ray directions rather than a single global target embedding, which would allow spatially varying customization of the Gaussian field.
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 / 4 minor

Summary. The paper proposes UniqueSplat, a feed-forward 3D Gaussian Splatting method that predicts view-dependent Gaussians by injecting target-view information into the predictor's weights. The model consists of a primary network (based on MVSplat) and a two-branch hypernetwork: a view-agnostic branch that produces a fixed weight tensor from a learned constant embedding, and a view-specific branch that extracts features from input views, projects them to the target view via epipolar geometry, and generates a weight update. The fused weights are applied to the convolutional layers of the predictor, which then estimates Gaussian parameters for rendering the target view. Experiments on RealEstate10K, ACID, and DTU report state-of-the-art results in both intra-dataset and cross-dataset settings, along with an efficiency comparison and ablations of the two branches.

Significance. If the proposed mechanism is responsible for the reported gains, the contribution is significant: UniqueSplat is among the first feed-forward 3DGS models to condition the Gaussian predictor on the query view, and the cross-dataset gains on DTU are large. The paper also reports a favorable efficiency trade-off relative to TranSplat (38.1M parameters vs. 110.5M) and includes failure cases and qualitative comparisons. However, the central attribution claim is not yet established. The ablations do not control for the large increase in parameter count or for the MVSplat initialization, and the view-agnostic branch is a fixed weight tensor rather than a dynamic sharing mechanism. With the missing capacity-matched and simple-conditioning baselines, the reported improvements could plausibly come from added model capacity rather than from view-conditioned Gaussian customization.

major comments (3)
  1. [Section IV-E, Table IV] The ablation design does not isolate the proposed view-conditioning mechanism. The full model (38.1M) is compared only against MVSplat (12.0M) and MVSplat plus the view-agnostic branch; the view-specific branch adds a ResNet18 encoder, a projection module, and a hypernetwork, so the 0.89 dB gain over MVSplat on RealEstate10K could be caused by additional capacity rather than by per-view conditioning. Please add (i) a capacity-matched static predictor with the same parameter count and the same MVSplat initialization, (ii) a variant that removes target-view information from the view-specific branch while keeping the architecture fixed (for example, replacing the target projection matrices with a fixed reference projection), and (iii) a simple conditioning baseline such as concatenating the target pose to the cost volume or using FiLM conditioning. Without these controls, the central claim that view-conditioning rather than capacity drives the improvements is not supported.
  2. [Section III-B1, Eq. (1), and Algorithm 1] The view-agnostic branch is described as extracting shareable information from diverse input views, but at inference it computes W_a = H_a(e_a) once from a single learned constant e_a that does not depend on the input images, the scene, or the target view. This is a fixed weight tensor, and adding it is essentially a reparameterization of the predictor's initial weights. The gains from variant (a) to variant (b) in Table IV (0.31 dB on RealEstate10K and 0.45 dB on ACID) may therefore reflect added capacity or the MVSplat initialization rather than any learned cross-view abstraction. Please clarify the intended interpretation and provide an ablation where e_a is conditioned on the input scene, or explicitly state that this branch serves as a learned initialization.
  3. [Section IV-B, Table II] The cross-dataset claim is subject to the same confound and to an additional transfer concern. The view-agnostic embedding is a constant learned on RealEstate10K and is not updated at test time, so it cannot provide scene- or dataset-specific adaptation on ACID or DTU; the only adaptive component is the view-specific branch. Given the large DTU improvement (16.01 vs. 13.94 for MVSplat), please report per-scene standard deviations or error bars and analyze whether the gain is attributable to target-view conditioning by including the capacity-matched and simple-conditioning baselines in the cross-dataset setting. Without this evidence, the generalization claim is not independently established.
minor comments (4)
  1. [Tables and captions] Table I has the header typo 'LPISP' instead of 'LPIPS'; Figure 4's caption contains 'Qualitatitive' and 'corss-dataset'; Table II's caption says 'Qualitative comparison' even though the table is quantitative. These should be corrected.
  2. [Section III-B2, Eq. (10)] The correlation function C_ij(M1, M2) is not precisely defined: the text writes (M_i^1, M_j^2) / (||M_i^1|| ||M_j^2||), but it is unclear whether the superscripts index the two matrices or whether M_i^1 means the i-th row of M1. Please define the indexing explicitly.
  3. [Section III-B2, Eqs. (12)-(13)] The notation is inconsistent: the text says the hypernetwork H_s produces a weight matrix W_s, but Eq. (12) defines ΔW = H_s(e_s) and Eq. (13) writes W_H = W + ΔW. Please align the notation (e.g., set W_s = ΔW) and specify which layers of the predictor are affected.
  4. [Section IV-D, Table III] Table III mixes settings: the ACID column reproduces the intra-dataset numbers from Table I, while the DTU column reproduces the cross-dataset numbers from Table II. Since the table is intended as an efficiency comparison, please state explicitly which training/test protocol is used for each dataset, or report the two settings separately.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity; UniqueSplat's central claim is an empirical architecture comparison against external benchmarks, not a self-referential derivation.

full rationale

UniqueSplat's derivation chain is architectural and empirical, not a fitted-then-predicted cycle. The view-conditioned hypernetwork takes input images and target camera matrices and produces predictor weights W_H = W_a + ΔW, where W_a = H_a(e_a) and ΔW = H_s(e_s). The target-view information is the target pose/camera matrix, not the ground-truth target image; ground-truth images appear only in the training loss L = λ1·L_mse + λ2·L_lpips. The reported improvements (Tables I and II) are evaluated on held-out test scenes of RealEstate10K, ACID, and DTU against external baselines such as MVSplat and TranSplat, with the comparison numbers reported independently in the paper. The ablation in Table IV varies branch presence but not capacity; this is an attribution confound and a correctness/rigor concern, not a circularity, because the ablated variants are still evaluated on held-out data. The only self-citation is reference [19] in a survey-style citation list, and it is not used to justify the architecture, the ablations, or any theorem. No equation defines an output in terms of its own target, no fitted parameter is renamed a prediction, and no uniqueness or prior result is imported from the authors' own work. Hence no significant circularity is present in the paper's derivation chain.

Assumptions & free parameters 7 free parameters · 6 assumptions · 2 invented entities

The method introduces two learned latent embeddings, view-agnostic and view-specific, that are internal to the network and have no falsifiable handle outside image metrics. No physical entities, forces, or conserved quantities are postulated. The hand-chosen hyperparameters above are the user-set degrees of freedom that affect the central claim; none are accompanied by sensitivity analyses. The main assumptions are inherited from MVSplat or introduced by the view-conditioning design.

free parameters (7)
  • MSE loss weight lambda_1 = 1
    Fixed to 1 in Eq. 17; no sensitivity analysis.
  • LPIPS loss weight lambda_2 = 0.05
    Set by hand; no search or ablation reported.
  • LPIPS warm-up iteration = 150,000
    LPIPS is added after 150k iterations; no ablation of this schedule.
  • Embedding dimension = 512
    Used for both view-agnostic and view-specific embeddings; no sensitivity analysis.
  • Hypernetwork depth = 3 layers
    Both hypernetworks are 3-layer MLPs; width and exact target layers are not specified.
  • Learning rate = 1e-6
    Adam learning rate for all experiments; no schedule or grid search reported.
  • Input and target view counts = K=2, N=4
    Follows pixelSplat/MVSplat protocol; not varied.
assumptions (6)
  • domain assumption MVSplat's cost-volume encoder and depth predictor are a suitable primary network for view-conditioned Gaussian prediction.
    The primary network is initialized from and based on MVSplat in Section IV-A; the paper does not test other backbones.
  • domain assumption Target-view camera projection matrices are available at inference time and are used to align source features to the query view.
    The view-specific branch builds feature volumes by projecting source features with P_t in Eqs. 4-10; this is standard for NVS but assumed.
  • ad hoc to paper The learnable view-agnostic embedding trained on the training distribution transfers to unseen scenes and datasets.
    Section III-B.1 defines the embedding as a learnable parameter fixed at test time; cross-dataset claims in Table II depend on its transfer, which is asserted but not analyzed.
  • ad hoc to paper Epipolar projection and normalized correlation in Eq. 10 are sufficient to fuse multi-view features for the target view.
    The projection and correlation operations are introduced without comparison to alternative feature fusion mechanisms.
  • domain assumption MSE plus LPIPS supervision at 256x256 is sufficient to learn high-quality Gaussian parameters.
    The training loss in Eq. 17 is standard and not ablated.
  • domain assumption Pixel-aligned Gaussians predicted from the input views can represent the target view's radiance field.
    The predictor outputs per-pixel Gaussian parameters from source views in Eq. 15; this representation is inherited from MVSplat and not re-derived.
invented entities (2)
  • View-specific embedding e_s
    purpose: Encodes target-view information to generate per-view weight deltas for the predictor.
    It is an internal learned vector with no measurable counterpart outside the trained network; it is validated only through rendered-image metrics.
  • View-agnostic embedding e_a
    purpose: Encodes scene knowledge shared across views and training scenes; fixed at test time.
    A learned constant with no independent falsifiable handle; its transfer to new datasets is asserted.

how reviews work

0 comments
Cite this review

Pith. "Pith review of UniqueSplat: View-conditioned 3D Gaussian Splatting for Generalizable 3D Reconstruction." pith.science (2026). https://pith.science/paper/EEZFTOFP

@misc{pith2026260802145,
  author       = {Pith},
  title        = {Pith review of: UniqueSplat: View-conditioned 3D Gaussian Splatting for Generalizable 3D Reconstruction},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/EEZFTOFP}},
  note         = {Machine review of arXiv:2608.02145}
}
read the original abstract

In this paper, we propose UniqueSplat, a view-conditioned feed-forward 3D Gaussian Splatting model to reconstruct customized 3D radiance fields for each view query. Existing feed-forward methods such as pixelSplat and MVSplat aim to generate fixed Gaussians across all views of each scene by minimizing the error between rendered views and ground-truth images. However, such fixed Gaussians generally render images from all views and lack the ability to adapt to specific viewpoints, as they do not incorporate target view information when predicting Gaussians. To address this, our UniqueSplat learns the view-conditioned information as a prior and incorporates this knowledge into network parameters, so that Gaussians are dynamically adjusted in accordance with different views. Specifically, we propose a two-branch view-conditioned hyperNetwork to simultaneously learn view-agnostic embeddings and view-specific knowledge, which not only explores the shareable knowledge from various views, but also adapts the model to specific views at test time. Extensive experiments on widely-used datasets including RealEstate10K, ACID and DTU demonstrate the superiority of UniqueSplat over the state-of-the-art methods. Moreover, UniqueSplat encouragingly outperforms existing methods in cross-dataset evaluation, showing its notable generalization ability.

Figures

Figures reproduced from arXiv: 2608.02145 by the authors.

Figure 1
Figure 1. Qualitative comparison between MVSplat [1] and UniqueSplat. UniqueSplat incorporates view-conditioned information as a prior to achieve superior [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Overview of our UniqueSplat framework, consisting of two key [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Qualitative comparison on RealEstate10K [65] and ACID [66] benchmarks. All models are trained and tested on the same datasets. Benefiting from [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Qualitatitive comparison for corss-dataset generalization on ACID [66] and DTU [68]. All models are trained on RealEstate10K [65] and tested on [PITH_FULL_IMAGE:figures/full_fig_p009_4.png]
Figure 5
Figure 5. Figure 5: More qualitative comparisons with MVSplat. We train models on RealEstate10K dataset and evaluate them on both RealEstate10K and ACID datasets. [PITH_FULL_IMAGE:figures/full_fig_p010_5.png]
Figure 6
Figure 6. Figure 6: Failure case. While UniqueSplat delivers impressive results, it fails [PITH_FULL_IMAGE:figures/full_fig_p011_6.png]

Review history (2 revisions) →

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

71 extracted references · 44 canonical work pages

  1. [1]

    Mvsplat: Efficient 3d gaussian splatting from sparse multi-view images,

    Y . Chen, H. Xu, C. Zheng, B. Zhuang, M. Pollefeys, A. Geiger, T.- J. Cham, and J. Cai, “Mvsplat: Efficient 3d gaussian splatting from sparse multi-view images,” inEuropean Conference on Computer Vision. Springer, 2024, pp. 370–386

    work page 2024

  2. [2]

    Gps-gaussian: Generalizable pixel-wise 3d gaussian splatting for real- time human novel view synthesis,

    S. Zheng, B. Zhou, R. Shao, B. Liu, S. Zhang, L. Nie, and Y . Liu, “Gps-gaussian: Generalizable pixel-wise 3d gaussian splatting for real- time human novel view synthesis,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 19 680–19 690

    2024

  3. [3]

    H2-mapping: Real-time dense mapping using hierarchical hybrid rep- resentation,

    C. Jiang, H. Zhang, P. Liu, Z. Yu, H. Cheng, B. Zhou, and S. Shen, “H2-mapping: Real-time dense mapping using hierarchical hybrid rep- resentation,”IEEE Robotics and Automation Letters, 2023

    work page 2023

  4. [4]

    Neurad: Neural rendering for autonomous driving,

    A. Tonderski, C. Lindstr ¨om, G. Hess, W. Ljungbergh, L. Svensson, and C. Petersson, “Neurad: Neural rendering for autonomous driving,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 14 895–14 904

    2024

  5. [5]

    Joint-confidence-guided multi- task learning for 3d reconstruction and understanding from monocular camera,

    Y . Wang, Q. Zhao, Y . Gan, and Z. Xia, “Joint-confidence-guided multi- task learning for 3d reconstruction and understanding from monocular camera,”IEEE Transactions on Image Processing, vol. 32, pp. 1120– 1133, 2023

    work page 2023

  6. [6]

    Vidsfm: Robust and accurate structure-from-motion for monocular videos,

    H. Cui, D. Tu, F. Tang, P. Xu, H. Liu, and S. Shen, “Vidsfm: Robust and accurate structure-from-motion for monocular videos,”IEEE Trans- actions on Image Processing, vol. 31, pp. 2449–2462, 2022

    work page 2022

  7. [7]

    Nr-mvsnet: Learning multi-view stereo based on normal consistency and depth refinement,

    J. Li, Z. Lu, Y . Wang, J. Xiao, and Y . Wang, “Nr-mvsnet: Learning multi-view stereo based on normal consistency and depth refinement,” IEEE Transactions on Image Processing, vol. 32, pp. 2649–2662, 2023

    work page 2023

  8. [8]

    Occlusion-aware unsupervised learning of depth from 4-d light fields,

    J. Jin and J. Hou, “Occlusion-aware unsupervised learning of depth from 4-d light fields,”IEEE Transactions on Image Processing, vol. 31, pp. 2216–2228, 2022

    work page 2022

Show all 71 references
  1. [9]

    Neural sparse voxel fields,

    L. Liu, J. Gu, K. Zaw Lin, T.-S. Chua, and C. Theobalt, “Neural sparse voxel fields,”Advances in Neural Information Processing Systems, vol. 33, pp. 15 651–15 663, 2020

    2020

  2. [10]

    Mip-nerf: A multiscale representation for anti- aliasing neural radiance fields,

    J. T. Barron, B. Mildenhall, M. Tancik, P. Hedman, R. Martin-Brualla, and P. P. Srinivasan, “Mip-nerf: A multiscale representation for anti- aliasing neural radiance fields,” inProceedings of the IEEE/CVF Inter- national Conference on Computer Vision, 2021, pp. 5855–5864

    2021

  3. [11]

    Plenoxels: Radiance fields without neural networks,

    S. Fridovich-Keil, A. Yu, M. Tancik, Q. Chen, B. Recht, and A. Kanazawa, “Plenoxels: Radiance fields without neural networks,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2022, pp. 5501–5510

    2022

  4. [12]

    Instant neural graphics primitives with a multiresolution hash encoding,

    T. M ¨uller, A. Evans, C. Schied, and A. Keller, “Instant neural graphics primitives with a multiresolution hash encoding,”ACM Transactions on Graphics, vol. 41, no. 4, pp. 1–15, 2022

    2022

  5. [13]

    Learning spherical radiance field for efficient 360 unbounded novel view synthesis,

    M. Chen, L. Wang, Y . Lei, Z. Dong, and Y . Guo, “Learning spherical radiance field for efficient 360 unbounded novel view synthesis,”IEEE Transactions on Image Processing, 2024

    2024

  6. [14]

    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, vol. 42, no. 4, pp. 1–14, 2023

    2023

  7. [15]

    Multi-scale 3d gaussian splatting for anti-aliased rendering,

    Z. Yan, W. F. Low, Y . Chen, and G. H. Lee, “Multi-scale 3d gaussian splatting for anti-aliased rendering,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 20 923–20 931

    2024

  8. [16]

    Mip-splatting: Alias-free 3d gaussian splatting,

    Z. Yu, A. Chen, B. Huang, T. Sattler, and A. Geiger, “Mip-splatting: Alias-free 3d gaussian splatting,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 19 447–19 456

    2024

  9. [17]

    Splatter image: Ultra- fast single-view 3d reconstruction,

    S. Szymanowicz, C. Rupprecht, and A. Vedaldi, “Splatter image: Ultra- fast single-view 3d reconstruction,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 10 208–10 217

    2024

  10. [18]

    Transplat: Generalizable 3d gaussian splatting from sparse multi-view images with transformers,

    C. Zhang, Y . Zou, Z. Li, M. Yi, and H. Wang, “Transplat: Generalizable 3d gaussian splatting from sparse multi-view images with transformers,” inProceedings of the AAAI Conference on Artificial Intelligence, vol. 39, no. 9, 2025, pp. 9869–9877

    2025

  11. [19]

    Gaussian graph network: Learning efficient and generalizable gaussian representations from multi-view images,

    S. Zhang, X. Fei, F. Liu, H. Song, and Y . Duan, “Gaussian graph network: Learning efficient and generalizable gaussian representations from multi-view images,”Advances in Neural Information Processing Systems, 2024

    2024

  12. [20]

    pixelsplat: 3d gaussian splats from image pairs for scalable generalizable 3d recon- struction,

    D. Charatan, S. L. Li, A. Tagliasacchi, and V . Sitzmann, “pixelsplat: 3d gaussian splats from image pairs for scalable generalizable 3d recon- struction,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 19 457–19 467

    2024

  13. [21]

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

    B. Mildenhall, P. P. Srinivasan, M. Tancik, J. T. Barron, R. Ramamoorthi, and R. Ng, “Nerf: Representing scenes as neural radiance fields for view synthesis,” inEuropean Conference on Computer Vision. Springer, 2020, pp. 405–421

    2020

  14. [22]

    Tensorf: Tensorial radiance fields,

    A. Chen, Z. Xu, A. Geiger, J. Yu, and H. Su, “Tensorf: Tensorial radiance fields,” inEuropean Conference on Computer Vision. Springer, 2022, pp. 333–350

    2022

  15. [23]

    Fast learn- ing radiance fields by shooting much fewer rays,

    W. Zhang, R. Xing, Y . Zeng, Y .-S. Liu, K. Shi, and Z. Han, “Fast learn- ing radiance fields by shooting much fewer rays,”IEEE Transactions on Image Processing, vol. 32, pp. 2703–2718, 2023

    2023

  16. [24]

    Gaus- sianshader: 3d gaussian splatting with shading functions for reflective surfaces,

    Y . Jiang, J. Tu, Y . Liu, X. Gao, X. Long, W. Wang, and Y . Ma, “Gaus- sianshader: 3d gaussian splatting with shading functions for reflective surfaces,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 5322–5332

    2024

  17. [25]

    Gs-ir: 3d gaussian splatting for inverse rendering,

    Z. Liang, Q. Zhang, Y . Feng, Y . Shan, and K. Jia, “Gs-ir: 3d gaussian splatting for inverse rendering,” inProceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition, 2024, pp. 21 644– 21 653

    2024

  18. [26]

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

    T. Lu, M. Yu, L. Xu, Y . Xiangli, L. Wang, D. Lin, and B. Dai, “Scaffold-gs: Structured 3d gaussians for view-adaptive rendering,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 20 654–20 664

    2024

  19. [27]

    Compgs: Smaller and faster gaussian splatting with vector quantiza- tion,

    K. Navaneet, K. P. Meibodi, S. A. Koohpayegani, and H. Pirsiavash, “Compgs: Smaller and faster gaussian splatting with vector quantiza- tion,”European Conference on Computer Vision, 2024

    2024

  20. [28]

    Eagles: Efficient accelerated 3d gaussians with lightweight encodings,

    S. Girish, K. Gupta, and A. Shrivastava, “Eagles: Efficient accelerated 3d gaussians with lightweight encodings,”European Conference on Computer Vision, 2024

    2024

  21. [29]

    Fsgs: Real-time few-shot view synthesis using gaussian splatting,

    Z. Zhu, Z. Fan, Y . Jiang, and Z. Wang, “Fsgs: Real-time few-shot view synthesis using gaussian splatting,” inEuropean Conference on Computer Vision. Springer, 2024, pp. 145–163

    2024

  22. [30]

    Sparsegs: Real-time 360° sparse view synthesis using gaussian splat- ting,

    H. Xiong, S. Muttukuru, R. Upadhyay, P. Chari, and A. Kadambi, “Sparsegs: Real-time 360° sparse view synthesis using gaussian splat- ting,”arXiv preprint arXiv:2312.00206, 2023

    2023 arXiv

  23. [31]

    4d gaussian splatting for real-time dynamic scene rendering,

    G. Wu, T. Yi, J. Fang, L. Xie, X. Zhang, W. Wei, W. Liu, Q. Tian, and X. Wang, “4d gaussian splatting for real-time dynamic scene rendering,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 20 310–20 320

    2024

  24. [32]

    Real-time photorealistic dynamic scene representation and rendering with 4d gaussian splatting,

    Z. Yang, H. Yang, Z. Pan, and L. Zhang, “Real-time photorealistic dynamic scene representation and rendering with 4d gaussian splatting,” inThe Twelfth International Conference on Learning Representations

  25. [33]

    Phys- gaussian: Physics-integrated 3d gaussians for generative dynamics,

    T. Xie, Z. Zong, Y . Qiu, X. Li, Y . Feng, Y . Yang, and C. Jiang, “Phys- gaussian: Physics-integrated 3d gaussians for generative dynamics,” in JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2021 13 Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern R...

    2021

  26. [34]

    3dgs-enhancer: Enhancing unbounded 3d gaussian splatting with view-consistent 2d diffusion priors,

    X. Liu, C. Zhou, and S. Huang, “3dgs-enhancer: Enhancing unbounded 3d gaussian splatting with view-consistent 2d diffusion priors,”Advances in Neural Information Processing Systems, vol. 37, pp. 133 305–133 327, 2024

    2024

  27. [35]

    Mvpgs: Excavating multi-view priors for gaussian splatting from sparse input views,

    W. Xu, H. Gao, S. Shen, R. Peng, J. Jiao, and R. Wang, “Mvpgs: Excavating multi-view priors for gaussian splatting from sparse input views,” inEuropean Conference on Computer Vision. Springer, 2024, pp. 203–220

    2024

  28. [36]

    Lita-gs: Illumination-agnostic novel view synthesis via reference-free 3d gaussian splatting and physical priors,

    H. Zhou, W. Dong, and J. Chen, “Lita-gs: Illumination-agnostic novel view synthesis via reference-free 3d gaussian splatting and physical priors,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2025, pp. 21 580–21 589

    2025

  29. [37]

    Depth-regularized optimization for 3d gaussian splatting in few-shot images,

    J. Chung, J. Oh, and K. M. Lee, “Depth-regularized optimization for 3d gaussian splatting in few-shot images,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops, 2024, pp. 811–820

    2024

  30. [38]

    pixelnerf: Neural radiance fields from one or few images,

    A. Yu, V . Ye, M. Tancik, and A. Kanazawa, “pixelnerf: Neural radiance fields from one or few images,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2021, pp. 4578–4587

    2021

  31. [39]

    Codenerf: Disentangled neural radiance fields for object categories,

    W. Jang and L. Agapito, “Codenerf: Disentangled neural radiance fields for object categories,” inProceedings of the IEEE/CVF International Conference on Computer Vision, 2021, pp. 12 949–12 958

    2021

  32. [40]

    Sharf: Shape- conditioned radiance fields from a single view,

    K. Rematas, R. Martin-Brualla, and V . Ferrari, “Sharf: Shape- conditioned radiance fields from a single view,” inInternational Con- ference on Machine Learning. PMLR, 2021, pp. 8948–8958

    2021

  33. [41]

    Mvsnerf: Fast generalizable radiance field reconstruction from multi-view stereo,

    A. Chen, Z. Xu, F. Zhao, X. Zhang, F. Xiang, J. Yu, and H. Su, “Mvsnerf: Fast generalizable radiance field reconstruction from multi-view stereo,” inProceedings of the IEEE/CVF International Conference on Computer Vision, 2021, pp. 14 124–14 133

    2021

  34. [42]

    Geonerf: Generalizing nerf with geometry priors,

    M. M. Johari, Y . Lepoittevin, and F. Fleuret, “Geonerf: Generalizing nerf with geometry priors,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2022, pp. 18 365–18 375

    2022

  35. [43]

    Neural rays for occlusion-aware image-based rendering,

    Y . Liu, S. Peng, L. Liu, Q. Wang, P. Wang, C. Theobalt, X. Zhou, and W. Wang, “Neural rays for occlusion-aware image-based rendering,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2022, pp. 7824–7833

    2022

  36. [44]

    Stereo radiance fields (srf): Learning view synthesis for sparse views of novel scenes,

    J. Chibane, A. Bansal, V . Lazova, and G. Pons-Moll, “Stereo radiance fields (srf): Learning view synthesis for sparse views of novel scenes,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2021, pp. 7911–7920

    2021

  37. [45]

    Generalizable patch- based neural rendering,

    M. Suhail, C. Esteves, L. Sigal, and A. Makadia, “Generalizable patch- based neural rendering,” inEuropean Conference on Computer Vision. Springer, 2022, pp. 156–174

    2022

  38. [46]

    Is attention all that nerf needs?

    M. Varma, P. Wang, X. Chen, T. Chen, S. Venugopalan, and Z. Wang, “Is attention all that nerf needs?” inThe Eleventh International Conference on Learning Representations

  39. [47]

    Ibrnet: Learning multi-view image-based rendering,

    Q. Wang, Z. Wang, K. Genova, P. P. Srinivasan, H. Zhou, J. T. Barron, R. Martin-Brualla, N. Snavely, and T. Funkhouser, “Ibrnet: Learning multi-view image-based rendering,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2021, pp. 4690–4699

    2021

  40. [48]

    Learning to render novel views from wide-baseline stereo pairs,

    Y . Du, C. Smith, A. Tewari, and V . Sitzmann, “Learning to render novel views from wide-baseline stereo pairs,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023, pp. 4970–4980

    2023

  41. [49]

    Scene representation transformer: Geometry-free novel view synthesis through set-latent scene representations,

    M. S. Sajjadi, H. Meyer, E. Pot, U. Bergmann, K. Greff, N. Radwan, S. V ora, M. Lu ˇci´c, D. Duckworth, A. Dosovitskiyet al., “Scene representation transformer: Geometry-free novel view synthesis through set-latent scene representations,” inProceedings of the IEEE/CVF Con- fer...

    2022

  42. [50]

    Direct voxel grid optimization: Super- fast convergence for radiance fields reconstruction,

    C. Sun, M. Sun, and H.-T. Chen, “Direct voxel grid optimization: Super- fast convergence for radiance fields reconstruction,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2022, pp. 5459–5469

    2022

  43. [51]

    Viewset diffusion:(0-) image-conditioned 3d generative models from 2d data,

    S. Szymanowicz, C. Rupprecht, and A. Vedaldi, “Viewset diffusion:(0-) image-conditioned 3d generative models from 2d data,” inProceedings of the IEEE/CVF International Conference on Computer Vision, 2023, pp. 8863–8873

    2023

  44. [52]

    Renderdiffusion: Image diffusion for 3d recon- struction, inpainting and generation,

    T. Anciukevi ˇcius, Z. Xu, M. Fisher, P. Henderson, H. Bilen, N. J. Mitra, and P. Guerrero, “Renderdiffusion: Image diffusion for 3d recon- struction, inpainting and generation,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023, pp. 12 ...

    2023

  45. [53]

    Nerfdiff: Single-image view synthesis with nerf- guided distillation from 3d-aware diffusion,

    J. Gu, A. Trevithick, K.-E. Lin, J. M. Susskind, C. Theobalt, L. Liu, and R. Ramamoorthi, “Nerfdiff: Single-image view synthesis with nerf- guided distillation from 3d-aware diffusion,” inInternational Conference on Machine Learning. PMLR, 2023, pp. 11 808–11 826

    2023

  46. [54]

    latentsplat: Autoencoding variational gaussians for fast generalizable 3d reconstruc- tion,

    C. Wewer, K. Raj, E. Ilg, B. Schiele, and J. E. Lenssen, “latentsplat: Autoencoding variational gaussians for fast generalizable 3d reconstruc- tion,” inEuropean Conference on Computer Vision. Springer, 2024, pp. 456–473

    2024

  47. [55]

    Hypernetworks,

    D. Ha, A. Dai, and Q. V . Le, “Hypernetworks,”arXiv preprint arXiv:1609.09106, 2016

    2016 arXiv

  48. [56]

    A brief review of hypernetworks in deep learning,

    V . K. Chauhan, J. Zhou, P. Lu, S. Molaei, and D. A. Clifton, “A brief review of hypernetworks in deep learning,”Artificial Intelligence Review, vol. 57, no. 9, p. 250, 2024

    2024

  49. [57]

    Hyperseg: Patch-wise hypernetwork for real-time semantic segmentation,

    Y . Nirkin, L. Wolf, and T. Hassner, “Hyperseg: Patch-wise hypernetwork for real-time semantic segmentation,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2021, pp. 4061–4070

    2021

  50. [58]

    Smash: One-shot model architecture search through hypernetworks,

    A. Brock, T. Lim, J. Ritchie, and N. Weston, “Smash: One-shot model architecture search through hypernetworks,” inInternational Conference on Learning Representations, 2018

    2018

  51. [59]

    Hypersound: Generating implicit neural representations of audio signals with hypernetworks,

    F. Szatkowski, K. J. Piczak, P. Spurek, J. Tabor, and T. Trzci ´nski, “Hypersound: Generating implicit neural representations of audio signals with hypernetworks,”arXiv preprint arXiv:2211.01839, 2022

    2022 arXiv

  52. [60]

    D’oh: Decoder-only random hypernetworks for implicit neural representa- tions,

    C. Gordon, L. E. MacDonald, H. Saratchandran, and S. Lucey, “D’oh: Decoder-only random hypernetworks for implicit neural representa- tions,” inProceedings of the Asian Conference on Computer Vision, 2024, pp. 2507–2526

    2024

  53. [61]

    Transformers as meta-learners for implicit neural representations,

    Y . Chen and X. Wang, “Transformers as meta-learners for implicit neural representations,” inEuropean Conference on Computer Vision. Springer, 2022, pp. 170–187

    2022

  54. [62]

    Hyperinr: A fast and pre- dictive hypernetwork for implicit neural representations via knowledge distillation,

    Q. Wu, D. Bauer, Y . Chen, and K.-L. Ma, “Hyperinr: A fast and pre- dictive hypernetwork for implicit neural representations via knowledge distillation,”arXiv preprint arXiv:2304.04188, 2023

    2023 arXiv

  55. [63]

    Hyp-nerf: Learning improved nerf priors using a hypernetwork,

    B. Sen, G. Singh, A. Agarwal, R. Agaram, M. Krishna, and S. Srid- har, “Hyp-nerf: Learning improved nerf priors using a hypernetwork,” Advances in Neural Information Processing Systems, vol. 36, 2023

    2023

  56. [64]

    Explicit correspondence matching for generalizable neural radiance fields,

    Y . Chen, H. Xu, Q. Wu, C. Zheng, T.-J. Cham, and J. Cai, “Explicit correspondence matching for generalizable neural radiance fields,”IEEE Transactions on Pattern Analysis and Machine Intelligence, 2025

    2025

  57. [65]

    Stereo magnification: Learning view synthesis using multiplane images,

    T. Zhou, R. Tucker, J. Flynn, G. Fyffe, and N. Snavely, “Stereo magnification: Learning view synthesis using multiplane images,”arXiv preprint arXiv:1805.09817, 2018

    2018 arXiv

  58. [66]

    Infinite nature: Perpetual view generation of natural scenes from a single image,

    A. Liu, R. Tucker, V . Jampani, A. Makadia, N. Snavely, and A. Kanazawa, “Infinite nature: Perpetual view generation of natural scenes from a single image,” inProceedings of the IEEE/CVF Inter- national Conference on Computer Vision, 2021, pp. 14 458–14 467

    2021

  59. [67]

    Murf: Multi-baseline radiance fields,

    H. Xu, A. Chen, Y . Chen, C. Sakaridis, Y . Zhang, M. Pollefeys, A. Geiger, and F. Yu, “Murf: Multi-baseline radiance fields,” inPro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 20 041–20 050

    2024

  60. [68]

    Large scale multi-view stereopsis evaluation,

    R. Jensen, A. Dahl, G. V ogiatzis, E. Tola, and H. Aanaes, “Large scale multi-view stereopsis evaluation,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2014, pp. 406– 413

    2014

  61. [69]

    Image quality assessment: from error visibility to structural similarity,

    Z. Wang, A. C. Bovik, H. R. Sheikh, and E. P. Simoncelli, “Image quality assessment: from error visibility to structural similarity,”IEEE Transactions on Image Processing, vol. 13, no. 4, pp. 600–612, 2004

    2004

  62. [70]

    The unreasonable effectiveness of deep features as a perceptual metric,

    R. Zhang, P. Isola, A. A. Efros, E. Shechtman, and O. Wang, “The unreasonable effectiveness of deep features as a perceptual metric,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2018, pp. 586–595

    2018

  63. [71]

    High- resolution image synthesis with latent diffusion models,

    R. Rombach, A. Blattmann, D. Lorenz, P. Esser, and B. Ommer, “High- resolution image synthesis with latent diffusion models,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recogni- tion, 2022, pp. 10 684–10 695. JOURNAL OF LATEX CLASS FILES, VOL. 14, ...

    2022

Pith tools

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