Pith. sign in

REVIEW 3 major objections 8 minor 80 references

Texture energy lets a feed-forward model keep far fewer 3D Gaussians without black patches or fine-tuning.

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

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

T0 review · grok-4.5

2026-07-11 20:33 UTC pith:HCQMDZQJ

load-bearing objection Solid engineering fix for controllable sparsity in feed-forward 3DGS; the DWT proxy is imperfect but the gains and single-pass design still hold up. the 3 major comments →

arxiv 2607.04256 v1 pith:HCQMDZQJ submitted 2026-07-05 cs.CV

AdaptiveSplat:Texture Aware Controllable 3D Gaussian Allocation for Feed-Forward Reconstruction

Badrinath Singhal , Srihari K G , Sreehari Iyer , Ankit Dhiman , Venkatesh Babu Radhakrishnan This is my paper
classification cs.CV
keywords 3D Gaussian Splattingfeed-forward reconstructiontexture-aware pruningcontrollable Gaussian allocationnovel view synthesisDiscrete Wavelet TransformAdaptive Gaussian Head
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

Feed-forward 3D reconstruction methods normally emit one Gaussian per pixel, so the representation is dense and redundant. Simply deleting Gaussians produces black holes or forces a slow per-scene fine-tune that destroys the feed-forward property. AdaptiveSplat treats natural-image statistics as a guide: low-frequency, smooth regions can be covered by fewer, larger Gaussians while high-frequency textured regions keep dense coverage. It estimates local texture energy with a discrete wavelet transform, groups points into SuperClusters, prunes aggressively inside low-energy clusters under a user budget β, and re-predicts the remaining Gaussians’ scales, opacities and appearance with a single adaptive head. The result is a controllable, still fully feed-forward pipeline that keeps novel-view quality high even when 80 % of the primitives are removed.

Core claim

A single feed-forward pass that ranks SuperClusters by wavelet texture energy, retains only a budgeted fraction of Gaussians inside low-energy clusters, and re-predicts the attributes of the survivors yields novel-view fidelity that substantially exceeds both naïve and state-of-the-art post-hoc pruning methods (with or without fine-tuning) across RE10K, ACID, DL3DV, Tanks and Temples and DTU.

What carries the argument

Texture-aware SuperCluster mask + Adaptive Gaussian Head: binary masks built from DWT energy order the SuperClusters and mark which pixel-aligned Gaussians survive; a DPT-style head then conditions on those masks to output corrected scales, rotations, opacities and spherical harmonics for the retained set, all without test-time optimisation.

Load-bearing premise

Local wavelet energy averaged inside colour-and-position SuperClusters is assumed to be a reliable enough proxy for how many Gaussians a region actually needs; specular highlights and light-induced high frequencies can mis-rank flat surfaces.

What would settle it

On a scene dominated by specular highlights or strong light-induced high frequencies, raise β to 0.8 and check whether the method still avoids black patches and over-smoothing while matching or exceeding the PSNR of a fine-tuned baseline; if quality collapses relative to a geometry-aware oracle, the texture proxy has failed.

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

If this is right

  • Users can dial Gaussian count at inference time (β ∈ [0, 0.8]) and still obtain artifact-free novel views without any per-scene optimisation.
  • Rendering FPS rises monotonically with pruning strength while PSNR remains competitive, giving a practical quality–speed knob for real-time pipelines.
  • The same mask-and-adapt module can be dropped onto existing feed-forward backbones (MASt3R, VGGT, MVSplat, InstantSplat) without architectural redesign.
  • Cross-domain zero-shot transfer (train on ACID/RE10K, test on DTU) remains stronger than competing pruning strategies, indicating the texture prior generalises beyond the training distribution.

Where Pith is reading between the lines

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

  • If the texture proxy can be replaced by a joint texture–geometry score, the remaining failure modes under specular surfaces would shrink, opening the door to even higher pruning ratios.
  • The same SuperCluster-plus-adaptive-head pattern could be applied to dynamic or multi-object scenes where temporal or semantic frequency varies, not only static image texture.
  • Because the method already produces larger Gaussians in smooth regions, it may serve as a drop-in initialiser for later compression or quantisation stages that currently start from uniform dense sets.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 8 minor

Summary. The paper proposes AdaptiveSplat, a feed-forward pipeline that makes the number of predicted 3D Gaussians controllable under a user budget β. Starting from pixel-aligned Gaussians produced by MASt3R/VGGT/MVSplat-style backbones, it (1) estimates local texture energy via single-level 2D DWT, (2) partitions the multi-view point cloud into SuperClusters (K-means in xyz+rgb) and builds a binary retention mask that prunes more aggressively in low-energy clusters (fixed retention ratio γ), and (3) re-predicts scale, rotation, opacity and SH of the retained primitives with an Adaptive Gaussian Head conditioned on features and the mask. Training samples β ~ U[0,1) with photometric+LPIPS loss so that a single head generalizes across budgets. Experiments on RE10K, ACID, DL3DV, Tanks & Temples and DTU, with and without per-scene fine-tuning of baselines, report large gains over EAGLES, LightGaussian, PUP-3DGS and EfficientGS at β ∈ {0.4,0.6,0.8}, plus ablations on texture ranking, the adaptive head, the mask, K and γ.

Significance. If the results hold under scrutiny, the work supplies a practical, plug-in mechanism for controllable Gaussian density in feed-forward reconstruction without test-time optimization—an important efficiency knob for real-time novel-view synthesis and storage. Strengths include multi-dataset evaluation (sparse and dense protocols), explicit with/without fine-tuning baselines, stronger two-stage adaptive-head baselines in the supplement, hyper-parameter sweeps on K and γ, cross-domain DTU tests, and FPS/Pareto latency analysis. The Adaptive Gaussian Head ablation is particularly clear: removing it collapses quality, supporting the claim that attribute re-prediction is necessary after pruning. The contribution is primarily empirical/systems rather than theoretical, but it is well-scoped to a real limitation of current feed-forward 3DGS pipelines.

major comments (3)
  1. Sec. 3.1–3.2 and Supp. G: The central allocation rule ranks SuperClusters by average single-level DWT energy E_SCi and retains only γ of primitives in low-energy clusters. Supp. G / Fig. S show that specular highlights and light-induced high frequencies can assign high energy inside otherwise flat SuperClusters, so the mask retains redundant Gaussians (or, conversely, can under-protect needed structure). Because the Adaptive Head only re-predicts attributes of the already-selected set and cannot restore deleted geometry, systematic mis-ranking is load-bearing for quality at high β. Please quantify how often this failure mode occurs on RE10K/ACID/DL3DV (e.g., fraction of SuperClusters whose DWT rank disagrees with a simple gradient/edge oracle, or PSNR drop restricted to specular regions), and either (a) add a simple robustification (multi-level DWT, gradient magnitude, or opacity/scale-a
  2. Tab. 1–2 vs. Supp. Tables A–B and naive baselines (Supp. D): The headline gains are largest against post-hoc pruning without fine-tuning, where black-patch artifacts make baselines near-unusable. The stronger two-stage baselines that also train an adaptive head on pruning masks (Supp. A–B) still lose, but the margin shrinks and is essential for the claim of texture-aware selection (not merely “any adaptive head”). Please promote a compact version of Tables A–B into the main paper and report the same β grid for all backbones, so readers can separate (i) benefit of adaptation, (ii) benefit of texture ranking, and (iii) benefit of staying single-pass feed-forward. Also clarify whether β is applied to the same N for every method (pixel count vs. post-fusion count) so B is comparable.
  3. Sec. 3.2 Binary Mask Construction and Tab. 4: The greedy rule that covers low-energy SuperClusters with γ|SCj| until budget B, with fixed γ=0.1 and K=300, is entirely empirical. Ablations show γ=0.1 is best among {0.05,0.10,0.15,0.20} and K≈300 saturates, but there is no analysis of sensitivity to scene scale, number of views, or resolution, nor a comparison to continuous soft allocation (e.g., energy-proportional retention). Given that the Adaptive Head is trained with random β, a soft or learned retention schedule would better match the training distribution and reduce dependence on hand-chosen γ. At minimum, report variance of metrics over a small grid of (K,γ) on held-out scenes and discuss failure when low-texture area fraction is very small or very large.
minor comments (8)
  1. Fig. 3 pipeline diagram is heavily duplicated/overplotted in the manuscript PDF text; please provide a single clean schematic with consistent labels for (a) backbone, (b) SuperCluster+DWT, (c) Adaptive GS Head.
  2. Notation: N = mHW is stated, but multi-view fusion and SuperClustering can change the effective primitive count before masking; define N and B after fusion explicitly in Sec. 3.
  3. Eq. (1): λ=0.001 is given; state whether LPIPS is computed on full resolution or patches and whether any depth/geometry regularizer was tried (even if unused).
  4. Sec. 4.1: training is “48 GPU hours on each dataset” on one A100—clarify whether the backbone is frozen and only the Adaptive Head (and any DPT layers) are trained, and whether the same schedule is used for MVSplat vs. MASt3R/VGGT variants.
  5. Fig. 5–8 qualitative panels would benefit from zoomed insets on high-frequency regions (text, edges) and a consistent β label per column; some composite figures appear to mix RE10K and ACID without clear row labels.
  6. Related Work: wavelet use in WaveNeRF is cited; a short sentence contrasting image-space frequency allocation vs. explicit 3D Gaussian count control would help position the contribution.
  7. Typos/style: “finetune/finetuning” vs. “fine-tuning”, occasional missing spaces after periods in the abstract/intro, and “SuperCluster” capitalization should be consistent.
  8. Supp. latency Tab. H: report mean±std over scenes and include mask/DWT overhead separately from Gaussian Head forward pass so the feed-forward claim is fully costed.

Circularity Check

0 steps flagged

No circularity: empirical feed-forward method with standard photometric training; texture proxy and hyperparameters are design choices validated by ablation, not fitted quantities re-presented as predictions.

full rationale

AdaptiveSplat proposes a controllable pruning pipeline (DWT texture energy, SuperCluster ranking, binary masks, Adaptive Gaussian Head) whose outputs are novel-view metrics under a user budget β. Nothing is defined in terms of the reported PSNR/SSIM/LPIPS; training samples β ~ U[0,1) and optimizes ordinary L2 + LPIPS on rendered targets (Eq. 1). The motivation that high-frequency regions need denser Gaussians is taken from external citations [43,54,56] and a toy experiment (Fig. 2), not from a self-derived uniqueness theorem. Hyperparameters K=300 and γ=0.1 are chosen by ablation (Tab. 4, Supp. H) and then held fixed; they are not fitted to a subset of the evaluation metrics and then called predictions. Failure cases (Supp. G) are openly acknowledged rather than papered over by definitional equivalence. The strongest claim is therefore an empirical performance comparison against external pruning baselines, fully self-contained against public benchmarks, with no reduction of any reported quantity to its own inputs by construction.

Axiom & Free-Parameter Ledger

3 free parameters · 3 axioms · 2 invented entities

The central claim rests on three free hyper-parameters chosen by ablation, the domain assumption that natural-image power spectra justify texture-based allocation, and two invented algorithmic constructs (SuperClusters and the Adaptive Gaussian Head) that have no independent existence outside this pipeline.

free parameters (3)
  • K (number of SuperClusters) = 300
    Fixed at 300 after ablation; controls granularity of texture ranking and directly affects which Gaussians survive.
  • γ (retention ratio inside low-energy SuperClusters) = 0.1
    Fixed at 0.1 after ablation; determines how aggressively low-texture regions are thinned.
  • λ (LPIPS weight) = 0.001
    Set to 0.001; balances photometric and perceptual terms in the training loss.
axioms (3)
  • domain assumption Natural images obey a power-law PSD (PSD ∝ 1/f^α, α≈2), so low-frequency regions can be represented by fewer, larger Gaussians.
    Invoked in Sec. 1 and used to justify texture-aware rather than uniform or geometry-only pruning.
  • ad hoc to paper Single-level 2-D DWT magnitude summed over horizontal/vertical/diagonal bands is a sufficient local texture energy measure.
    Defined in Sec. 3.1; no comparison to other texture descriptors is provided.
  • ad hoc to paper K-means in (x,y,z,r,g,b) yields SuperClusters that are homogeneous enough for uniform retention ratios.
    Sec. 3.2; the clustering space and distance are design choices without theoretical guarantee.
invented entities (2)
  • SuperCluster no independent evidence
    purpose: Spatially and photometrically coherent groups of points that receive a single texture energy score and a shared retention budget.
    Defined via K-means in Sec. 3.2; no independent geometric or photometric validation outside the pruning pipeline.
  • Adaptive Gaussian Head no independent evidence
    purpose: DPT-style network that re-predicts scale/rotation/opacity/SH of retained Gaussians conditioned on the binary mask.
    Sec. 3.3; the architecture is a standard DPT with extra conv layers, but the mask-conditioning mechanism is paper-specific.

pith-pipeline@v1.1.0-grok45 · 35931 in / 2805 out tokens · 26489 ms · 2026-07-11T20:33:58.395248+00:00 · methodology

0 comments
read the original abstract

Current feed-forward 3D reconstruction methods predict pixel aligned Gaussian primitives, resulting in highly redundant representations. A natural solution is to prune the redundant Gaussians, but naive pruning introduces severe artifacts and often requires inference time fine-tuning, breaking the feed-forward paradigm. Based on previous works, high frequency regions require more Gaussian primitives, while low frequency regions can be represented with significantly fewer primitives. Motivated by this, we propose a novel approach to explicitly control the number of Gaussians by leveraging local texture information. Our approach achieves this through three key components: (1) texture estimation to capture spatial variation in scene detail, (2) texture-aware pruning that removes redundant Gaussians from low frequency regions, and (3) an adaptive Gaussian head that predicts the modified attributes of the retained primitives without breaking the feed-forward paradigm. Experiments on RE10K, ACID, DL3DV, Tanks and Temples, and DTU demonstrate the effectiveness of our approach, while ablation studies validate the contributions of its key components.

Figures

Figures reproduced from arXiv: 2607.04256 by Ankit Dhiman, Badrinath Singhal, Sreehari Iyer, Srihari K G, Venkatesh Babu Radhakrishnan.

Figure 1
Figure 1. Figure 1: Overview: We demonstrate AdaptiveSplat’s ability to maintain scene fidelity while controlling the allocation of primitives. Given input images and a target Gaussian budget, our feed-forward approach produces sparse yet high-quality reconstructions. In contrast, prior methods exhibit artifacts and over-smoothing at high sparsity. Abstract. Current feed-forward 3D reconstruction methods predict pixel aligned… view at source ↗
Figure 2
Figure 2. Figure 2: Toy experiment demonstrating core insight for AdaptiveSplat. (Left) A ren￾dered view shows visible artifacts, such as "black patches" that are indicative of naive pruning in low-texture regions. (Right) The same scene is rendered after refining the remaining Gaussians. The rendering quality is significantly restored as the surviving Gaussians in the low-texture areas have adaptively expanded to fill the ga… view at source ↗
Figure 3
Figure 3. Figure 3: Our pipeline allocates the desired number of 3D Gaussians for real-time, novel￾view synthesis. (a) Initial Representation: We infer a 3D point cloud from context images using the MASt3R/VGGT backbone. (b) SuperCluster Formation and Texture Analysis: The point cloud is partitioned into SuperClusters, and texture information is simultaneously analyzed using DWT to identify low-texture regions. (c) Gaussian p… view at source ↗
Figure 4
Figure 4. Figure 4: Within each SuperCluster, Gaussians are ranked by texture en￾ergy, and the top 10% (γ = 0.1) are kept while the rest are masked out. This rule is applied only to low-energy clusters; high-energy clusters are fully retained. {SC1, . . . , SCK} denote the regions or￾dered by increasing energy ESCj . Each region SCj has a nominal retention of γ|SCj | Gaussians, where γ = 0.1 (empir￾ically validated, ablation … view at source ↗
Figure 5
Figure 5. Figure 5: Comparison of results on RE10K and ACID scenes using β = 0.4 and β = 0.8 of target budget. We allocate fewer but larger 3D Gaussians to low texture SuperClusters by the feed-forward backbone. AdaptiveSplat reaches 20.45dB PSNR versus 11.36dB from LightGaussian. No￾tably, while baseline methods degrade significantly at high pruning (β = 0.8), our method maintains strong performance across all view counts. O… view at source ↗
Figure 6
Figure 6. Figure 6: Qualitative comparison of our method with baseline methods (with fine￾tuning) on a single scene from RE10K and ACID datasets. Our method faithfully represents a scene even with β = 0.8 (80% pruning) [PITH_FULL_IMAGE:figures/full_fig_p011_6.png] view at source ↗
Figure 8
Figure 8. Figure 8: Texture Analysis (varying scene): Across the scenes we observe oversmoothed regions whereas we retain the texture of the scene. These results consistently demonstrate that competing methods introduce excessive smoothing in high-texture areas, whereas our method preserves struc￾tural details more effectively under increasing pruning strengths [PITH_FULL_IMAGE:figures/full_fig_p014_8.png] view at source ↗
Figure 10
Figure 10. Figure 10: Plot of rendering speed in Frame Per Second (FPS) with increasing pruning strength [PITH_FULL_IMAGE:figures/full_fig_p014_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Pareto Plot for inference time and quality comparison for RE10K at β = 0.8 6 Conclusions and Future Work We introduce AdaptiveSplat, a novel, feed-forward framework for controllable 3D Gaussian pruning. Our method’s core novelty is texture-aware Gaussian con￾trol that significantly reduces primitives by targeting low-texture regions. By eliminating the need for expensive per-scene optimization, AdaptiveSp… view at source ↗

discussion (0)

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

Reference graph

Works this paper leans on

80 extracted references · 1 canonical work pages

  1. [1]

    International Journal of Computer Vision pp

    Aanæs, H., Jensen, R.R., Vogiatzis, G., Tola, E., Dahl, A.B.: Large-scale data for multiple-view stereopsis. International Journal of Computer Vision pp. 1–16 (2016)

    2016

  2. [2]

    arXiv preprint arXiv:2406.18214 (2024)

    Ali, M.S., Qamar, M., Bae, S.H., Tartaglione, E.: Trimming the fat: Efficient com- pression of 3d gaussian splats through pruning. arXiv preprint arXiv:2406.18214 (2024)

    Pith/arXiv arXiv 2024

  3. [3]

    Asim, M., Wewer, C., Wimmer, T., Schiele, B., Lenssen, J.E.: Met3r: Measuring multi-view consistency in generated images (2024)

    2024

  4. [4]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

    Charatan,D.,Li,S.L.,Tagliasacchi,A.,Sitzmann,V.:pixelsplat:3dgaussiansplats from image pairs for scalable generalizable 3d reconstruction. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 19457– 19467 (2024)

    2024

  5. [5]

    arXiv preprint arXiv:2404.18929 (2024)

    Chen, M., Laina, I., Vedaldi, A.: Dge: Direct gaussian 3d editing by consistent multi-view editing. arXiv preprint arXiv:2404.18929 (2024)

    Pith/arXiv arXiv 2024

  6. [6]

    Chen,Y.,Wu,Q.,Lin,W.,Harandi,M.,Cai,J.:Hac:Hash-gridassistedcontextfor 3d gaussian splatting compression (2024),https://arxiv.org/abs/2403.14530

    Pith/arXiv arXiv 2024

  7. [7]

    arXiv preprint arXiv:2403.14627 (2024)

    Chen, Y., Xu, H., Zheng, C., Zhuang, B., Pollefeys, M., Geiger, A., Cham, T.J., Cai, J.: Mvsplat: Efficient 3d gaussian splatting from sparse multi-view images. arXiv preprint arXiv:2403.14627 (2024)

    Pith/arXiv arXiv 2024

  8. [8]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

    Dhiman, A., Lu, T., Srinath, R., Arslan, E., Xing, A., Xiangli, Y., Radhakrishnan, V.B., Sridhar, S.: Turbo-gs: Accelerating 3d gaussian fitting for high-resolution radiance fields. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 15454–15464 (2026)

    2026

  9. [9]

    In: Proceed- ings of the IEEE/CVF International Conference on Computer Vision (ICCV)

    Dhiman, A., Srinath, R., Rangwani, H., Parihar, R., Boregowda, L.R., Sridhar, S., Babu, R.V.: Strata-nerf : Neural radiance fields for stratified scenes. In: Proceed- ings of the IEEE/CVF International Conference on Computer Vision (ICCV). pp. 17603–17614 (October 2023)

    2023

  10. [10]

    In: The Thirteenth International Conference on Learning Representations (2025),https://openreview.net/forum?id=7JUrBLDjCq

    Dong, J., Wang, Y.X.: 3DGS-drag: Dragging gaussians for intuitive point-based 3d editing. In: The Thirteenth International Conference on Learning Representations (2025),https://openreview.net/forum?id=7JUrBLDjCq

    2025

  11. [11]

    arXiv preprint arXiv:2010.11929 (2020)

    Dosovitskiy, A., Beyer, L., Kolesnikov, A., Weissenborn, D., Zhai, X., Unterthiner, T., Dehghani, M., Minderer, M., Heigold, G., Gelly, S., et al.: An image is worth 16x16 words: Transformers for image recognition at scale. arXiv preprint arXiv:2010.11929 (2020)

    Pith/arXiv arXiv 2010

  12. [12]

    arXiv preprint arXiv:2403.20309 (2024)

    Fan, Z., Cong, W., Wen, K., Wang, K., Zhang, J., Ding, X., Xu, D., Ivanovic, B., Pavone, M., Pavlakos, G., et al.: Instantsplat: Sparse-view gaussian splatting in seconds. arXiv preprint arXiv:2403.20309 (2024)

    Pith/arXiv arXiv 2024

  13. [13]

    Fan, Z., Wang, K., Wen, K., Zhu, Z., Xu, D., Wang, Z.: Lightgaussian: Unbounded 3d gaussian compression with 15x reduction and 200+ fps (2023)

    2023

  14. [14]

    In: Proceedings of the European Conference on Computer Vision (ECCV)

    Fang,G.,Wang,B.:Mini-splatting:Representingsceneswithaconstrainednumber of gaussians. In: Proceedings of the European Conference on Computer Vision (ECCV). pp. 165–181. Springer (2024)

    2024

  15. [15]

    In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) (2024)

    Fang, J., Wang, J., Zhang, X., Xie, L., Tian, Q.: Gaussianeditor: Editing 3d gaus- sians delicately with text instructions. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) (2024)

    2024

  16. [16]

    Advances in Neural Information Processing Systems (2024) AdaptiveSplat 17

    Gao, R., Holynski, A., Henzler, P., Brussee, A., Martin-Brualla, R., Srinivasan, P., Barron, J.T., Poole, B.: Cat3d: Create anything in 3d with multi-view diffusion models. Advances in Neural Information Processing Systems (2024) AdaptiveSplat 17

    2024

  17. [17]

    In: Proceedings of the European Conference on Com- puter Vision (ECCV)

    Girish, S., Gupta, K., Shrivastava, A.: Eagles: Efficient accelerated 3d gaussians with lightweight encodings. In: Proceedings of the European Conference on Com- puter Vision (ECCV). pp. 54–71. Springer (2024)

    2024

  18. [18]

    In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR)

    Hanson, A., Tu, A., Singla, V., Jayawardhana, M., Zwicker, M., Goldstein, T.: Pup 3d-gs: Principled uncertainty pruning for 3d gaussian splatting. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR). pp. 5949–5958 (June 2025),https://pup3dgs.github.io/

    2025

  19. [19]

    In: Proceedings of the IEEE/CVF Interna- tional Conference on Computer Vision (2023)

    Haque, A., Tancik, M., Efros, A., Holynski, A., Kanazawa, A.: Instruct-nerf2nerf: Editing 3d scenes with instructions. In: Proceedings of the IEEE/CVF Interna- tional Conference on Computer Vision (2023)

    2023

  20. [20]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

    Hyung,J.,Hwang,S.,Kim,D.,Lee,H.,Choo,J.:Local3deditingvia3ddistillation of clip knowledge. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 12674–12684 (2023)

    2023

  21. [21]

    In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) (June 2014)

    Jensen, R., Dahl, A., Vogiatzis, G., Tola, E., Aanaes, H.: Large scale multi-view stereopsis evaluation. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) (June 2014)

    2014

  22. [22]

    In: ACM SIGGRAPH (2021)

    Jiakai, Z., Xinhang, L., Xinyi, Y., Fuqiang, Z., Yanshun, Z., Minye, W., Yingliang, Z., Lan, X., Jingyi, Y.: Editable free-viewpoint video using a layered neural repre- sentation. In: ACM SIGGRAPH (2021)

    2021

  23. [23]

    ACM Transactions on Graphics (TOG)44(6), 1–16 (2025)

    Jiang, L., Mao, Y., Xu, L., Lu, T., Ren, K., Jin, Y., Xu, X., Yu, M., Pang, J., Zhao, F., et al.: Anysplat: Feed-forward 3d gaussian splatting from unconstrained views. ACM Transactions on Graphics (TOG)44(6), 1–16 (2025)

    2025

  24. [24]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (2024)

    Keetha, N., Karhade, J., Jatavallabhula, K.M., Yang, G., Scherer, S., Ramanan, D., Luiten, J.: Splatam: Splat, track & map 3d gaussians for dense rgb-d slam. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (2024)

    2024

  25. [25]

    In: International Con- ference on 3D Vision (3DV)

    Keetha, N., Müller, N., Schönberger, J., Porzi, L., Zhang, Y., Fischer, T., Knapitsch, A., Zauss, D., Weber, E., Antunes, N., Luiten, J., Lopez-Antequera, M., Bulò, S.R., Richardt, C., Ramanan, D., Scherer, S., Kontschieder, P.: MapA- nything: Universal feed-forward metric 3D reconstruction. In: International Con- ference on 3D Vision (3DV). IEEE (2026)

    2026

  26. [26]

    ACM Transactions on Graphics42(4) (July 2023),https://repo-sam.inria.fr/fungraph/3d-gaussian-splatting/

    Kerbl, B., Kopanas, G., Leimkühler, T., Drettakis, G.: 3d gaussian splatting for real-time radiance field rendering. ACM Transactions on Graphics42(4) (July 2023),https://repo-sam.inria.fr/fungraph/3d-gaussian-splatting/

    2023

  27. [27]

    arXiv:2403.14370 (2024)

    Kim, J., Koo, J., Yeo, K., Sung, M.: Synctweedies: A general generative framework based on synchronized diffusions. arXiv:2403.14370 (2024)

    Pith/arXiv arXiv 2024

  28. [28]

    ACM Transactions on Graphics36(4) (2017)

    Knapitsch, A., Park, J., Zhou, Q.Y., Koltun, V.: Tanks and temples: Benchmarking large-scale scene reconstruction. ACM Transactions on Graphics36(4) (2017)

    2017

  29. [29]

    In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) (2024)

    Koo, J., Park, C., Sung, M.: Posterior distillation sampling. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) (2024)

    2024

  30. [30]

    In: Proceedings of the European Conference on Computer Vision (ECCV)

    Leroy, V., Cabon, Y., Revaud, J.: Grounding image matching in 3d with mast3r. In: Proceedings of the European Conference on Computer Vision (ECCV). pp. 71–91. Springer (2024)

    2024

  31. [31]

    Li, X., Wang, T., Gu, Z., Zhang, S., Guo, C., Cao, L.: Flashworld: High-quality 3d scene generation within seconds (2025)

    2025

  32. [32]

    In: International Conference on Learning Representations (ICLR) (2025)

    Lin, C., Pan, P., Yang, B., Li, Z., Mu, Y.: Diffsplat: Repurposing image diffusion models for scalable 3d gaussian splat generation. In: International Conference on Learning Representations (ICLR) (2025)

    2025

  33. [33]

    3d vision

    Ling, L., Sheng, Y., Tu, Z., Zhao, W., Xin, C., Wan, K., Yu, L., Guo, Q., Yu, Z., Lu, Y., et al.: Dl3dv-10k: A large-scale scene dataset for deep learning-based 18 Singhal et al. 3d vision. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 22160–22169 (2024)

    2024

  34. [34]

    Liu, A., Tucker, R., Jampani, V., Makadia, A., Snavely, N., Kanazawa, A.: Infinite nature: Perpetual view generation of natural scenes from a single image (2021), https://arxiv.org/abs/2012.09855

    Pith/arXiv arXiv 2021

  35. [35]

    In: The Thirteenth International Conference on Learning Representations (2025),https: //openreview.net/forum?id=2prShxdLkX

    LIU, Q., Liu, Y., Wang, J., Lyu, X., Wang, P., Wang, W., Hou, J.: MoDGS: Dynamic gaussian splatting from casually-captured monocular videos. In: The Thirteenth International Conference on Learning Representations (2025),https: //openreview.net/forum?id=2prShxdLkX

    2025

  36. [36]

    Liu, R., Wu, R., Hoorick, B.V., Tokmakov, P., Zakharov, S., Vondrick, C.: Zero-1- to-3: Zero-shot one image to 3d object (2023)

    2023

  37. [37]

    IEEE MultiMedia32(1), 61–71 (2025).https://doi.org/10.1109/ MMUL.2025.3543224

    Liu, W., Guan, T., Zhu, B., Xu, L., Song, Z., Li, D., Wang, Y., Yang, W.: Effi- cientgs: Streamlining gaussian splatting for large-scale high-resolution scene rep- resentation. IEEE MultiMedia32(1), 61–71 (2025).https://doi.org/10.1109/ MMUL.2025.3543224

    arXiv 2025

  38. [38]

    Lu, T., Yu, M., Xu, L., Xiangli, Y., Wang, L., Lin, D., Dai, B.: Scaffold-gs: Struc- tured 3d gaussians for view-adaptive rendering (2023),https://arxiv.org/abs/ 2312.00109

    Pith/arXiv arXiv 2023

  39. [39]

    In: 3DV (2024)

    Luiten, J., Kopanas, G., Leibe, B., Ramanan, D.: Dynamic 3d gaussians: Tracking by persistent dynamic view synthesis. In: 3DV (2024)

    2024

  40. [40]

    arXiv preprint arXiv:2501.05427 (2024)

    Meng, X., Wang, C., Lei, J., Daniilidis, K., Gu, J., Liu, L.: Zero-1-to-g: Tam- ing pretrained 2d diffusion model for direct 3d generation. arXiv preprint arXiv:2501.05427 (2024)

    Pith/arXiv arXiv 2024

  41. [41]

    In: Pro- ceedings of the European Conference on Computer Vision (ECCV) (2020)

    Mildenhall, B., Srinivasan, P.P., Tancik, M., Barron, J.T., Ramamoorthi, R., Ng, R.: Nerf: Representing scenes as neural radiance fields for view synthesis. In: Pro- ceedings of the European Conference on Computer Vision (ECCV) (2020)

    2020

  42. [42]

    In: Proceedings of the European Conference on Computer Vision (ECCV)

    Morgenstern, W., Barthel, F., Hilsmann, A., Eisert, P.: Compact 3d scene rep- resentation via self-organizing gaussian grids. In: Proceedings of the European Conference on Computer Vision (ECCV). pp. 18–34. Springer Nature Switzer- land, Cham (2025).https://doi.org/10.1007/978-3-031-73013-9_2,https: //fraunhoferhhi.github.io/Self-Organizing-Gaussians/

    work page doi:10.1007/978-3-031-73013-9_2 2025

  43. [43]

    arXiv preprint arXiv:2412.06234 (2024)

    Nam, S., Sun, X., Kang, G., Lee, Y., Oh, S., Park, E.: Generative densification: Learning to densify gaussians for high-fidelity generalizable 3d reconstruction. arXiv preprint arXiv:2412.06234 (2024)

    Pith/arXiv arXiv 2024

  44. [44]

    arXiv preprint arXiv:2311.18159 (2023)

    Navaneet, K., Meibodi, K.P., Koohpayegani, S.A., Pirsiavash, H.: Compact3d: Smaller and faster gaussian splatting with vector quantization. arXiv preprint arXiv:2311.18159 (2023)

    Pith/arXiv arXiv 2023

  45. [45]

    In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR)

    Niedermayr, S., Stumpfegger, J., Westermann, R.: Compressed 3d gaussian splat- ting for accelerated novel view synthesis. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR). pp. 10349–10358 (June 2024)

    2024

  46. [46]

    arXiv (2022)

    Poole, B., Jain, A., Barron, J.T., Mildenhall, B.: Dreamfusion: Text-to-3d using 2d diffusion. arXiv (2022)

    2022

  47. [47]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (2020)

    Pumarola, A., Corona, E., Pons-Moll, G., Moreno-Noguer, F.: D-NeRF: Neural Radiance Fields for Dynamic Scenes. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (2020)

    2020

  48. [48]

    In: Proceedings of the IEEE/CVF international conference on computer vision

    Ranftl, R., Bochkovskiy, A., Koltun, V.: Vision transformers for dense prediction. In: Proceedings of the IEEE/CVF international conference on computer vision. pp. 12179–12188 (2021) AdaptiveSplat 19

    2021

  49. [49]

    In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

    Rombach, R., Blattmann, A., Lorenz, D., Esser, P., Ommer, B.: High-resolution image synthesis with latent diffusion models. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 10684–10695 (2022)

    2022

  50. [50]

    In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) (June 2016)

    Schonberger, J.L., Frahm, J.M.: Structure-from-motion revisited. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) (June 2016)

    2016

  51. [51]

    Smart, B., Zheng, C., Laina, I., Prisacariu, V.A.: Splatt3r: Zero-shot gaussian splatting from uncalibrated image pairs (2024),https://arxiv.org/abs/2408. 13912

    2024

  52. [52]

    In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) (2021)

    Srinivasan, P.P., Deng, B., Zhang, X., Tancik, M., Mildenhall, B., Barron, J.T.: Nerv: Neural reflectance and visibility fields for relighting and view synthesis. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) (2021)

    2021

  53. [53]

    In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) (2024)

    Szymanowicz, S., Rupprecht, C., Vedaldi, A.: Splatter image: Ultra-fast single-view 3d reconstruction. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) (2024)

    2024

  54. [54]

    In: The Thirteenth International Conference on Learning Representations (2025), https://openreview.net/forum?id=SBzIbJojs8

    Tang, S., Ye, W., Ye, P., Lin, W., Zhou, Y., Chen, T., Ouyang, W.: Hisplat: Hierarchical 3d gaussian splatting for generalizable sparse-view reconstruction. In: The Thirteenth International Conference on Learning Representations (2025), https://openreview.net/forum?id=SBzIbJojs8

    2025

  55. [55]

    arXiv preprint arXiv:2411.12592 (2024)

    Tang, Y., Guo, Y., Li, D., Peng, C.: Spars3r: Semantic prior alignment and regu- larization for sparse 3d reconstruction. arXiv preprint arXiv:2411.12592 (2024)

    Pith/arXiv arXiv 2024

  56. [56]

    In: LBMV07 Pre- sentations

    Torralba, A., Oliva, A.: Statistics of natural image categories. In: LBMV07 Pre- sentations. Carnegie Mellon (2007)

    2007

  57. [57]

    In: Proceedings of the IEEE/CVF Confer- ence on Computer Vision and Pattern Recognition (2025)

    Wang, J., Chen, M., Karaev, N., Vedaldi, A., Rupprecht, C., Novotny, D.: Vggt: Visual geometry grounded transformer. In: Proceedings of the IEEE/CVF Confer- ence on Computer Vision and Pattern Recognition (2025)

    2025

  58. [58]

    In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR)

    Wang, S., Leroy, V., Cabon, Y., Chidlovskii, B., Revaud, J.: Dust3r: Geometric 3d vision made easy. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR). pp. 20697–20709 (June 2024)

    2024

  59. [59]

    arXiv preprint arXiv:2507.13347 (2025)

    Wang, Y., Zhou, J., Zhu, H., Chang, W., Zhou, Y., Li, Z., Chen, J., Pang, J., Shen, C., He, T.:π 3: Permutation-equivariant visual geometry learning. arXiv preprint arXiv:2507.13347 (2025)

    Pith/arXiv arXiv 2025

  60. [60]

    arXiv preprint arXiv:2501.06714 (2025)

    Wang, Y., Wu, Q., Xu, D.: F3d-gaus: Feed-forward 3d-aware generation on ima- genet with cycle-aggregative gaussian splatting. arXiv preprint arXiv:2501.06714 (2025)

    Pith/arXiv arXiv 2025

  61. [61]

    IEEE transactions on image processing 13(4), 600–612 (2004)

    Wang, Z., Bovik, A.C., Sheikh, H.R., Simoncelli, E.P.: Image quality assessment: from error visibility to structural similarity. IEEE transactions on image processing 13(4), 600–612 (2004)

    2004

  62. [62]

    In: arXiv (2024)

    Wewer, C., Raj, K., Ilg, E., Schiele, B., Lenssen, J.E.: latentsplat: Autoencoding variational gaussians for fast generalizable 3d reconstruction. In: arXiv (2024)

    2024

  63. [63]

    Proceedings of the European Conference on Computer Vision (ECCV) (2024)

    Wu, J., Bian, J.W., Li, X., Wang, G., Reid, I., Torr, P., Prisacariu, V.: GaussCtrl: Multi-View Consistent Text-Driven 3D Gaussian Splatting Editing. Proceedings of the European Conference on Computer Vision (ECCV) (2024)

    2024

  64. [64]

    arXiv:2411.18613 (2024)

    Wu, R., Gao, R., Poole, B., Trevithick, A., Zheng, C., Barron, J.T., Holynski, A.: CAT4D: Create Anything in 4D with Multi-View Video Diffusion Models. arXiv:2411.18613 (2024)

    Pith/arXiv arXiv 2024

  65. [65]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

    Wu, R., Mildenhall, B., Henzler, P., Park, K., Gao, R., Watson, D., Srinivasan, P.P., Verbin, D., Barron, J.T., Poole, B., et al.: Reconfusion: 3d reconstruction with diffusion priors. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 21551–21561 (2024) 20 Singhal et al

    2024

  66. [66]

    In: Proceedings of the IEEE/CVF International Conference on Computer Vision

    Xu, M., Zhan, F., Zhang, J., Yu, Y., Zhang, X., Theobalt, C., Shao, L., Lu, S.: Wavenerf: Wavelet-based generalizable neural radiance fields. In: Proceedings of the IEEE/CVF International Conference on Computer Vision. pp. 18195–18204 (2023)

    2023

  67. [67]

    arxiv: 2403.14621 (2024)

    Xu, Y., Shi, Z., Yifan, W., Peng, S., Yang, C., Shen, Y., Gordon, W.: Grm: Large gaussian reconstruction model for efficient 3d reconstruction and generation. arxiv: 2403.14621 (2024)

    Pith/arXiv arXiv 2024

  68. [68]

    ACM Transactions on Graphics43(6) (December 2024)

    Yang, C., Li, S., Fang, J., Liang, R., Xie, L., Zhang, X., Shen, W., Tian, Q.: Gaus- sianobject: High-quality 3d object reconstruction from four views with gaussian splatting. ACM Transactions on Graphics43(6) (December 2024)

    2024

  69. [69]

    arXiv preprint (2024)

    Yao, Y., Zeng, Z., Gu, C., Zhu, X., Zhang, L.: Reflective gaussian splatting. arXiv preprint (2024)

    2024

  70. [70]

    arXiv preprint arXiv:2410.24207 (2024)

    Ye, B., Liu, S., Xu, H., Xueting, L., Pollefeys, M., Yang, M.H., Songyou, P.: No pose, no problem: Surprisingly simple 3d gaussian splats from sparse unposed im- ages. arXiv preprint arXiv:2410.24207 (2024)

    Pith/arXiv arXiv 2024

  71. [71]

    In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) (2021)

    Yu, A., Ye, V., Tancik, M., Kanazawa, A.: pixelNeRF: Neural radiance fields from one or few images. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) (2021)

    2021

  72. [72]

    In: Proceedings of the European Conference on Computer Vision (ECCV) (2024)

    Yu, J.J., Aumentado-Armstrong, T., Forghani, F., Derpanis, K.G., Brubaker, M.A.: Polyoculus: Simultaneous multi-view image-based novel view synthesis. In: Proceedings of the European Conference on Computer Vision (ECCV) (2024)

    2024

  73. [73]

    arXiv preprint arxiv:2410.03825 (2024)

    Zhang, J., Herrmann, C., Hur, J., Jampani, V., Darrell, T., Cole, F., Sun, D., Yang, M.H.: Monst3r: A simple approach for estimating geometry in the presence of motion. arXiv preprint arxiv:2410.03825 (2024)

    Pith/arXiv arXiv 2024

  74. [74]

    In: Proceedings of the IEEE conference on computer vision and pattern recognition

    Zhang, R., Isola, P., Efros, A.A., Shechtman, E., Wang, O.: The unreasonable effectiveness of deep features as a perceptual metric. In: Proceedings of the IEEE conference on computer vision and pattern recognition. pp. 586–595 (2018)

    2018

  75. [75]

    Advances in Neural Information Processing Systems37, 50361–50380 (2024)

    Zhang, S., Fei, X., Liu, F., Song, H., Duan, Y.: Gaussian graph network: Learn- ing efficient and generalizable gaussian representations from multi-view images. Advances in Neural Information Processing Systems37, 50361–50380 (2024)

    2024

  76. [76]

    In: Advances in Neural Information Processing Systems (NeurIPS) (2024)

    Zhou, J., Zhang, W., Liu, Y.S.: Diffgs: Functional gaussian splatting diffusion. In: Advances in Neural Information Processing Systems (NeurIPS) (2024)

    2024

  77. [77]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

    Zhou,S.,Chang,H.,Jiang,S.,Fan,Z.,Zhu,Z.,Xu,D.,Chari,P.,You,S.,Wang,Z., Kadambi, A.: Feature 3dgs: Supercharging 3d gaussian splatting to enable distilled feature fields. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 21676–21685 (2024)

    2024

  78. [78]

    Zhou, S., Fan, Z., Xu, D., Chang, H., Chari, P., Bharadwaj, T., You, S., Wang, Z., Kadambi, A.: Dreamscene360: Unconstrained text-to-3d scene generation with panoramicgaussiansplatting.In:ProceedingsoftheEuropeanConferenceonCom- puter Vision (ECCV). pp. 324–342. Springer (2024)

    2024

  79. [79]

    In: SIGGRAPH (2018)

    Zhou, T., Tucker, R., Flynn, J., Fyffe, G., Snavely, N.: Stereo magnification: Learn- ing view synthesis using multiplane images. In: SIGGRAPH (2018)

    2018

  80. [80]

    Ziwen, C., Tan, H., Zhang, K., Bi, S., Luan, F., Hong, Y., Fuxin, L., Xu, Z.: Long- lrm: Long-sequence large reconstruction model for wide-coverage gaussian splats. In: Proceedings of the IEEE/CVF International Conference on Computer Vision (2025) AdaptiveSplat A Supplementary Material: AdaptiveSplat: Texture Aware Controllable 3D Gaussian Allocation for ...

    arXiv 2025