Pith. sign in

REVIEW 4 major objections 6 minor 96 references

Seeing World Dynamics in a Nutshell

T0 review · 4 major / 6 minor · reviewed 2026-08-09 · deepseek-v4-flash

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

Pith's one-line read Monocular video becomes dynamic 3D Gaussians in a single pass.

desk verdict Fast, clever feed-forward video-to-Gaussian method, but the 'true 3D' claim is not yet backed by quantitative geometry tests. read the letter →

arxiv 2502.03465 v2 pith:P37A2BW4 submitted 2025-02-05 cs.CV cs.AIcs.GRcs.MM

Qiuhong Shen , Xuanyu Yi , Mingbao Lin , Hanwang Zhang , Shuicheng Yan , Xinchao Wang This is my paper · ORCID
classification cs.CVcs.AIcs.GRcs.MM
keywords dynamicGaussiansplattingmonocularvideorepresentationfeed-forwardreconstructionspatial-temporalaligneddepthregularizationopticalflow
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

This paper claims that a casually captured monocular video can be lifted, in a single feed-forward pass, into a dynamic 3D Gaussian representation, without per-scene optimization or camera-pose estimation. The proposed model, NutWorld, outputs a spatial-temporal aligned Gaussian (STAG) cloud in a fixed orthographic canonical volume, so each primitive is tied to a pixel and timestamp and carries a continuous deformation field. Calibrated depth and optical-flow priors regularize the ill-posed monocular lift, resolving depth collapse and motion-appearance ambiguity. The payoff would be video representations that reconstruct and render in seconds rather than tens of minutes, while supporting frame interpolation, segmentation, editing, consistent depth, and limited novel views. On 50 test clips drawn from two large video datasets, NutWorld reports PSNR 29.18, SSIM 0.9015, and LPIPS 0.1415, with 1.8-second inference for a 90-frame video and 450 FPS rendering.

What carries the argument

The central object is the spatial-temporal aligned Gaussian (STAG), a structured dynamic Gaussian representation in an orthographic canonical camera space. Each Gaussian is bound to a pixel coordinate (u, v), a depth d predicted by bin regression, and a reference timestamp; a temporal slicing indicator suppresses the deformation field at the reference frame to preserve spatial alignment, and a small MLP conditioned on a sinusoidal time encoding produces the continuous deformation field $\mu(t)$ that interpolates frames at arbitrary timestamps. The network pipeline concatenates RGB frames with a depth prior channel, processes the tokens with self-attention and hierarchical upsampling, and decodes Gaussian attributes (position, opacity, scale, rotation, color) from the resulting feature map. Two regularizers carry the argument: a scale-and-shift invariant depth loss that aligns rendered depth with the pseudo-depth prior, and a global flow loss that compares projected Gaussian trajectories against cumulative optical-flow paths, both with outlier filtering.

What would settle it

Render held-out viewpoints with increasing angular or translational offsets and compare against ground-truth multi-view capture; if image quality collapses at modest parallax or recovered depth disagrees with known scene geometry, STAG is a layered 2.5D reconstruction rather than a true 3D representation.

Watch

Extended reading notes

Core claim

NutWorld's central claim is that a monocular video can be represented directly in its intrinsic 3D form by a structured spatial-temporal aligned Gaussian point cloud, predicted from six sparsely sampled frames by a transformer in a single forward pass. The key to making this feed-forward prediction stable is the spatial-temporal alignment: each Gaussian is anchored to a specific pixel location and timestamp in a fixed orthographic canonical volume, so its static position is $\mu = (u + \Delta x, v + \Delta y, d)$, with bounded offsets and a depth coordinate, while a continuous deformation field $\mu(t)$ carries it through time. Calibrated depth and optical-flow priors provide the geometric and motion supervision that disambiguates the monocular input. The paper supports the claim with reconstruction metrics that exceed optimization-based baselines and with ablations showing that loosening the STAG constraint degrades PSNR by about 10 dB, while removing the flow or depth losses also causes clear drops.

Load-bearing premise

The load-bearing premise is that a fixed orthographic camera plus a monocular depth prior provides enough geometric signal for the resulting STAG to be treated as genuine 3D structure rather than a 2.5D image mosaic.

Editorial extensions

If this is right

  • NutWorld reconstructs a 90-frame video in 1.8 seconds and renders at 450 FPS, roughly a 1000x speedup over the optimization-based baselines it is compared with.
  • Because every Gaussian is tied to a pixel and timestamp, segment-based inference can extend the representation to arbitrarily long videos while preserving trajectory continuity through overlapping frames.
  • The continuous deformation field supports rendering intermediate frames at arbitrary timestamps, giving frame interpolation without per-frame generation.
  • The explicit, maskable Gaussians enable video object segmentation by propagating an initial mask through the deformation field, with no task-specific training.
  • The ablations show the representation is not merely cosmetic: removing the flow loss drops PSNR from 29.18 to 26.39, removing the depth loss drops it to 28.15, and removing the STAG structure drops it to 19.58.

Reading between the lines

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

  • Editorial inference: because NutWorld uses a fixed orthographic camera, its '3D' is a single-view layered geometry; a strict test would be whether rendering quality survives larger camera translation than the small dolly or zoom effects shown.
  • Editorial inference: the method's ceiling is set by the external depth and flow priors; corrupting those priors at test time would quantify how much geometry and motion come from the network rather than from the priors.
  • Editorial inference: the pixel-anchored structure suggests a natural extension to video generation, where the deformation field could be sampled or predicted from text rather than learned from the input video alone.
  • Editorial inference: distilling semantic features into each Gaussian, which the authors name as future work, would let segmentation and editing run directly on STAG without extra inference-time models.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 6 minor

Summary. The paper proposes NutWorld, a feed-forward framework that maps a monocular video into a set of spatially and temporally aligned 3D Gaussians (STAG) in a fixed orthographic canonical space. A transformer encoder with hierarchical upsampling predicts per-pixel Gaussian attributes and deformation fields, regularized by off-the-shelf video depth and optical-flow priors. The authors report reconstruction PSNR/SSIM/LPIPS on 50 clips from MiraData and RealEstate10K, a 1.8-second inference time for a 90-frame video, 450 FPS rendering, and qualitative demonstrations of segmentation, editing, interpolation, consistent depth, and novel-view synthesis. The main claimed contribution is an optimization-free dynamic 3D representation for casually captured monocular videos.

Significance. If the geometric component is validated, this is a useful advance: single-forward-pass dynamic Gaussian reconstruction with real-time rendering would be relevant to video representation and downstream video editing tasks. The STAG structure is well motivated, the ablation study shows that the depth and flow losses matter, and the reported speed improvements are substantial. The paper currently provides strong evidence for a high-quality video codec under a fixed orthographic camera, but the evidence that this codec is a genuine recovered dynamic 3D representation is incomplete; the missing quantitative geometry checks are the main gap.

major comments (4)
  1. [Section 4.1 and Section 5.2] The reconstruction metrics in Table 1 are computed under the same fixed orthographic camera used to define the canonical space in Section 4.1. With pixel-aligned Gaussian centers and per-frame colors, a depth-ordered 2.5D mosaic can plausibly match the reported PSNR/SSIM/LPIPS numbers without representing a true dynamic 3D volume. The authors should provide a quantitative geometry check, for example held-out novel-view synthesis using the COLMAP poses that are available for RealEstate10K (which the paper explicitly declines to use in the Section 5.1 footnote and Appendix 2), or depth error against a reliable reference, and should compare against a 2.5D baseline under the same protocol.
  2. [Section 5.2 and Appendix 2] The 50 test clips are drawn from MiraData and RealEstate10K, the same datasets used for training, with a random 95:5 split. This gives NutWorld an in-distribution advantage over optimization-based baselines that are evaluated on the same clips but do not share the training distribution. The authors should report per-dataset numbers, add cross-dataset evaluation or a held-out-domain clip set, and state the train/test overlap explicitly in the main text rather than only in the appendix.
  3. [Section 5.3 and Figure 12] Novel view synthesis and consistent depth are presented as evidence of spatial 3D structure, but the evidence is purely qualitative. This is load-bearing for the central claim: a 2.5D representation can also produce plausible depth visualizations and small-baseline view changes. The authors should quantify novel-view accuracy on RealEstate10K, where camera poses are available, or otherwise temper the claim that STAG is a genuine 3D representation.
  4. [Equations (6) and (8)] The depth and flow losses provide useful regularization, but they supervise against external priors rather than ground-truth scene geometry. The qualitative ablation in Figure 7(a) shows depth collapse without the depth loss, which is informative, yet it does not establish the absolute correctness or temporal consistency of the recovered depths. Since Appendix 4 candidly acknowledges dependence on these priors, the paper should either add a quantitative geometry evaluation or carefully rephrase the abstract's claim that NutWorld 'transforms monocular videos into dynamic 3D Gaussian representations' to reflect what is actually measured.
minor comments (6)
  1. [Equation (6)] The denominator written as 'X M' is unclear; it should be the sum of the mask values over all pixels, and the summation index in Ldepth should be made explicit.
  2. [Section 4.2 and Table 3] The main text states that the deformation output is bounded by tanh within [−b,b]^3, while Table 3 lists the dynamic position as unbounded. This inconsistency should be resolved because it affects reproducibility.
  3. [Section 4.2] The claim that positional embeddings are unnecessary because 'depth coordinates inherently encode spatial information' is imprecise: the concatenated depth channel is a scalar per pixel and does not by itself encode the 2D grid coordinate. Please clarify how spatial position is injected.
  4. [Section 5.2] The sentence 'all methods incorporate the confined canonical space, depth and flow supervision' needs more detail: it should state how each optimization-based baseline was adapted to this protocol, since the comparison may otherwise be difficult to reproduce.
  5. [Table 4 and Table 3] There are typos in the activation names: 'Sof tplus' should be 'Softplus' and 'N ormalize' should be 'Normalize'.
  6. [Appendix 2, Segment-based inference] The statement 'Through quantitative comparison shown in Table.1 in the manuscript' is confusing because Table 1 does not directly report segment-based inference. Please cite the specific experiment or figure that supports this claim.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: self-citations are peripheral, and the core feed-forward reconstruction claim is empirically tested against external priors and optimization-based baselines.

full rationale

NutWorld is an empirical systems paper rather than a derivation chain, and its central quantitative claim is reconstruction quality on held-out video clips. The depth and flow priors come from external foundation models (Video Depth Anything and UniMatch), and they are used both as input channels and as regularizers; the paper is transparent about this and does not present the regularized depth or flow outputs as an independent prediction. The RGB reconstruction metrics in Table 1 are standard image-similarity measures on test clips, so the main result does not reduce by construction to the training losses. The cited works by the same authors (Gamba, MVGamba, GFlow, FlashSplat) appear in related-work context or as tools for downstream segmentation; none is invoked as a uniqueness theorem or as the justification for the STAG representation. The reviewer concern that the fixed orthographic camera and per-pixel alignment make STAG close to a 2.5D video codec, and that no quantitative novel-view or depth-accuracy evaluation is reported, is a legitimate verification gap about whether the representation is genuinely 3D, but it is not a circularity step: the paper does not define its 3D claim in terms of the metrics it reports. Overall, the derivation is self-contained with respect to external benchmarks, and no fitted parameter is renamed as a prediction.

Assumptions & free parameters 5 free parameters · 4 assumptions · 1 invented entities

The central claim is empirical: a transformer predicts STAGs from videos, supervised by RGB, depth, and flow. The load-bearing axioms are modeling choices (orthographic canonical space) and external prior reliability. Free parameters are training hyperparameters; the most influential is the number of upsampler blocks, which changes PSNR by about 2 dB.

free parameters (5)
  • loss weights lambda_MSE, lambda_depth, lambda_flow = 0.5, 0.001, 0.2
    Set manually in Section 4.4; the training objective and reconstruction quality depend on their balance.
  • number of upsampler blocks n = 2 (main) / 3 (variant)
    Chosen by hand; PSNR rises from 29.18 to 31.15 when n goes from 2 to 3 (Table 2).
  • depth discretization bins = 20
    Appendix Table 3; hand-chosen discretization of the z-axis in [0,1].
  • deformation amplitude bound b = not specified
    Section 4.2 Eq. 5 says tanh bounds the deformation to [-b,b]^3, but the value of b is not given.
  • outlier thresholds for depth and flow = top 10% depth values, top 20% flow magnitudes
    Defined in Section 4.3 and the Appendix Flow Prior Calibration; post-hoc filtering of pseudo-label noise.
assumptions (4)
  • domain assumption A monocular video can be interpreted as a projection of a dynamic 3D world represented by Gaussian primitives in an orthographic canonical volume.
    Section 4.1 states: 'we employ an orthographic camera coordinate system to interpret the input as a quasi-3D canonical volume.' This is the modeling choice that enables pose-free prediction.
  • domain assumption Pretrained depth (Video Depth Anything) and optical flow (UniMatch) priors provide reliable supervision for spatial and temporal structure.
    Section 4.3 uses these as calibration losses; the Limitations section acknowledges that performance 'is inherently dependent on external models'.
  • domain assumption The training datasets (MiraData, RealEstate10K) are representative of casually captured monocular videos.
    Section 5.1; experiments are only on these datasets, with no in-the-wild evaluation despite the claim of casual videos.
  • domain assumption The orthographic camera model and depth-in-[0,1] parameterization eliminate scale ambiguity without losing essential structure.
    Section 4.1; this assumption makes feed-forward prediction tractable but limits true perspective and parallax recovery.
invented entities (1)
  • Spatial-Temporal Aligned Gaussian (STAG)
    purpose: Pixel-aligned dynamic Gaussian representation with per-pixel deformation in a canonical orthographic space; the learned building block of the video representation.
    A model construct, not a physical entity. Its geometric validity is not independently verified: there is no quantitative novel-view or depth accuracy evaluation, only qualitative figures.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Seeing World Dynamics in a Nutshell." pith.science (2026). https://pith.science/paper/P37A2BW4

@misc{pith2026250203465,
  author       = {Pith},
  title        = {Pith review of: Seeing World Dynamics in a Nutshell},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/P37A2BW4}},
  note         = {Machine review of arXiv:2502.03465}
}
read the original abstract

We consider the problem of efficiently representing casually captured monocular videos in a spatially- and temporally-coherent manner. While existing approaches predominantly rely on 2D/2.5D techniques treating videos as collections of spatiotemporal pixels, they struggle with complex motions, occlusions, and geometric consistency due to absence of temporal coherence and explicit 3D structure. Drawing inspiration from monocular video as a projection of the dynamic 3D world, we explore representing videos in their intrinsic 3D form through continuous flows of Gaussian primitives in space-time. In this paper, we propose NutWorld, a novel framework that efficiently transforms monocular videos into dynamic 3D Gaussian representations in a single forward pass. At its core, NutWorld introduces a structured spatial-temporal aligned Gaussian (STAG) representation, enabling optimization-free scene modeling with effective depth and flow regularization. Through comprehensive experiments, we demonstrate that NutWorld achieves high-fidelity video reconstruction quality while enabling various downstream applications in real-time. Demos and code will be available at https://github.com/Nut-World/NutWorld.

Figures

Figures reproduced from arXiv: 2502.03465 by the authors.

Figure 1
Figure 1. We introduce NutWorld, a feed-forward framework representing casual monocular videos parameterized by spatial-temporal aligned Gaussian (STAG), which empowers various video downstream processing tasks. Abstract We consider the problem of efficiently representing casually captured monocular videos in a spatially- and temporally￾coherent manner. While existing approaches predominantly rely on 2D/2.5D techniques treati… view at source ↗
Figure 2
Figure 2. The illustration of STAG to represent dynamic scenes. [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Overview of NutWorld. We directly predict STAG in a canonical space from sparse input frames via a transformer-based reconstruction model, where calibrated depth and flow priors are leveraged to avoid depth ambiguity and motion uncertainty. nates the need for explicit camera estimation, enabling joint modeling of both camera and object motion. We detail the orthographic rasterization pipeline in the Appendix. Struct… view at source ↗
Figures from the paper (10 more)
Figure 4
Figure 4. Figure 4: Qualitative comparison of video reconstruction using our NutWorld and other optimization-based methods. [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 5
Figure 5. Figure 5: Qualitative results in various downstream tasks, including video segmentation, editing, frame interpolation and consistent depth [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 6
Figure 6. Figure 6: Visualization of Gaussian trajectories. Trajectories of [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]
Figure 7
Figure 7. Figure 7: Qualitative ablation on 2D prior regularization. [PITH_FULL_IMAGE:figures/full_fig_p008_7.png]
Figure 8
Figure 8. Figure 8: More qualitative results on video reconstruction. [PITH_FULL_IMAGE:figures/full_fig_p013_8.png]
Figure 9
Figure 9. Figure 9: More qualitative results on video editing. [PITH_FULL_IMAGE:figures/full_fig_p014_9.png]
Figure 10
Figure 10. Figure 10: More qualitative results on consistent depth prediction. [PITH_FULL_IMAGE:figures/full_fig_p014_10.png]
Figure 11
Figure 11. Figure 11: More qualitative results on frame interpolation. Note that [PITH_FULL_IMAGE:figures/full_fig_p015_11.png]
Figure 12
Figure 12. Figure 12: Qualitative results on novel view synthesis. [PITH_FULL_IMAGE:figures/full_fig_p016_12.png]
Figure 13
Figure 13. Figure 13: Qualitative results on video object segmentation. [PITH_FULL_IMAGE:figures/full_fig_p016_13.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

96 extracted references · 43 canonical work pages

  1. [1]

    Meissonic: Revitalizing masked gen- erative transformers for efficient high-resolution text- to-image synthesis

    Jinbin Bai, Tian Ye, Wei Chow, Enxin Song, Qing- Guo Chen, Xiangtai Li, Zhen Dong, Lei Zhu, and Shuicheng Yan. Meissonic: Revitalizing masked gen- erative transformers for efficient high-resolution text- to-image synthesis. arXiv preprint arXiv:2410.08261,

    arXiv

  2. [2]

    Vidu: a highly consis- tent, dynamic and skilled text-to-video generator with diffusion models

    Fan Bao, Chendong Xiang, Gang Yue, Guande He, Hongzhou Zhu, Kaiwen Zheng, Min Zhao, Shilong Liu, Yaole Wang, and Jun Zhu. Vidu: a highly consis- tent, dynamic and skilled text-to-video generator with diffusion models. arXiv preprint arXiv:2405.04233 ,

    arXiv

  3. [3]

    Zoedepth: Zero-shot transfer by combining relative and metric depth.arXiv preprint arXiv:2302.12288, 2023

    Shariq Farooq Bhat, Reiner Birkl, Diana Wofk, Peter Wonka, and Matthias M ¨uller. Zoedepth: Zero-shot transfer by combining relative and metric depth.arXiv preprint arXiv:2302.12288, 2023. 5

    arXiv 2023

  4. [4]

    Nope-nerf: Optimising neural radiance field with no pose prior

    Wenjing Bian, Zirui Wang, Kejie Li, Jia-Wang Bian, and Victor Adrian Prisacariu. Nope-nerf: Optimising neural radiance field with no pose prior. In Proceed- ings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 4160–4169, 2023. 2

    2023

  5. [5]

    The essential guide to video process- ing

    Alan C Bovik. The essential guide to video process- ing. Academic Press, 2009. 2

    2009

  6. [6]

    Pix2video: Video editing using image diffusion

    Duygu Ceylan, Chun-Hao P Huang, and Niloy J Mitra. Pix2video: Video editing using image diffusion. In Proceedings of the IEEE/CVF International Confer- ence on Computer Vision, pages 23206–23217, 2023. 2

    2023

  7. [7]

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

    David Charatan, Sizhe Lester Li, Andrea Tagliasacchi, and Vincent Sitzmann. pixelsplat: 3d gaussian splats from image pairs for scalable generalizable 3d recon- struction. In CVPR, pages 19457–19467, 2024. 3, 4, 5, 6, 1, 2

    2024

  8. [8]

    Video depth anything: Consistent depth estimation for super- long videos

    Sili Chen, Hengkai Guo, Shengnan Zhu, Feihu Zhang, Zilong Huang, Jiashi Feng, and Bingyi Kang. Video depth anything: Consistent depth estimation for super- long videos. arXiv preprint arXiv:2501.12375, 2025. 2, 4, 5

    arXiv 2025

Show all 96 references
  1. [9]

    Learning continuous image representation with local implicit image function

    Yinbo Chen, Sifei Liu, and Xiaolong Wang. Learning continuous image representation with local implicit image function. In Proceedings of the IEEE/CVF con- ference on computer vision and pattern recognition , pages 8628–8638, 2021. 3

    2021

  2. [10]

    Mvsplat: Efficient 3d gaus- sian splatting from sparse multi-view images

    Yuedong Chen, Haofei Xu, Chuanxia Zheng, Bo- han Zhuang, Marc Pollefeys, Andreas Geiger, Tat-Jen Cham, and Jianfei Cai. Mvsplat: Efficient 3d gaus- sian splatting from sparse multi-view images. InEuro- pean Conference on Computer Vision, pages 370–386. Springer, 2025. 3, 6, 2

    2025

  3. [11]

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

    Jaeyoung Chung, Jeongtaek Oh, and Kyoung Mu Lee. Depth-regularized optimization for 3d gaussian splatting in few-shot images. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pat- tern Recognition, pages 811–820, 2024. 2

    2024

  4. [12]

    Flashattention-2: Faster attention with bet- ter parallelism and work partitioning

    Tri Dao. Flashattention-2: Faster attention with bet- ter parallelism and work partitioning. arXiv preprint arXiv:2307.08691, 2023. 7

    2023 arXiv

  5. [13]

    Transformers are SSMs: Generalized models and efficient algorithms through structured state space duality

    Tri Dao and Albert Gu. Transformers are SSMs: Generalized models and efficient algorithms through structured state space duality. In International Con- ference on Machine Learning (ICML), 2024. 3

    2024

  6. [14]

    Flashattention: Fast and memory- efficient exact attention with io-awareness

    Tri Dao, Dan Fu, Stefano Ermon, Atri Rudra, and Christopher R ´e. Flashattention: Fast and memory- efficient exact attention with io-awareness. Advances in Neural Information Processing Systems, 35:16344– 16359, 2022. 7

    2022

  7. [15]

    Video frame interpolation: A comprehensive survey

    Jiong Dong, Kaoru Ota, and Mianxiong Dong. Video frame interpolation: A comprehensive survey. ACM Transactions on Multimedia Computing, Communica- tions and Applications, 19(2s):1–31, 2023. 2

    2023

  8. [16]

    An image is worth 16x16 words: Transformers for image recognition at scale

    Alexey Dosovitskiy. An image is worth 16x16 words: Transformers for image recognition at scale. arXiv preprint arXiv:2010.11929, 2020. 5

    2010 arXiv

  9. [17]

    4d gaussian splatting: Towards efficient novel view synthesis for dynamic scenes

    Yuanxing Duan, Fangyin Wei, Qiyu Dai, Yuhang He, Wenzheng Chen, and Baoquan Chen. 4d gaussian splatting: Towards efficient novel view synthesis for dynamic scenes. arXiv preprint arXiv:2402.03307 ,

    arXiv

  10. [18]

    Slowfast networks for video recog- nition

    Christoph Feichtenhofer, Haoqi Fan, Jitendra Malik, and Kaiming He. Slowfast networks for video recog- nition. In Proceedings of the IEEE/CVF interna- tional conference on computer vision , pages 6202– 6211, 2019. 2

    2019

  11. [19]

    Guiding instruction-based image editing via multimodal large language models

    Tsu-Jui Fu, Wenze Hu, Xianzhi Du, William Yang Wang, Yinfei Yang, and Zhe Gan. Guiding instruction-based image editing via multimodal large language models. arXiv preprint arXiv:2309.17102 ,

    arXiv

  12. [20]

    Efros, and Xiaolong Wang

    Yang Fu, Sifei Liu, Amey Kulkarni, Jan Kautz, Alexei A. Efros, and Xiaolong Wang. Colmap-free 3d gaussian splatting. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recogni- tion (CVPR), pages 20796–20805, 2024. 2

    2024

  13. [21]

    Vista: A generalizable driving world model with high fidelity and versatile controllability

    Shenyuan Gao, Jiazhi Yang, Li Chen, Kashyap Chitta, Yihang Qiu, Andreas Geiger, Jun Zhang, and Hongyang Li. Vista: A generalizable driving world model with high fidelity and versatile controllability. arXiv preprint arXiv:2405.17398, 2024. 2

    2024 arXiv

  14. [22]

    Epipolar transformers

    Yihui He, Rui Yan, Katerina Fragkiadaki, and Shoou- I Yu. Epipolar transformers. In Proceedings of the ieee/cvf conference on computer vision and pattern recognition, pages 7779–7788, 2020. 3

    2020

  15. [23]

    8 Semantic anything in 3d gaussians

    Xu Hu, Yuxi Wang, Lue Fan, Junsong Fan, Jun- ran Peng, Zhen Lei, Qing Li, and Zhaoxiang Zhang. 8 Semantic anything in 3d gaussians. arXiv preprint arXiv:2401.17857, 2024. 8

    2024 arXiv

  16. [24]

    2d gaussian splatting for geometrically accurate radiance fields

    Binbin Huang, Zehao Yu, Anpei Chen, Andreas Geiger, and Shenghua Gao. 2d gaussian splatting for geometrically accurate radiance fields. In ACM SIG- GRAPH 2024 Conference Papers, pages 1–11, 2024. 2

    2024

  17. [25]

    V oxposer: Composable 3d value maps for robotic manipulation with language models

    Wenlong Huang, Chen Wang, Ruohan Zhang, Yunzhu Li, Jiajun Wu, and Li Fei-Fei. V oxposer: Composable 3d value maps for robotic manipulation with language models. arXiv preprint arXiv:2307.05973, 2023. 2

    2023 arXiv

  18. [26]

    Rekep: Spatio-temporal rea- soning of relational keypoint constraints for robotic manipulation

    Wenlong Huang, Chen Wang, Yunzhu Li, Ruohan Zhang, and Li Fei-Fei. Rekep: Spatio-temporal rea- soning of relational keypoint constraints for robotic manipulation. arXiv preprint arXiv:2409.01652 ,

    arXiv

  19. [27]

    Miradata: A large-scale video dataset with long durations and structured captions, 2024

    Xuan Ju, Yiming Gao, Zhaoyang Zhang, Ziyang Yuan, Xintao Wang, Ailing Zeng, Yu Xiong, Qiang Xu, and Ying Shan. Miradata: A large-scale video dataset with long durations and structured captions, 2024. 2, 6

    2024

  20. [28]

    Cotracker: It is better to track together

    Nikita Karaev, Ignacio Rocco, Benjamin Graham, Na- talia Neverova, Andrea Vedaldi, and Christian Rup- precht. Cotracker: It is better to track together. arXiv preprint arXiv:2307.07635, 2023. 2

    2023 arXiv

  21. [29]

    A compact dynamic 3d gaussian representation for real-time dynamic view synthesis

    Kai Katsumata, Duc Minh V o, and Hideki Nakayama. A compact dynamic 3d gaussian representation for real-time dynamic view synthesis. In European Con- ference on Computer Vision, pages 394–412. Springer,

  22. [30]

    3d gaussian splatting for real-time radiance field rendering

    Bernhard Kerbl, Georgios Kopanas, Thomas Leimk¨uhler, and George Drettakis. 3d gaussian splatting for real-time radiance field rendering. ACM Trans. Graph., 42(4):139–1, 2023. 2, 3

    2023

  23. [31]

    Segment anything

    Alexander Kirillov, Eric Mintun, Nikhila Ravi, Hanzi Mao, Chloe Rolland, Laura Gustafson, Tete Xiao, Spencer Whitehead, Alexander C Berg, Wan-Yen Lo, et al. Segment anything. In Proceedings of the IEEE/CVF International Conference on Computer Vi- sion, pages 4015–4026, 2023. 8

    2023

  24. [32]

    Videopoet: A large language model for zero-shot video generation

    Dan Kondratyuk, Lijun Yu, Xiuye Gu, Jos ´e Lezama, Jonathan Huang, Grant Schindler, Rachel Hornung, Vighnesh Birodkar, Jimmy Yan, Ming-Chang Chiu, et al. Videopoet: A large language model for zero-shot video generation. arXiv preprint arXiv:2312.14125 ,

    arXiv

  25. [33]

    Dynmf: Neural motion factorization for real-time dynamic view synthesis with 3d gaussian splatting

    Agelos Kratimenos, Jiahui Lei, and Kostas Daniilidis. Dynmf: Neural motion factorization for real-time dynamic view synthesis with 3d gaussian splatting. In European Conference on Computer Vision , pages 252–269. Springer, 2025. 2, 3

    2025

  26. [34]

    Mosca: Dynamic gaus- sian fusion from casual videos via 4d motion scaf- folds

    Jiahui Lei, Yijia Weng, Adam Harley, Leonidas Guibas, and Kostas Daniilidis. Mosca: Dynamic gaus- sian fusion from casual videos via 4d motion scaf- folds. arXiv preprint arXiv:2405.17421, 2024. 2, 3

    2024 arXiv

  27. [35]

    Grounding image matching in 3d with mast3r, 2024

    Vincent Leroy, Yohann Cabon, and Jerome Revaud. Grounding image matching in 3d with mast3r, 2024. 3

    2024

  28. [36]

    Vdg: Vision- only dynamic gaussian for driving simulation

    Hao Li, Jingfeng Li, Dingwen Zhang, Chenming Wu, Jieqi Shi, Chen Zhao, Haocheng Feng, Errui Ding, Jingdong Wang, and Junwei Han. Vdg: Vision- only dynamic gaussian for driving simulation. arXiv preprint arXiv:2406.18198, 2024. 2

    2024 arXiv

  29. [37]

    Barf: Bundle-adjusting neural ra- diance fields

    Chen-Hsuan Lin, Wei-Chiu Ma, Antonio Torralba, and Simon Lucey. Barf: Bundle-adjusting neural ra- diance fields. In Proceedings of the IEEE/CVF inter- national conference on computer vision, pages 5741– 5751, 2021. 2

    2021

  30. [38]

    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. In Proceed- ings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, ...

    2024

  31. [39]

    In- finite nature: Perpetual view generation of natural scenes from a single image

    Andrew Liu, Richard Tucker, Varun Jampani, Ameesh Makadia, Noah Snavely, and Angjoo Kanazawa. In- finite nature: Perpetual view generation of natural scenes from a single image. In Proceedings of the IEEE/CVF International Conference on Computer Vi- sion, pages 14458–14467, 2021. 3

    2021

  32. [40]

    Robust dynamic radiance fields

    Yu-Lun Liu, Chen Gao, Andreas Meuleman, Hung-Yu Tseng, Ayush Saraf, Changil Kim, Yung-Yu Chuang, Johannes Kopf, and Jia-Bin Huang. Robust dynamic radiance fields. In Proceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition, pages 13–23, 2023. 3, 6, 7

    2023

  33. [41]

    Video frame interpolation with trans- former

    Liying Lu, Ruizheng Wu, Huaijia Lin, Jiangbo Lu, and Jiaya Jia. Video frame interpolation with trans- former. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 3532–3542, 2022. 2

    2022

  34. [42]

    Dynamicfusion: Reconstruction and tracking of non-rigid scenes in real-time

    Richard A Newcombe, Dieter Fox, and Steven M Seitz. Dynamicfusion: Reconstruction and tracking of non-rigid scenes in real-time. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 343–352, 2015. 2

    2015

  35. [43]

    Open x-embodiment: Robotic learning datasets and rt-x models

    Abby O’Neill, Abdul Rehman, Abhinav Gupta, Abhi- ram Maddukuri, Abhishek Gupta, Abhishek Padalkar, Abraham Lee, Acorn Pooley, Agrim Gupta, Ajay Mandlekar, et al. Open x-embodiment: Robotic learning datasets and rt-x models. arXiv preprint arXiv:2310.08864, 2023. 2

    arXiv

  36. [44]

    Codef: Content deformation fields for temporally consistent video processing

    Hao Ouyang, Qiuyu Wang, Yuxi Xiao, Qingyan Bai, Juntao Zhang, Kecheng Zheng, Xiaowei Zhou, Qifeng Chen, and Yujun Shen. Codef: Content deformation fields for temporally consistent video processing. In 9 Proceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Re...

    2024

  37. [45]

    Global structure-from-motion revisited

    Linfei Pan, D ´aniel Bar ´ath, Marc Pollefeys, and Jo- hannes L Sch ¨onberger. Global structure-from-motion revisited. In European Conference on Computer Vi- sion (ECCV), 2024. 3

    2024

  38. [46]

    Fatezero: Fusing attentions for zero-shot text-based video editing

    Chenyang Qi, Xiaodong Cun, Yong Zhang, Chenyang Lei, Xintao Wang, Ying Shan, and Qifeng Chen. Fatezero: Fusing attentions for zero-shot text-based video editing. In Proceedings of the IEEE/CVF In- ternational Conference on Computer Vision , pages 15932–15942, 2023. 2

    2023

  39. [47]

    Langsplat: 3d language gaussian splatting

    Minghan Qin, Wanhua Li, Jiawei Zhou, Haoqian Wang, and Hanspeter Pfister. Langsplat: 3d language gaussian splatting. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recogni- tion, pages 20051–20060, 2024. 2

    2024

  40. [48]

    Learning transferable visual models from natural language supervision

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sas- try, Amanda Askell, Pamela Mishkin, Jack Clark, et al. Learning transferable visual models from natural language supervision. In International conference on machine learning, ...

    2021

  41. [49]

    Towards robust monocular depth estimation: Mixing datasets for zero- shot cross-dataset transfer

    Ren ´e Ranftl, Katrin Lasinger, David Hafner, Kon- rad Schindler, and Vladlen Koltun. Towards robust monocular depth estimation: Mixing datasets for zero- shot cross-dataset transfer. IEEE transactions on pat- tern analysis and machine intelligence , 44(3):1623– 1637, 2020. 5

    2020

  42. [50]

    L4gm: Large 4d gaussian reconstruction model

    Jiawei Ren, Kevin Xie, Ashkan Mirzaei, Hanxue Liang, Xiaohui Zeng, Karsten Kreis, Ziwei Liu, An- tonio Torralba, Sanja Fidler, Seung Wook Kim, et al. L4gm: Large 4d gaussian reconstruction model. arXiv preprint arXiv:2406.10324, 2024. 2, 8

    2024 arXiv

  43. [51]

    Structure-from-motion revisited

    Johannes L Schonberger and Jan-Michael Frahm. Structure-from-motion revisited. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 4104–4113, 2016. 2, 3

    2016

  44. [52]

    Gamba: Marry gaussian splatting with mamba for single view 3d reconstruction

    Qiuhong Shen, Zike Wu, Xuanyu Yi, Pan Zhou, Han- wang Zhang, Shuicheng Yan, and Xinchao Wang. Gamba: Marry gaussian splatting with mamba for single view 3d reconstruction. arXiv preprint arXiv:2403.18795, 2024. 3

    2024 arXiv

  45. [53]

    Flashsplat: 2d to 3d gaussian splatting segmentation solved optimally

    Qiuhong Shen, Xingyi Yang, and Xinchao Wang. Flashsplat: 2d to 3d gaussian splatting segmentation solved optimally. In European Conference on Com- puter Vision, pages 456–472. Springer, 2025. 2, 8

    2025

  46. [54]

    Real-time single image and video super-resolution using an efficient sub-pixel convolu- tional neural network

    Wenzhe Shi, Jose Caballero, Ferenc Husz ´ar, Johannes Totz, Andrew P Aitken, Rob Bishop, Daniel Rueckert, and Zehan Wang. Real-time single image and video super-resolution using an efficient sub-pixel convolu- tional neural network. In Proceedings of the IEEE conference on com...

    2016

  47. [55]

    Implicit neural representations with periodic activation functions.Ad- vances in neural information processing systems , 33: 7462–7473, 2020

    Vincent Sitzmann, Julien Martel, Alexander Bergman, David Lindell, and Gordon Wetzstein. Implicit neural representations with periodic activation functions.Ad- vances in neural information processing systems , 33: 7462–7473, 2020. 2, 3

    2020

  48. [56]

    Low- memory neural network training: A technical report

    Nimit S Sohoni, Christopher R Aberger, Megan Leszczynski, Jian Zhang, and Christopher R ´e. Low- memory neural network training: A technical report. arXiv preprint arXiv:1904.10631, 2019. 7

    1904 arXiv

  49. [57]

    Pwc-net: Cnns for optical flow using pyramid, warping, and cost volume

    Deqing Sun, Xiaodong Yang, Ming-Yu Liu, and Jan Kautz. Pwc-net: Cnns for optical flow using pyramid, warping, and cost volume. InProceedings of the IEEE conference on computer vision and pattern recogni- tion, pages 8934–8943, 2018. 2

    2018

  50. [58]

    From sora what we can see: A survey of text-to-video generation

    Rui Sun, Yumin Zhang, Tejal Shah, Jiahao Sun, Shuoying Zhang, Wenqi Li, Haoran Duan, Bo Wei, and Rajiv Ranjan. From sora what we can see: A survey of text-to-video generation. arXiv preprint arXiv:2405.10674, 2024. 2

    2024 arXiv

  51. [59]

    Splatter a video: Video gaussian representation for versatile processing

    Yang-Tian Sun, Yi-Hua Huang, Lin Ma, Xiaoyang Lyu, Yan-Pei Cao, and Xiaojuan Qi. Splatter a video: Video gaussian representation for versatile processing. arXiv preprint arXiv:2406.13870, 2024. 2, 3, 6, 7, 8

    2024 arXiv

  52. [60]

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

    Stanislaw Szymanowicz, Chrisitian Rupprecht, and Andrea Vedaldi. Splatter image: Ultra-fast single- view 3d reconstruction. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pat- tern Recognition, pages 10208–10217, 2024. 4

    2024

  53. [61]

    Fourier features let networks learn high fre- quency functions in low dimensional domains

    Matthew Tancik, Pratul Srinivasan, Ben Mildenhall, Sara Fridovich-Keil, Nithin Raghavan, Utkarsh Sing- hal, Ravi Ramamoorthi, Jonathan Barron, and Ren Ng. Fourier features let networks learn high fre- quency functions in low dimensional domains. Ad- vances in neural informatio...

    2020

  54. [62]

    Lgm: Large multi- view gaussian model for high-resolution 3d content creation

    Jiaxiang Tang, Zhaoxi Chen, Xiaokang Chen, Tengfei Wang, Gang Zeng, and Ziwei Liu. Lgm: Large multi- view gaussian model for high-resolution 3d content creation. In European Conference on Computer Vi- sion, pages 1–18. Springer, 2025. 8, 1

    2025

  55. [63]

    Digital video processing

    A Murat Tekalp. Digital video processing . Prentice Hall Press, 2015. 2

    2015

  56. [64]

    The interpretation of structure from motion

    Shimon Ullman. The interpretation of structure from motion. Proceedings of the Royal Society of Lon- don. Series B. Biological Sciences , 203(1153):405– 426, 1979. 2, 3

    1979

  57. [65]

    Attention is all you need

    A Vaswani. Attention is all you need. Advances in Neural Information Processing Systems, 2017. 3, 5

    2017

  58. [66]

    Vggsfm: Visual geometry grounded deep structure from motion

    Jianyuan Wang, Nikita Karaev, Christian Rupprecht, and David Novotny. Vggsfm: Visual geometry grounded deep structure from motion. In Proceedings 10 of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 21686–21697, 2024. 3

    2024

  59. [67]

    Tracking everything everywhere all at once

    Qianqian Wang, Yen-Yu Chang, Ruojin Cai, Zhengqi Li, Bharath Hariharan, Aleksander Holynski, and Noah Snavely. Tracking everything everywhere all at once. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 19795–19806,

  60. [68]

    Shape of motion: 4d reconstruction from a single video

    Qianqian Wang, Vickie Ye, Hang Gao, Jake Austin, Zhengqi Li, and Angjoo Kanazawa. Shape of motion: 4d reconstruction from a single video. arXiv preprint arXiv:2407.13764, 2024. 2, 3

    2024

  61. [69]

    Dust3r: Geometric 3d vision made easy

    Shuzhe Wang, Vincent Leroy, Yohann Cabon, Boris Chidlovskii, and Jerome Revaud. Dust3r: Geometric 3d vision made easy. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recogni- tion, pages 20697–20709, 2024. 3

    2024

  62. [70]

    Gflow: Recover- ing 4d world from monocular video

    Shizun Wang, Xingyi Yang, Qiuhong Shen, Zhenx- iang Jiang, and Xinchao Wang. Gflow: Recover- ing 4d world from monocular video. arXiv preprint arXiv:2405.18426, 2024. 2, 3, 8

    2024 arXiv

  63. [71]

    Mvster: Epipolar transformer for efficient multi-view stereo

    Xiaofeng Wang, Zheng Zhu, Guan Huang, Fangbo Qin, Yun Ye, Yijia He, Xu Chi, and Xingang Wang. Mvster: Epipolar transformer for efficient multi-view stereo. In European Conference on Computer Vision, pages 573–591. Springer, 2022. 3

    2022

  64. [72]

    Nerf–: Neural radi- ance fields without known camera parameters

    Zirui Wang, Shangzhe Wu, Weidi Xie, Min Chen, and Victor Adrian Prisacariu. Nerf–: Neural radi- ance fields without known camera parameters. arXiv preprint arXiv:2102.07064, 2021. 2

    2021 arXiv

  65. [73]

    latentsplat: Au- toencoding variational gaussians for fast generalizable 3d reconstruction

    Christopher Wewer, Kevin Raj, Eddy Ilg, Bernt Schiele, and Jan Eric Lenssen. latentsplat: Au- toencoding variational gaussians for fast generalizable 3d reconstruction. arXiv preprint arXiv:2403.16292,

    arXiv

  66. [74]

    4d gaussian splatting for real- time dynamic scene rendering

    Guanjun Wu, Taoran Yi, Jiemin Fang, Lingxi Xie, Xiaopeng Zhang, Wei Wei, Wenyu Liu, Qi Tian, and Xinggang Wang. 4d gaussian splatting for real- time dynamic scene rendering. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pat- tern Recognition, pages 20310–203...

    2024

  67. [75]

    Unifying flow, stereo and depth estimation

    Haofei Xu, Jing Zhang, Jianfei Cai, Hamid Rezatofighi, Fisher Yu, Dacheng Tao, and Andreas Geiger. Unifying flow, stereo and depth estimation. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2023. 2, 5

    2023

  68. [76]

    Grm: Large gaussian reconstruction model for efficient 3d reconstruction and generation

    Yinghao Xu, Zifan Shi, Wang Yifan, Hansheng Chen, Ceyuan Yang, Sida Peng, Yujun Shen, and Gordon Wetzstein. Grm: Large gaussian reconstruction model for efficient 3d reconstruction and generation. arXiv preprint arXiv:2403.14621, 2024. 5, 1

    2024 arXiv

  69. [77]

    Depth anything v2

    Lihe Yang, Bingyi Kang, Zilong Huang, Zhen Zhao, Xiaogang Xu, Jiashi Feng, and Hengshuang Zhao. Depth anything v2. arXiv preprint arXiv:2406.09414,

    arXiv

  70. [78]

    Real-time photorealistic dynamic scene representa- tion and rendering with 4d gaussian splatting

    Zeyu Yang, Hongye Yang, Zijie Pan, and Li Zhang. Real-time photorealistic dynamic scene representa- tion and rendering with 4d gaussian splatting. arXiv preprint arXiv:2310.10642, 2023. 2, 3

    2023 arXiv

  71. [79]

    Deformable 3d gaussians for high-fidelity monocular dynamic scene reconstruction

    Ziyi Yang, Xinyu Gao, Wen Zhou, Shaohui Jiao, Yuqing Zhang, and Xiaogang Jin. Deformable 3d gaussians for high-fidelity monocular dynamic scene reconstruction. In Proceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition, pages 20331–20341, 2024. 2, 3

    2024

  72. [80]

    Joint feature learning and rela- tion modeling for tracking: A one-stream framework

    Botao Ye, Hong Chang, Bingpeng Ma, Shiguang Shan, and Xilin Chen. Joint feature learning and rela- tion modeling for tracking: A one-stream framework. In ECCV, 2022. 2

    2022

  73. [81]

    Deformable sprites for unsupervised video decomposition

    Vickie Ye, Zhengqi Li, Richard Tucker, Angjoo Kanazawa, and Noah Snavely. Deformable sprites for unsupervised video decomposition. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 2657–2666, 2022. 2, 3

    2022

  74. [82]

    Invariant training 2d-3d joint hard samples for few-shot point cloud recognition

    Xuanyu Yi, Jiajun Deng, Qianru Sun, Xian-Sheng Hua, Joo-Hwee Lim, and Hanwang Zhang. Invariant training 2d-3d joint hard samples for few-shot point cloud recognition. In Proceedings of the IEEE/CVF International Conference on Computer Vision , pages 14463–14474, 2023. 2

    2023

  75. [83]

    Mvgamba: Unify 3d content generation as state space sequence modeling

    Xuanyu Yi, Zike Wu, Qiuhong Shen, Qingshan Xu, Pan Zhou, Joo-Hwee Lim, Shuicheng Yan, Xinchao Wang, and Hanwang Zhang. Mvgamba: Unify 3d content generation as state space sequence modeling. arXiv preprint arXiv:2406.06367, 2024. 3

    2024 arXiv

  76. [84]

    Diffusion time- step curriculum for one image to 3d generation

    Xuanyu Yi, Zike Wu, Qingshan Xu, Pan Zhou, Joo- Hwee Lim, and Hanwang Zhang. Diffusion time- step curriculum for one image to 3d generation. In Proceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition , pages 9948– 9958, 2024. 2

    2024

  77. [85]

    Mip-splatting: Alias-free 3d gaussian splatting

    Zehao Yu, Anpei Chen, Binbin Huang, Torsten Sat- tler, and Andreas Geiger. Mip-splatting: Alias-free 3d gaussian splatting. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recogni- tion, pages 19447–19456, 2024. 2

    2024

  78. [86]

    Gaus- sian opacity fields: Efficient and compact surface re- construction in unbounded scenes

    Zehao Yu, Torsten Sattler, and Andreas Geiger. Gaus- sian opacity fields: Efficient and compact surface re- construction in unbounded scenes. arXiv preprint arXiv:2404.10772, 2024. 2

    2024 arXiv

  79. [87]

    Revisiting bfloat16 training

    Pedram Zamirai, Jian Zhang, Christopher R Aberger, and Christopher De Sa. Revisiting bfloat16 training. arXiv preprint arXiv:2010.06192, 2020. 7 11

    2010 arXiv

  80. [88]

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

    Chuanrui Zhang, Yingshuang Zou, Zhuoling Li, Min- min Yi, and Haoqian Wang. Transplat: Generalizable 3d gaussian splatting from sparse multi-view images with transformers. arXiv preprint arXiv:2408.13770,

    arXiv

  81. [89]

    A comprehensive survey of vision-based hu- man action recognition methods

    Hong-Bo Zhang, Yi-Xiang Zhang, Bineng Zhong, Qing Lei, Lijie Yang, Ji-Xiang Du, and Duan-Sheng Chen. A comprehensive survey of vision-based hu- man action recognition methods. Sensors, 19(5):1005,

  82. [90]

    Monst3r: A simple approach for estimating geometry in the presence of motion

    Junyi Zhang, Charles Herrmann, Junhwa Hur, Varun Jampani, Trevor Darrell, Forrester Cole, Deqing Sun, and Ming-Hsuan Yang. Monst3r: A simple approach for estimating geometry in the presence of motion. arXiv preprint arXiv:2410.03825, 2024. 3

    arXiv

  83. [91]

    Gs- lrm: Large reconstruction model for 3d gaussian splat- ting

    Kai Zhang, Sai Bi, Hao Tan, Yuanbo Xiangli, Nanx- uan Zhao, Kalyan Sunkavalli, and Zexiang Xu. Gs- lrm: Large reconstruction model for 3d gaussian splat- ting. In European Conference on Computer Vision , pages 1–19. Springer, 2025. 3, 5

    2025

  84. [92]

    The unreasonable ef- fectiveness of deep features as a perceptual metric

    Richard Zhang, Phillip Isola, Alexei A Efros, Eli Shechtman, and Oliver Wang. The unreasonable ef- fectiveness of deep features as a perceptual metric. In CVPR, 2018. 7

    2018

  85. [93]

    Stereo magnifica- tion: Learning view synthesis using multiplane im- ages

    Tinghui Zhou, Richard Tucker, John Flynn, Gra- ham Fyffe, and Noah Snavely. Stereo magnifica- tion: Learning view synthesis using multiplane im- ages. arXiv preprint arXiv:1805.09817 , 2018. 2, 3, 6

    2018 arXiv

  86. [94]

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

    Zehao Zhu, Zhiwen Fan, Yifan Jiang, and Zhangyang Wang. Fsgs: Real-time few-shot view synthesis using gaussian splatting. In European Conference on Com- puter Vision, pages 145–163. Springer, 2025. 2

    2025

  87. [95]

    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. arXiv preprint arXiv:2410.12781, 2024. 3

    2024 arXiv

  88. [96]

    Ewa splatting

    Matthias Zwicker, Hanspeter Pfister, Jeroen Van Baar, and Markus Gross. Ewa splatting. IEEE Transactions on Visualization and Computer Graphics , 8(3):223– 238, 2002. 1 12

    2002

Pith tools

Reviewed August 9, 2026 · model on record in the stance chip above.