Pith. sign in

REVIEW 4 major objections 6 minor 1 cited by

F3D-Gaus: Feed-forward 3D-aware Generation on ImageNet with Cycle-Aggregative Gaussian Splatting

T0 review · 4 major / 6 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read A feed-forward Gaussian-splatting model can learn 3D-aware generation from monocular ImageNet images alone, using cycle-aggregative self-supervision and geometry-guided video-inpainting refinement.

desk verdict The cycle-aggregation idea is genuinely clever, but Eq. (5) merges Gaussians from two different camera frames without a coordinate transform, and the main FID comparison is reconstruction, not generation; both need fixing before the headline claims hold. read the letter →

arxiv 2501.06714 v3 pith:CJ3GU32Z submitted 2025-01-12 cs.CV

Yuxin Wang , Qianyi Wu , Dan Xu This is my paper · ORCID
classification cs.CV
keywords 3D-awaregenerationGaussianSplattingmonoculartrainingnovelviewsynthesiscycleconsistencyself-supervisedlearningfeed-forwardinferenceImageNet
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 tries to establish that 3D-aware generation can be trained from monocular image datasets alone, without multi-view or video supervision. It proposes F3D-Gaus, a feed-forward pipeline that predicts a pixel-aligned Gaussian Splatting representation from a single RGB-D image and renders novel views directly. To make novel views consistent, it renders a novel view from the first representation, predicts a second representation from that rendered image, and aggregates the two Gaussian sets before re-rendering the canonical view for supervision. The paper argues this cycle-aggregative self-supervision lets the two views fill in each other's missing geometry, and that a geometry-guided video-inpainting refinement cleans up wide-angle artifacts. If correct, the method would make 3D content generation scalable to abundant single-image datasets while remaining fast at inference.

What carries the argument

The load-bearing object is the cycle-aggregative self-supervision built on pixel-aligned Gaussian Splatting. Each pixel predicts a Gaussian primitive (position, opacity, color, scale, rotation), and rendering is differentiable alpha-blending. The cycle works as follows: from input GS0, render a novel view; predict GS1 from that render; compute complementary masks from binarized alpha maps in both views, so primitives visible in one representation fill holes in the other; concatenate the selected primitives to form aggregated sets; re-render the canonical view and apply the reconstruction and cycle losses, with gradients blocked from flowing back through the rendered novel view into GS0. This aggregation is what carries the argument: it is the mechanism supposed to teach the model to extrapolate geometry beyond the visible frontal view without ever seeing a real second view.

What would settle it

Take an object with known ground-truth 3D geometry, train the model on only single views of it, and then measure the depth error of held-out novel views against the true depth maps; if the cycle-aggregative constraint only enforces appearance consistency without true geometric alignment, depth error will grow sharply as the novel-view angle increases even when the rendered images look plausible.

Watch

Extended reading notes

Core claim

On its own terms, the paper's discovery is that a pixel-aligned Gaussian Splatting model can learn generalizable 3D-aware generation from monocular data by closing a render-and-reaggregate cycle. The model takes an image and its depth, predicts Gaussian primitives aligned to image pixels, renders a random novel view, feeds that rendered view back through the same network to get a second set of primitives, and then uses masks computed from alpha maps to concatenate the two sets. Re-rendering the aggregated set back at the canonical view and comparing with the input provides self-supervision that propagates through both Gaussian sets, forcing them to be complementary and multi-view consistent. The paper further adds artifact masks derived from normal maps and a fixed video-inpainting model to refine textures in wide-viewpoint renderings. The reported result is that on ImageNet and single-class datasets this pipeline outperforms the tri-plane and diffusion baselines on fidelity, depth continuity, and speed.

Load-bearing premise

The self-supervised strategy assumes that Gaussian primitives predicted from two different camera views can be directly concatenated as if they were already in the same 3D coordinate frame; if pixel-aligned primitives remain in their own camera coordinates, the aggregation merges geometry that is not actually aligned.

Editorial extensions

If this is right

  • 3D-aware generation can be trained directly on massive monocular datasets, removing the need for expensive multi-view or video capture.
  • Feed-forward inference produces novel views and depth maps in a single pass at 256 resolution, with no per-scene optimization or super-resolution post-processing.
  • Aggregating the canonical and novel-view Gaussian sets at inference time improves rendering fidelity and reduces ghosting artifacts for wide viewpoint changes.
  • The geometry-guided video-inpainting stage suppresses artifacts near object boundaries and in large-angle views.
  • The model generalizes to out-of-domain scene-level images and to images produced by other generators, not just the training distribution.

Reading between the lines

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

  • Editorial inference: the cycle-aggregation can be read as a form of self-distillation in which the model learns to complete its own occlusions; the same render-and-reaggregate idea could be applied to other pixel-aligned representations, such as triplanes or voxel grids, and might deliver similar consistency gains.
  • Editorial inference: because the paper concatenates primitives from different camera coordinate frames without an explicit world-frame transform, a cleaner reformulation would transform GS1 into GS0's coordinate system before aggregation; a controlled comparison would show whether most of the benefit comes from geometric alignment or merely from having more primitives.
  • Editorial inference: the method's reliance on pseudo-depth from a monocular estimator is a practical bottleneck; the paper's own ablation shows depth accuracy degrades when the U-Net predicts depth jointly, suggesting that improved depth priors or a depth-free formulation would be a natural next step.
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 F3D-Gaus, a feed-forward pixel-aligned Gaussian Splatting model for 3D-aware generation from monocular datasets such as ImageNet. The method takes an RGB-D image as input, predicts a set of Gaussian primitives, and then uses a self-supervised cycle-aggregative strategy in which a rendered novel view is fed back through the same network to obtain a second set of Gaussians; the two sets are then concatenated and re-rendered to the canonical view for supervision. A second training stage uses a fixed video inpainting model, guided by rendered alpha and normal maps, to refine artifacts in novel views. Experiments on ImageNet, Dogs, Horses, and Elephants report FID, Inception Score, and Non-Flatness Score, along with training and inference time comparisons against G3DR and IVID.

Significance. If the method worked as described, it would be a significant contribution: a feed-forward 3D-aware generation pipeline trained without multi-view or video data, with efficient rendering and improved image quality over prior monocular tri-plane approaches. The idea of cycling a rendered view back through a pixel-aligned Gaussian predictor is interesting, and the reported efficiency numbers are useful. However, the central aggregation mechanism is geometrically undefined, and the main quantitative comparison does not actually measure generative performance. These issues affect the load-bearing claims of the paper, so the current significance is not established. I also note that the paper does not provide code or reproducible implementation details, which makes evaluation of the method harder.

major comments (4)
  1. [§3.2, Eq. (5)] The aggregation in Eq. (5) concatenates GS0 and GS1 without any transformation into a common coordinate frame. According to §3.1, pixel-aligned 3DGS coordinates are computed by adding the input depth to a predicted offset along the input camera's rays, so GS0 is expressed in camera-0 coordinates while GS1, predicted from the rendered novel view \tilde I1/\tilde D1, is expressed in camera-1 coordinates. Rendering \hat GS0 = Concat(GS0, GS1[M_{1→0}]) from view0 requires every selected GS1 primitive to be transformed by the relative pose T_{1→0}, but no such transformation appears in Eq. (5) or anywhere in the cycle-supervision description. Without this transformation, the renderer treats two disjoint coordinate systems as one, and the cycle loss can be minimized by point sets that merely co-render a plausible image without representing a geometrically consistent 3D scene. This is the load-bearing step of the paper's claimed cross-view alignment, and it is undefined as stated.
  2. [§4.2, Table 1] The headline quantitative comparison is not a generative evaluation. The text states that 'we use ImageNet images as input to reconstruct the original view for comparison with G3DR,' and FID is computed on the full ImageNet training set. F3D-Gaus is therefore evaluated as an autoencoder on the training distribution, not as a generator producing new samples. The FID gap (1.6 vs 24.9) largely reflects that F3D-Gaus is trained to reconstruct real ImageNet images, while G3DR's ftrigen is evaluated on real images that are out of its training distribution. The reported Inception Score of 308.6 similarly measures the classifiability of reconstructions rather than the quality of newly generated content. This does not support the abstract's claim of 'state-of-the-art realism in 3D rendering' for generation. A proper generative evaluation would require sampling from the model's prior or class condition and comparing against generative baselines in that setting.
  3. [§3.2, cycle supervision] The cycle-consistency objective is internal: the novel view \tilde I1 and depth \tilde D1 used as input to predict GS1 are themselves rendered by the model's own GS0. The cycle loss Lcycle then aligns the aggregated representation back to the original view. As a result, the supervision can be minimized by predictions that co-render a plausible canonical image without learning externally consistent 3D structure; the only external anchors are Lphoto and the perceptual/CLIP losses, which are 2D measures. The paper's claim that cycle-aggregative training 'enforces cross-view consistency in the learned 3D representation' would require, at minimum, a valid coordinate alignment as described in the first major comment, and ideally some external multi-view or geometric prior. In its current form, the consistency is between the model's own predictions and does not provide evidence of geometric correctness.
  4. [§4.2, Table 2] Even in the two-stage evaluation of Table 2, F3D-Gaus is not evaluated as a standalone generative model; it operates on 10k images sampled by IVID's conditional generator. The reported FID of 15.5 on these generated inputs, while better than IVID and G3DR on two of three metrics, does not demonstrate that the overall system achieves state-of-the-art 3D-aware generation, because the first-stage generator is shared and the metric reflects the composite pipeline. For a paper positioned as 3D-aware generation, the authors should demonstrate a full generative capability, for example by integrating F3D-Gaus with a class-conditioned RGBD generator and sampling from scratch. As it stands, the contribution is a reconstruction/rendering module, which is narrower than the title and abstract claim.
minor comments (6)
  1. [Supplementary Sec. A] The opening sentence of the supplementary material is duplicated: 'In this supplementary material, we provide more additional experiments In this supplementary material, we provide more additional experiments'.
  2. [§3.2, Eq. (4)] The notation M_{1→0} and M_{0→1} is confusing; please specify on which image plane each mask is defined and how a 2D mask in the novel view selects pixel-aligned primitives that are then inserted into a set expressed in the canonical view.
  3. [Figure 2] Figure 2 contains placeholder text fragments such as 'rendered input ... rendered' that appear to be unfinished; the figure should be cleaned up.
  4. [§1, contributions] The paper claims to be the first to apply pixel-aligned Gaussian Splatting to generalizable 3D-aware generation on monocular datasets, but the related work already includes feed-forward pixel-aligned GS models such as Splatter-Image and Flash3D; the novelty should be clarified to be the monocular training setting and the cycle-aggregative mechanism rather than the representation itself.
  5. [§4.2, Table 1] The comparison at 2562 resolution is not apples-to-apples: G3DR is rendered at 128 and upsampled with Real-ESRGAN, while F3D-Gaus is trained and rendered directly at 256; the paper should either train both at the same resolution or clearly state this limitation in the table caption.
  6. [§4.2, Table 3] The training time difference between F3D-Gaus (13.0 A100 days) and G3DR (14.5 A100 days) is modest; the text says 'significantly improves training and inference efficiency,' which overstates the measured training improvement.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the cycle-aggregative loss is a self-supervised objective, not an identity reduction; the camera-frame concern is a geometric soundness issue, not circularity.

full rationale

The paper's central training mechanism (Sec. 3.2, Eqs. 2-6 and 10) is self-supervised cycle consistency: a novel view (I1_tilde, D1_tilde) is rendered from the model's own GS0, fed through the same U-Net to produce GS1, the two sets are aggregated via Eq. (5), and the result is re-rendered back to the canonical view and compared with the original I0, D0. This is a pseudo-labeling/cycle-consistency training objective, not a fitted parameter renamed as a prediction: no parameter is fitted to a subset of data and then reported as an independent prediction, and no load-bearing mathematical claim is imported from a self-citation. Citations to the authors' own prior work (e.g., [1], [47], [48], [53], [64]) are contextual related-work references and do not ground the cycle derivation. The geometric concern that GS0 is expressed in camera-0 coordinates while GS1 is predicted from a rendered novel view in camera-1 coordinates, with no explicit transform appearing in Eq. (5), is a real validity/soundness issue: it questions whether the learned consistency corresponds to genuine 3D alignment. However, that is not circularity under the required standard, because the loss is not equivalent to its input by construction; the model could in principle learn an aligned representation, and the paper reports external metrics (ImageNet FID/IS/NFS, efficiency) against G3DR and IVID. The acknowledged limitation in Supplementary Sec. A.3 about depth-input dependence is a limitation, not a circular step. Therefore no circular step meeting the specified evidentiary standard is present.

Assumptions & free parameters 4 free parameters · 3 assumptions · 0 invented entities

The central claim rests on three unstated assumptions: LeReS pseudo-depth is trusted as ground truth, the two predicted Gaussian sets are implicitly assumed to share a coordinate frame, and ProPainter's inpainted output is trusted as supervision. These are not free parameters in the classic sense but are load-bearing external inputs.

free parameters (4)
  • Alpha mask binarization threshold tau = 0.5
    Used in Eq. 4 (Sec 3.2) to construct complementary masks; chosen by hand and not ablated.
  • Artifact mask thresholds tau and tau_theta = unspecified
    Used in Eq. 11 (Sec 3.3) to detect artifacts; values are not reported.
  • Loss weights lambda_perp, lambda_CLIP, lambda_reg, lambda_video = unspecified
    Weights in Eqs. 8, 10, 13; not reported in the main text, so the objective is under-specified.
  • Camera sampling variances for novel views = 0.3 yaw, 0.15 pitch
    Gaussian variances used for novel-view sampling in Sec 4.1; chosen by hand, may affect all downstream metrics.
assumptions (3)
  • domain assumption Pseudo-depth maps from LeReS are treated as ground-truth geometry for supervision and input.
    The model takes RGB-D as input and all reconstruction/cycle losses (Eqs. 2, 6) use D0 as the target; any LeReS bias (scale, smoothness) is baked into the learned representation. See Sec 3.2 and Sec 4.1.
  • domain assumption GS0 and GS1 are in a common coordinate frame and can be concatenated directly.
    Eq. 5 concatenates primitives from two views without a transform; the paper never states how view1-camera-space coordinates are mapped into view0 space. This is load-bearing for the whole cycle-aggregation.
  • domain assumption The video inpainting model ProPainter produces reliable pseudo-ground-truth for refinement.
    In Sec 3.3, in-painted frames are used as supervision (Eq. 13); errors in inpainting (especially for novel objects) would be learned as target textures.

how reviews work

0 comments
Cite this review

Pith. "Pith review of F3D-Gaus: Feed-forward 3D-aware Generation on ImageNet with Cycle-Aggregative Gaussian Splatting." pith.science (2026). https://pith.science/paper/CJ3GU32Z

@misc{pith2026250106714,
  author       = {Pith},
  title        = {Pith review of: F3D-Gaus: Feed-forward 3D-aware Generation on ImageNet with Cycle-Aggregative Gaussian Splatting},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/CJ3GU32Z}},
  note         = {Machine review of arXiv:2501.06714}
}
read the original abstract

This paper tackles the problem of generalizable 3D-aware generation from monocular datasets, e.g., ImageNet. The key challenge of this task is learning a robust 3D-aware representation without multi-view or dynamic data, while ensuring consistent texture and geometry across different viewpoints. Although some baseline methods are capable of 3D-aware generation, the quality of the generated images still lags behind state-of-the-art 2D generation approaches, which excel in producing high-quality, detailed images. To address this severe limitation, we propose a novel feed-forward pipeline based on pixel-aligned Gaussian Splatting, coined as F3D-Gaus, which can produce more realistic and reliable 3D renderings from monocular inputs. In addition, we introduce a self-supervised cycle-aggregative constraint to enforce cross-view consistency in the learned 3D representation. This training strategy naturally allows aggregation of multiple aligned Gaussian primitives and significantly alleviates the interpolation limitations inherent in single-view pixel-aligned Gaussian Splatting. Furthermore, we incorporate video model priors to perform geometry-aware refinement, enhancing the generation of fine details in wide-viewpoint scenarios and improving the model's capability to capture intricate 3D textures. Extensive experiments demonstrate that our approach not only achieves high-quality, multi-view consistent 3D-aware generation from monocular datasets, but also significantly improves training and inference efficiency.

Figures

Figures reproduced from arXiv: 2501.06714 by the authors.

Figure 1
Figure 1. Illustration of our motivation for cycle self￾supervised training. For monocular datasets: (a) supervision is naturally available for the canonical view. (b) For novel views, where supervision is absent, we use the rendered novel-view im￾age as input to obtain its 3D representation. This 3D representation is then re-rendered from the canonical view, where supervision is available. Red arrows indicate feed-forward 3D… view at source ↗
Figure 2
Figure 2. Illustration of our overall framework. Given a single RGB image I0 and depth map D0, our model directly feeds them forward to output the pixel-aligned Gaussian Splatting representation GS0, which can be used for novel view synthesis. After obtaining the 3DGS representation, we render the image I˜1 and depth maps D˜1 for the novel view, and then output its corresponding 3DGS GS1. These two 3DGS representations are su… view at source ↗
Figure 3
Figure 3. Illustration of the proposed cycle-aggregative self￾supervised strategy. We guide complementary aggregation by leveraging the differences between the alpha maps of the two 3DGS from different viewpoints. and complementary. As shown in [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (9 more)
Figure 4
Figure 4. Figure 4: Illustration of geometry-guided texture refinement. (a) illustrates artifact localization in novel views, while (b) shows geometry mask-guided sequence in-painting. view. Next, we run another inference step to obtain novel￾view representation GS1, which are concatenate…
Figure 5
Figure 5. Figure 5: Qualitative visualization of rendered images and depth maps on the ImageNet dataset. Our method can generate novel view images along with corresponding depth maps for input images across various categories. alpha and normal maps, to fill in the missing regions in these…
Figure 6
Figure 6. Figure 6: Qualitative results of rendered images and depth maps on SDIP Elephants [27], LSUN Horses [58], and Dogs [27]. demonstrating its impressive generalization capabilities. Efficiency. Tab. 3 compares our method’s efficiency with IVID and G3DR. Our approach reduces the tra…
Figure 7
Figure 7. Figure 7: Performance comparison of three 1282 models across varying yaw angle ranges. The FID, IS, and NFS metrics are evaluated on the filtered subset of ImageNet. The x-axis represents the yaw angle ranges, while the y-axis denotes the corresponding metric values. (a) Input (…
Figure 8
Figure 8. Figure 8: Results on two complex indoor scene images out￾side the ImageNet dataset. The rendered novel view images, along with their corresponding depth and normal maps, effectively demonstrate the generalization capability of our model. three variants perform similarly, with th…
Figure 10
Figure 10. Figure 10: Qualitative comparison with G3DR [30]. We use Real-ESRGAN[46] to upsample G3DR’s 1282 outputs to 2562 . For our F3D-Gaus, we accept 2562 resolution inputs and ren￾der directly at the same 2562 resolution, without requiring post￾processing. In this supplementary materi…
Figure 12
Figure 12. Figure 12: Additional qualitative visualization of rendered images and depth maps on the ImageNet dataset [PITH_FULL_IMAGE:figures/full_fig_p013_12.png]
Figure 13
Figure 13. Figure 13: Another qualitative visualization of rendered images and depth maps on the ImageNet dataset [PITH_FULL_IMAGE:figures/full_fig_p014_13.png]
Figure 14
Figure 14. Figure 14: Qualitative results on VAR [41] output (left) and FFHQ images (right). B. Video Demo As a supplement to Sec. A.1, we have included a video demo to showcase the quantitative results in a video for￾mat. As shown in Fig.10, we use Real-ESRGAN [46] to upsample 1282 output…

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

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

    cs.CV 2026-07 conditional novelty 6.0 of 10

    Texture-aware SuperCluster pruning plus an adaptive Gaussian head lets feed-forward 3DGS models hit a user budget β while outperforming post-hoc pruners on RE10K, ACID, DL3DV and DTU.

Reference graph

Works this paper leans on

69 extracted references · 54 canonical work pages · cited by 1 Pith paper

  1. [1]

    3dgs-det: Em- power 3d gaussian splatting with boundary guidance and box-focused sampling for 3d object detection

    Yang Cao, Yuanliang Jv, and Dan Xu. 3dgs-det: Em- power 3d gaussian splatting with boundary guidance and box-focused sampling for 3d object detection. arXiv preprint arXiv:2410.01647, 2024. 3

    arXiv 2024

  2. [2]

    Efficient geometry-aware 3d generative adversarial networks

    Eric R Chan, Connor Z Lin, Matthew A Chan, Koki Nagano, Boxiao Pan, Shalini De Mello, Orazio Gallo, Leonidas J Guibas, Jonathan Tremblay, Sameh Khamis, et al. Efficient geometry-aware 3d generative adversarial networks. In CVPR, 2022. 2

    work page 2022

  3. [3]

    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, 2024. 3

    work page 2024

  4. [4]

    Pgsr: Planar-based gaussian splatting for efficient and high-fidelity sur- face reconstruction

    Danpeng Chen, Hai Li, Weicai Ye, Yifan Wang, Wei- jian Xie, Shangjin Zhai, Nan Wang, Haomin Liu, Hu- jun Bao, and Guofeng Zhang. Pgsr: Planar-based gaussian splatting for efficient and high-fidelity sur- face reconstruction. arxiv preprint arxiv:2406.06521,

    arXiv

  5. [5]

    Gaussianeditor: Swift and controllable 3d editing with gaussian splat- ting

    Yiwen Chen, Zilong Chen, Chi Zhang, Feng Wang, Xiaofeng Yang, Yikai Wang, Zhongang Cai, Lei Yang, Huaping Liu, and Guosheng Lin. Gaussianeditor: Swift and controllable 3d editing with gaussian splat- ting. In CVPR, 2024. 3

    work page 2024

  6. [6]

    Mvsplat360: Feed-forward 360 scene synthesis from sparse views

    Yuedong Chen, Chuanxia Zheng, Haofei Xu, Bohan Zhuang, Andrea Vedaldi, Tat-Jen Cham, and Jianfei Cai. Mvsplat360: Feed-forward 360 scene synthesis from sparse views. NeurIPS, 2024. 3

    work page 2024

  7. [7]

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

    work page

  8. [8]

    3d-r2n2: A unified approach for single and multi-view 3d object recon- struction

    Christopher B Choy, Danfei Xu, JunYoung Gwak, Kevin Chen, and Silvio Savarese. 3d-r2n2: A unified approach for single and multi-view 3d object recon- struction. In ECCV, 2016. 2

    work page 2016

Show all 69 references
  1. [9]

    Imagenet: A large-scale hierarchi- cal image database

    Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei-Fei. Imagenet: A large-scale hierarchi- cal image database. In CVPR, 2009. 1, 6

    2009

  2. [10]

    Cat3d: Cre- ate anything in 3d with multi-view diffusion models

    Ruiqi Gao, Aleksander Holynski, Philipp Henzler, Arthur Brussee, Ricardo Martin-Brualla, Pratul Srini- vasan, Jonathan T Barron, and Ben Poole. Cat3d: Cre- ate anything in 3d with multi-view diffusion models. arXiv preprint arXiv:2405.10314, 2024. 1, 2

    2024 arXiv

  3. [11]

    Mesh r-cnn

    Georgia Gkioxari, Jitendra Malik, and Justin Johnson. Mesh r-cnn. In ICCV, 2019. 2

    2019

  4. [12]

    Animatediff: Animate your personalized text-to-image diffusion models without specific tuning

    Yuwei Guo, Ceyuan Yang, Anyi Rao, Zhengyang Liang, Yaohui Wang, Yu Qiao, Maneesh Agrawala, Dahua Lin, and Bo Dai. Animatediff: Animate your personalized text-to-image diffusion models without specific tuning. ICLR, 2024. 2

    2024

  5. [13]

    Gans trained by a two time-scale update rule converge to a local nash equilibrium

    Martin Heusel, Hubert Ramsauer, Thomas Un- terthiner, Bernhard Nessler, and Sepp Hochreiter. Gans trained by a two time-scale update rule converge to a local nash equilibrium. NeurIPS, 2017. 6

    2017

  6. [14]

    Text2room: Extracting textured 3d meshes from 2d text-to-image models

    Lukas H ¨ollein, Ang Cao, Andrew Owens, Justin John- son, and Matthias Nießner. Text2room: Extracting textured 3d meshes from 2d text-to-image models. In ICCV, 2023. 2

    2023

  7. [15]

    Viewdiff: 3d-consistent image generation with text-to-image models

    Lukas H ¨ollein, Alja ˇz Bo ˇziˇc, Norman M ¨uller, David Novotny, Hung-Yu Tseng, Christian Richardt, Michael Zollh ¨ofer, and Matthias Nießner. Viewdiff: 3d-consistent image generation with text-to-image models. In CVPR, 2024. 2

    2024

  8. [16]

    Lrm: Large reconstruc- tion model for single image to 3d

    Yicong Hong, Kai Zhang, Jiuxiang Gu, Sai Bi, Yang Zhou, Difan Liu, Feng Liu, Kalyan Sunkavalli, Trung Bui, and Hao Tan. Lrm: Large reconstruc- tion model for single image to 3d. arXiv preprint arXiv:2311.04400, 2023. 1, 2, 3

    2023 arXiv

  9. [17]

    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. InSIGGRAPH,

  10. [18]

    Real3d: Scaling up large reconstruction models with real-world images

    Hanwen Jiang, Qixing Huang, and Georgios Pavlakos. Real3d: Scaling up large reconstruction models with real-world images. arXiv preprint arXiv:2406.08479,

    arXiv

  11. [19]

    Per- ceptual losses for real-time style transfer and super- resolution

    Justin Johnson, Alexandre Alahi, and Li Fei-Fei. Per- ceptual losses for real-time style transfer and super- resolution. In ECCV, 2016. 5

    2016

  12. [20]

    A style- based generator architecture for generative adversarial networks

    Tero Karras, Samuli Laine, and Timo Aila. A style- based generator architecture for generative adversarial networks. In CVPR, 2019. 5

    2019

  13. [21]

    Repurposing diffusion-based image generators for monocular depth estimation

    Bingxin Ke, Anton Obukhov, Shengyu Huang, Nando Metzger, Rodrigo Caye Daudt, and Konrad Schindler. Repurposing diffusion-based image generators for monocular depth estimation. In CVPR, 2024. 3

    2024

  14. [22]

    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. ToG,

  15. [23]

    Infinitenature-zero: Learning per- petual view generation of natural scenes from single images

    Zhengqi Li, Qianqian Wang, Noah Snavely, and Angjoo Kanazawa. Infinitenature-zero: Learning per- petual view generation of natural scenes from single images. In ECCV, 2022. 2

    2022

  16. [24]

    3dgsr: Implicit surface recon- struction with 3d gaussian splatting

    Xiaoyang Lyu, Yang-Tian Sun, Yi-Hua Huang, Xi- uzhe Wu, Ziyi Yang, Yilun Chen, Jiangmiao Pang, and Xiaojuan Qi. 3dgsr: Implicit surface recon- struction with 3d gaussian splatting. arXiv preprint arXiv:2404.00409, 2024. 3

    2024 arXiv

  17. [25]

    Im-3d: Iterative multiview diffu- sion and reconstruction for high-quality 3d generation

    Luke Melas-Kyriazi, Iro Laina, Christian Rupprecht, Natalia Neverova, Andrea Vedaldi, Oran Gafni, and Filippos Kokkinos. Im-3d: Iterative multiview diffu- sion and reconstruction for high-quality 3d generation. ICML, 2024. 2

    2024

  18. [26]

    Nerf: Representing scenes as neural radiance fields for view synthesis

    Ben Mildenhall, Pratul P Srinivasan, Matthew Tancik, Jonathan T Barron, Ravi Ramamoorthi, and Ren Ng. Nerf: Representing scenes as neural radiance fields for view synthesis. Communications of the ACM, 2021. 2

    2021

  19. [27]

    Self-distilled stylegan: Towards gener- ation from internet photos

    Ron Mokady, Omer Tov, Michal Yarom, Oran Lang, Inbar Mosseri, Tali Dekel, Daniel Cohen-Or, and Michal Irani. Self-distilled stylegan: Towards gener- ation from internet photos. In SIGGRAPH, 2022. 6, 7

    2022

  20. [28]

    Magic123: One image to high- quality 3d object generation using both 2d and 3d dif- fusion priors

    Guocheng Qian, Jinjie Mai, Abdullah Hamdi, Jian Ren, Aliaksandr Siarohin, Bing Li, Hsin-Ying Lee, Ivan Skorokhodov, Peter Wonka, Sergey Tulyakov, and Bernard Ghanem. Magic123: One image to high- quality 3d object generation using both 2d and 3d dif- fusion priors. In ICLR, 2024. 2

    2024

  21. [29]

    Learning transferable visual models from nat- ural 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 nat- ural language supervision. In ICML, 2021. 5

    2021

  22. [30]

    G3dr: Generative 3d reconstruction in imagenet

    Pradyumna Reddy, Ismail Elezi, and Jiankang Deng. G3dr: Generative 3d reconstruction in imagenet. In CVPR, 2024. 2, 5, 6, 7

    2024

  23. [31]

    Vq3d: Learning a 3d-aware generative model on imagenet

    Kyle Sargent, Jing Yu Koh, Han Zhang, Huiwen Chang, Charles Herrmann, Pratul Srinivasan, Jiajun Wu, and Deqing Sun. Vq3d: Learning a 3d-aware generative model on imagenet. In ICCV, 2023. 2

    2023

  24. [32]

    ZeroNVS: Zero-shot 360-degree view syn- thesis from a single real image

    Kyle Sargent, Zizhang Li, Tanmay Shah, Charles Her- rmann, Hong-Xing Yu, Yunzhi Zhang, Eric Ryan Chan, Dmitry Lagun, Li Fei-Fei, Deqing Sun, and Ji- ajun Wu. ZeroNVS: Zero-shot 360-degree view syn- thesis from a single real image. CVPR, 2024. 1, 2

    2024

  25. [33]

    Language embedded 3d gaussians for open-vocabulary scene understanding

    Jin-Chuan Shi, Miao Wang, Hao-Bin Duan, and Shao- Hua Guan. Language embedded 3d gaussians for open-vocabulary scene understanding. In CVPR,

  26. [34]

    Mvdream: Multi-view diffusion for 3d generation

    Yichun Shi, Peng Wang, Jianglong Ye, Long Mai, Ke- jie Li, and Xiao Yang. Mvdream: Multi-view diffusion for 3d generation. arXiv:2308.16512, 2023. 1, 2

    2023 arXiv

  27. [35]

    3d generation on imagenet

    Ivan Skorokhodov, Aliaksandr Siarohin, Yinghao Xu, Jian Ren, Hsin-Ying Lee, Peter Wonka, and Sergey Tulyakov. 3d generation on imagenet. In ICLR, 2023. 2, 6

    2023

  28. [36]

    Splatt3r: Zero-shot gaus- sian splatting from uncalibarated image pairs

    Brandon Smart, Chuanxia Zheng, Iro Laina, and Vic- tor Adrian Prisacariu. Splatt3r: Zero-shot gaus- sian splatting from uncalibarated image pairs. arXiv preprint arXiv:2408.13912, 2024. 3

    2024 arXiv

  29. [37]

    Flash3d: Feed-forward generalisable 3d scene reconstruction from a single image

    Stanislaw Szymanowicz, Eldar Insafutdinov, Chuanxia Zheng, Dylan Campbell, Joao Henriques, Christian Rupprecht, and Andrea Vedaldi. Flash3d: Feed-forward generalisable 3d scene reconstruction from a single image. arxiv, 2024. 3

    2024

  30. [38]

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

    Stanislaw Szymanowicz, Chrisitian Rupprecht, and Andrea Vedaldi. Splatter image: Ultra-fast single- view 3d reconstruction. In CVPR, 2024. 1, 2, 3, 4

    2024

  31. [39]

    Make-it-3d: High-fidelity 3d creation from a single image with dif- fusion prior

    Junshu Tang, Tengfei Wang, Bo Zhang, Ting Zhang, Ran Yi, Lizhuang Ma, and Dong Chen. Make-it-3d: High-fidelity 3d creation from a single image with dif- fusion prior. In ICCV, 2023. 1, 2

    2023

  32. [40]

    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 ECCV, 2024. 1, 2, 3

    2024

  33. [41]

    Visual autoregressive modeling: Scalable image generation via next-scale prediction

    Keyu Tian, Yi Jiang, Zehuan Yuan, Bingyue Peng, and Liwei Wang. Visual autoregressive modeling: Scalable image generation via next-scale prediction. NeurIPS, 2024. 5

    2024

  34. [42]

    SV3D: Novel multi-view synthesis and 3D generation from a single image using latent video diffusion

    Vikram V oleti, Chun-Han Yao, Mark Boss, Adam Letts, David Pankratz, Dmitrii Tochilkin, Christian Laforte, Robin Rombach, and Varun Jampani. SV3D: Novel multi-view synthesis and 3D generation from a single image using latent video diffusion. In ECCV,

  35. [43]

    Gaussianeditor: Editing 3d gaus- sians delicately with text instructions

    Junjie Wang, Jiemin Fang, Xiaopeng Zhang, Lingxi Xie, and Qi Tian. Gaussianeditor: Editing 3d gaus- sians delicately with text instructions. In CVPR, 2024. 3

    2024

  36. [44]

    Pixel2mesh: Generating 3d mesh models from single rgb images

    Nanyang Wang, Yinda Zhang, Zhuwen Li, Yanwei Fu, Wei Liu, and Yu-Gang Jiang. Pixel2mesh: Generating 3d mesh models from single rgb images. In ECCV,

  37. [45]

    Imagedream: Image- prompt multi-view diffusion for 3d generation

    Peng Wang and Yichun Shi. Imagedream: Image- prompt multi-view diffusion for 3d generation. arXiv preprint arXiv:2312.02201, 2023. 1, 2

    2023 arXiv

  38. [46]

    Real-esrgan: Training real-world blind super- resolution with pure synthetic data

    Xintao Wang, Liangbin Xie, Chao Dong, and Ying Shan. Real-esrgan: Training real-world blind super- resolution with pure synthetic data. In ICCVW, 2021. 6, 7, 2, 5

    2021

  39. [47]

    Learning 3d geometry and feature consistent gaussian splatting for object removal

    Yuxin Wang, Qianyi Wu, Guofeng Zhang, and Dan Xu. Learning 3d geometry and feature consistent gaussian splatting for object removal. In ECCV, 2025. 3

    2025

  40. [48]

    Pygs: Large-scale scene representation with pyramidal 3d gaussian splatting

    Zipeng Wang and Dan Xu. Pygs: Large-scale scene representation with pyramidal 3d gaussian splatting. arXiv preprint arXiv:2405.16829, 2024. 3

    2024 arXiv

  41. [49]

    Motionctrl: A unified and flexible motion con- troller for video generation

    Zhouxia Wang, Ziyang Yuan, Xintao Wang, Yaowei Li, Tianshui Chen, Menghan Xia, Ping Luo, and Ying Shan. Motionctrl: A unified and flexible motion con- troller for video generation. In ACM SIGGRAPH ,

  42. [50]

    Sur- face reconstruction from 3d gaussian splatting via lo- cal structural hints

    Qianyi Wu, Jianmin Zheng, and Jianfei Cai. Sur- face reconstruction from 3d gaussian splatting via lo- cal structural hints. In European Conference on Com- puter Vision, 2024. 3

    2024

  43. [51]

    Unsupervised learning of probably symmet- ric deformable 3d objects from images in the wild

    Shangzhe Wu, Christian Rupprecht, and Andrea Vedaldi. Unsupervised learning of probably symmet- ric deformable 3d objects from images in the wild. In Proceedings of the IEEE/CVF conference on com- puter vision and pattern recognition , pages 1–10,

  44. [52]

    3d-aware image generation using 2d dif- fusion models

    Jianfeng Xiang, Jiaolong Yang, Binbin Huang, and Xin Tong. 3d-aware image generation using 2d dif- fusion models. In ICCV, 2023. 2, 6, 7, 5

    2023

  45. [53]

    Gs-slam: Dense visual slam with 3d gaussian splatting

    Chi Yan, Delin Qu, Dan Xu, Bin Zhao, Zhigang Wang, Dong Wang, and Xuelong Li. Gs-slam: Dense visual slam with 3d gaussian splatting. In CVPR, 2024. 3

    2024

  46. [54]

    Depth any- thing: Unleashing the power of large-scale unlabeled data

    Lihe Yang, Bingyi Kang, Zilong Huang, Xiaogang Xu, Jiashi Feng, and Hengshuang Zhao. Depth any- thing: Unleashing the power of large-scale unlabeled data. In CVPR, 2024. 3

    2024

  47. [55]

    Gaussian grouping: Segment and edit anything in 3d scenes

    Mingqiao Ye, Martin Danelljan, Fisher Yu, and Lei Ke. Gaussian grouping: Segment and edit anything in 3d scenes. In ECCV, 2025. 3

    2025

  48. [56]

    Learning to recover 3d scene shape from a single im- age

    Wei Yin, Jianming Zhang, Oliver Wang, Simon Niklaus, Long Mai, Simon Chen, and Chunhua Shen. Learning to recover 3d scene shape from a single im- age. In CVPR, 2021. 2, 3, 6

    2021

  49. [57]

    Nvs- solver: Video diffusion model as zero-shot novel view synthesizer

    Meng You, Zhiyu Zhu, Hui Liu, and Junhui Hou. Nvs- solver: Video diffusion model as zero-shot novel view synthesizer. arXiv preprint arXiv:2405.15364, 2024. 1, 2

    2024 arXiv

  50. [58]

    Lsun: Con- struction of a large-scale image dataset using deep learning with humans in the loop

    Fisher Yu, Ari Seff, Yinda Zhang, Shuran Song, Thomas Funkhouser, and Jianxiong Xiao. Lsun: Con- struction of a large-scale image dataset using deep learning with humans in the loop. arXiv preprint arXiv:1506.03365, 2015. 6, 7

    2015 arXiv

  51. [59]

    Viewcrafter: Tam- ing video diffusion models for high-fidelity novel view synthesis

    Wangbo Yu, Jinbo Xing, Li Yuan, Wenbo Hu, Xiaoyu Li, Zhipeng Huang, Xiangjun Gao, Tien-Tsin Wong, Ying Shan, and Yonghong Tian. Viewcrafter: Tam- ing video diffusion models for high-fidelity novel view synthesis. arXiv preprint arXiv:2409.02048, 2024. 1, 2

    2024 arXiv

  52. [60]

    Gaus- sian opacity fields: Efficient adaptive surface recon- struction in unbounded scenes

    Zehao Yu, Torsten Sattler, and Andreas Geiger. Gaus- sian opacity fields: Efficient adaptive surface recon- struction in unbounded scenes. TOG, 2024. 3, 2

    2024

  53. [61]

    Egogaussian: Dynamic scene understanding from egocentric video with 3d gaussian splatting

    Daiwei Zhang, Gengyan Li, Jiajie Li, Micka ¨el Bressieux, Otmar Hilliges, Marc Pollefeys, Luc Van Gool, and Xi Wang. Egogaussian: Dynamic scene understanding from egocentric video with 3d gaussian splatting. In 3DV, 2025. 3

    2025

  54. [62]

    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 ECCV, 2025. 1

    2025

  55. [63]

    Infusion: Inpainting 3d gaus- sians via learning depth completion from diffusion prior

    Liu Zhiheng, Ouyang Hao, Wang Qiuyu, Cheng Ka Leong, Xiao Jie, Zhu Kai, Xue Nan, Liu Yu, Shen Yujun, and Cao Yang. Infusion: Inpainting 3d gaus- sians via learning depth completion from diffusion prior. arXiv preprint arXiv:2404.11613, 2024. 3

    2024 arXiv

  56. [64]

    Cvt-xrf: Contrastive in-voxel transformer for 3d consistent radiance fields from sparse inputs

    Yingji Zhong, Lanqing Hong, Zhenguo Li, and Dan Xu. Cvt-xrf: Contrastive in-voxel transformer for 3d consistent radiance fields from sparse inputs. In CVPR, 2024. 2

    2024

  57. [65]

    Hugs: Holistic urban 3d scene understanding via gaussian splatting

    Hongyu Zhou, Jiahao Shao, Lu Xu, Dongfeng Bai, Weichao Qiu, Bingbing Liu, Yue Wang, Andreas Geiger, and Yiyi Liao. Hugs: Holistic urban 3d scene understanding via gaussian splatting. In CVPR, 2024. 3

    2024

  58. [66]

    ProPainter: Improving propagation and transformer for video inpainting

    Shangchen Zhou, Chongyi Li, Kelvin C.K Chan, and Chen Change Loy. ProPainter: Improving propagation and transformer for video inpainting. In ICCV, 2023. 2, 5

    2023

  59. [67]

    Feature 3dgs: Supercharging 3d gaussian splatting to enable distilled feature fields

    Shijie Zhou, Haoran Chang, Sicheng Jiang, Zhiwen Fan, Zehao Zhu, Dejia Xu, Pradyumna Chari, Suya You, Zhangyang Wang, and Achuta Kadambi. Feature 3dgs: Supercharging 3d gaussian splatting to enable distilled feature fields. In CVPR, 2024. 3

    2024

  60. [68]

    Ewa volume splatting

    Matthias Zwicker, Hanspeter Pfister, Jeroen Van Baar, and Markus Gross. Ewa volume splatting. In VIS,

  61. [2001]

    class-conditioned RGBD generation model

    3 F3D-Gaus: Feed-forward 3D-aware Generation on ImageNet with Cycle-Aggregative Gaussian Splatting Supplementary Material (a) G3DR + Real-ESRGAN (b) Our F3D-Gaus Figure 10. Qualitative comparison with G3DR [30]. We use Real-ESRGAN[46] to upsample G3DR’s 1282 outputs to 2562. F...

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.