Sparse2DGS: Sparse-View Surface Reconstruction using 2D Gaussian Splatting with Dense Point Cloud

Hwann-Tzong Chen; Koichi Ito; Natsuki Takama; Shintaro Ito; Takafumi Aoki

REVIEW 3 major objections 4 minor 34 references

Sparse2DGS: Sparse-View Surface Reconstruction using 2D Gaussian Splatting with Dense Point Cloud

T0 review · 3 major / 4 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read Sparse2DGS reconstructs accurate 3D surfaces from just three images by initializing 2D Gaussian splatting with a dense point cloud.

desk verdict A plausible sparse-view 2DGS initialization trick that is useful but under-supported by the current ablation, which omits the isolation needed to attribute the gain to the MVS integration. read the letter →

arxiv 2505.19854 v2 pith:VQVXFV2T submitted 2025-05-26 cs.CV

Natsuki Takama , Shintaro Ito , Koichi Ito , Hwann-Tzong Chen , Takafumi Aoki This is my paper · ORCID

classification cs.CV

keywords sparse-view3Dreconstruction2DGaussiansplattingDUSt3RCOLMAPMVSpointcloudinitializationsurfacethree-view

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

The paper claims that the main obstacle to sparse-view Gaussian splatting is the sparse, unreliable point cloud that seeds the Gaussians. It proposes to replace that seed with an integrated dense point cloud built from DUSt3R's per-pixel stereo predictions and COLMAP's multi-view stereo output. With this initialization, 2D Gaussian splatting can reconstruct object surfaces from just three images, reaching a mean Chamfer Distance of 1.51 on twelve DTU scenes, matching a state-of-the-art sparse-view method that requires days of pre-training, in about 100 seconds per scene. The central load-bearing claim is that the fused dense point cloud supplies enough correct Gaussian centers to keep optimization from collapsing when views are few.

What carries the argument

The central object is the 2D Gaussian primitive from 2D Gaussian splatting: an ellipse lying in the local tangent plane of the surface, defined by a center, two tangent vectors, two scale factors, opacity, and color. The paper's carrying mechanism is the integrated dense point cloud that initializes these centers. DUSt3R produces a per-pixel point map for each input view; the paper downsamples it with voxels of size 0.005, removes statistical outliers, and aligns it to the COLMAP MVS point cloud via iterative closest point (ICP) registration. Each fused point becomes the initial position of a 2D Gaussian, so no densification is performed during optimization. The 2DGS losses—color reconstruction, depth distortion, and normal consistency—then optimize the Gaussians, and the rendered depth maps are merged by truncated signed distance function (TSDF) integration into a mesh.

What would settle it

Run the pipeline on a scene pair where the DUSt3R and COLMAP point clouds are known to be misaligned, for example a highly reflective object, and measure the Chamfer Distance; if the error jumps to or above the DUSt3R-only baseline, the alignment step is the failure point. A more direct test is to replace the ICP alignment with ground-truth alignment of the two clouds and check whether the mean Chamfer Distance improves; if it does not, misalignment is not the limiting factor.

Watch

Extended reading notes

Core claim

The discovery is that dense, accurate initialization is what makes Gaussian splatting viable in the sparse-view regime, and that the fusion of DUSt3R and COLMAP MVS provides exactly that dense initialization. The paper shows that using only the sparse COLMAP SfM point cloud, as in standard 2DGS, yields a mean Chamfer Distance of 2.54; using COLMAP MVS alone gives 1.90; using the post-processed DUSt3R cloud alone gives 1.76; and fusing both gives 1.51. The method does not densify Gaussians during optimization; it optimizes the parameters of the Gaussians placed at the dense initial points, and renders depth maps that are integrated into a mesh via TSDF fusion.

Load-bearing premise

The pipeline assumes that the ICP alignment between the DUSt3R point cloud and the COLMAP MVS point cloud converges to the correct global alignment, so that the fused Gaussian centers actually lie on the object surface.

Editorial extensions

If this is right

Three-image reconstruction is now practical without pre-training, since the method runs in about 100 seconds per scene on a single GPU.
The ablation quantifies the contribution of each point cloud source: COLMAP MVS alone lowers the mean Chamfer Distance to 1.90, post-processed DUSt3R alone to 1.76, and their fusion to 1.51.
Because no densification is used, the number of Gaussians is fixed by the fused point cloud, making the optimization faster and more deterministic than standard Gaussian splatting.
The method matches the accuracy of a pre-trained sparse-view neural surface method while requiring only three images, ground-truth camera parameters, and no offline training for each scene.

Reading between the lines

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

The same fused-initialization idea could be dropped into other Gaussian-splatting reconstruction methods, for example 3DGS-based surface extractors, to test whether the dense seed helps beyond the 2DGS formulation.
The reliance on ICP suggests a robustness bottleneck: a learned or globally consistent alignment, such as the one DUSt3R can produce for stereo pairs, might remove the need for a separate alignment step and make the method work on scenes where ICP is prone to local optima.
The method's success with three views hints that the key difficulty in sparse-view reconstruction is initialization density rather than the optimization objective itself; if so, even simpler reconstruction losses could benefit from the same dense seed.
A natural stress test would be running the method with only two input images, where the DUSt3R stereo prior is stronger and COLMAP MVS is weaker, to see whether the fusion still holds up.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Desk Editor's Note

read the letter

What you should know: this paper reports a simple, fast pipeline for three-view surface reconstruction by initializing 2DGS with a merged DUSt3R + COLMAP MVS point cloud. The idea is straightforward and likely useful, but the main ablation does not actually isolate the integration step from the post-processing, and the comparison is missing some key baselines.

What's new: the specific combination of DUSt3R and COLMAP MVS point clouds as Gaussian initialization, with a 2DGS backbone, and a quantitative result on 12 DTU scans. The speed (~100s) is attractive. The paper is honest enough to show an ablation where MVS alone and DUSt3R alone each beat COLMAP SfM, and the full method gets the best mean CD. That is reproducible and useful.

The stress-test concern is correct. Table 2's caption says 'DUSt3R (Ours)' is the post-processed version, but the table only shows the post-processed DUSt3R in the combined row, never alone. So you cannot tell whether the 0.25 mean-CD gain from 1.76 to 1.51 comes from adding COLMAP MVS points or just from voxel downsampling and outlier removal. This is a load-bearing missing control for the paper's central mechanism.

Other issues: reflective scans (69, 97, 110) are excluded, which narrows the claim; view selection is 'difficult' without a reproducible protocol (citing SPARF); UFORecon is evaluated without its random-set training strategy, possibly handicapping it; no variance or code; and pixelSplat/MVSplat are not cited, so the novelty is not sharply positioned. These are correctable.

Balanced view: the core empirical result—that dense point cloud initialization helps 2DGS from three views—is plausible and the paper shows it. The weakness is in the attribution, not in the motivating problem. The paper is short (workshop length), and the missing row is a small experimental fix.

I would send it to peer review, not desk reject. The ablation gap should be flagged as major, but the pipeline is simple enough that one missing row and a better baseline comparison could make the claim stand. I'd want the missing 'DUSt3R (Ours) alone' row, per-scan values, and ideally code. For a reading group, it's a maybe—useful as an example of a practical sparse-view system, not as a methodological breakthrough.

Referee Report

3 major / 4 minor

Summary. The paper proposes Sparse2DGS, a method for surface reconstruction from three input images. The method uses DUSt3R to generate a dense point cloud, post-processes it with voxel downsampling and statistical outlier removal, aligns it to a COLMAP MVS point cloud via ICP, and uses the merged cloud to initialize 2D Gaussian Splatting without densification. Surface meshes are obtained by rendering depth maps and fusing them with TSDF integration. On 12 selected DTU scenes, the method reports a mean Chamfer Distance of 1.51, matching UFORecon and improving over 2DGS (2.54) and raw DUSt3R initialization (2.09/1.76 depending on protocol), with about 100 seconds per scene.

Significance. If the reported results are reproducible, Sparse2DGS would be a practically attractive sparse-view surface reconstruction method: it avoids the multi-day pretraining of methods such as UFORecon, requires only three images, and reaches comparable mean Chamfer Distance on the chosen DTU subset. The paper's internal ablation in Table 2 shows a clear improvement of 2DGS with different initialization sources, and the qualitative results suggest meaningful surface completeness gains on difficult scans such as scan 24. However, the central mechanism — that integrating COLMAP MVS points with post-processed DUSt3R points is what drives the improvement — is not isolated by the reported ablation, and the benchmark protocol lacks the specificity needed to reproduce the exact numbers.

major comments (3)

[§4.5, Table 2]
[§3.2]
[§4.1, Table 1]

minor comments (4)

[§1]
[References]
[§5]
[§4.2]

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the central claim is an empirical initialization comparison against external DTU benchmarks; the missing ablation control is an experimental gap, not a derivation-level circularity.

full rationale

The paper makes an empirical claim: initializing 2DGS with a merged DUSt3R + COLMAP MVS dense point cloud improves sparse-view surface reconstruction from three images. No equation in the manuscript defines the output in terms of the input; the reported Chamfer distances are measured on held-out DTU scans against external baselines (SparseNeuS, ReTR, UFORecon, COLMAP, 2DGS, DUSt3R). DUSt3R and COLMAP outputs are external inputs, not parameters fitted to the evaluation numbers, and the 2DGS optimization follows the existing baseline formulation. The openly stated choices — voxel size 0.005 in Sec. 3.2, loss weights in Sec. 3.3, and the selection of three difficult views per scan following SPARF while excluding reflective scans 69, 97, 110 — are benchmark-specific tuning and dataset selection, not self-definitional predictions. The ablation in Table 2 is confounded: the caption defines 'DUSt3R (Ours)' as post-processed DUSt3R, yet there is no row using post-processed DUSt3R alone, so part of the 0.25 mean-CD gain from 1.76 to 1.51 could come from voxel downsampling and statistical outlier removal rather than from ICP integration with the MVS cloud. This is a legitimate experimental control gap and a correctness risk, but it is not circularity: the reported gain is not forced by construction, by a fitted equation, or by a load-bearing self-citation chain. No self-citations are load-bearing, and no uniqueness theorem or answer is imported from the authors' prior work. Therefore the score is 0, with the ablation confound noted as an experimental weakness rather than a circular step.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

The central result rests on the reliability of two external point-cloud generators (DUSt3R and COLMAP MVS), on ICP alignment between their outputs, and on the availability of ground-truth camera poses and masks. No new entities are introduced.

free parameters (4)

Voxel downsampling size = 0.005
Set empirically in Sec. 3.2; controls the density of the DUSt3R point cloud before integration. No sensitivity analysis is reported.
Statistical outlier removal parameters = not specified
Sec. 3.2 cites Open3D statistical outlier removal but gives no values for the number of neighbors or the standard deviation threshold, which affect the integrated cloud.
TSDF reconstruction parameters = not specified
TSDF integration in Sec. 3.3 is used for the final mesh, but voxel size, truncation distance, and mask handling are not stated; mesh quality depends on them.
Loss weights alpha and beta = 1000 and 0.05
Copied from 2DGS [10]; not tuned in this paper, but they are adjustable hyperparameters of the total loss in Eq. (6).

assumptions (5)

domain assumption DUSt3R produces sufficiently accurate dense point maps from three input images.
Relied on in Sec. 3.1; the entire pipeline depends on DUSt3R's geometric accuracy and point map alignment.
domain assumption COLMAP MVS produces usable dense points from only three views with ground-truth camera poses.
Sec. 3.1 combines COLMAP MVS with DUSt3R, assuming COLMAP MVS succeeds in a regime it was not designed for.
domain assumption ICP aligns the DUSt3R and COLMAP MVS point clouds into a common coordinate system.
Sec. 3.2 uses ICP without reporting alignment error or failure cases; a wrong alignment would corrupt the initialization.
domain assumption Ground-truth camera parameters are available for the three input images.
Sec. 4.2 states that ground-truth camera parameters are used, so the three-image claim assumes known poses.
domain assumption Masked TSDF integration of rendered depth maps produces an unbiased surface.
Sec. 3.3 and Sec. 4.2 reconstruct the final mesh by integrating masked depth maps, assuming the mask removes background without eroding the object surface.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Sparse2DGS: Sparse-View Surface Reconstruction using 2D Gaussian Splatting with Dense Point Cloud." pith.science (2026). https://pith.science/paper/VQVXFV2T

@misc{pith2026250519854,
  author       = {Pith},
  title        = {Pith review of: Sparse2DGS: Sparse-View Surface Reconstruction using 2D Gaussian Splatting with Dense Point Cloud},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/VQVXFV2T}},
  note         = {Machine review of arXiv:2505.19854}
}

read the original abstract

Gaussian Splatting (GS) has gained attention as a fast and effective method for novel view synthesis. It has also been applied to 3D reconstruction using multi-view images and can achieve fast and accurate 3D reconstruction. However, GS assumes that the input contains a large number of multi-view images, and therefore, the reconstruction accuracy significantly decreases when only a limited number of input images are available. One of the main reasons is the insufficient number of 3D points in the sparse point cloud obtained through Structure from Motion (SfM), which results in a poor initialization for optimizing the Gaussian primitives. We propose a new 3D reconstruction method, called Sparse2DGS, to enhance 2DGS in reconstructing objects using only three images. Sparse2DGS employs DUSt3R, a fundamental model for stereo images, along with COLMAP MVS to generate highly accurate and dense 3D point clouds, which are then used to initialize 2D Gaussians. Through experiments on the DTU dataset, we show that Sparse2DGS can accurately reconstruct the 3D shapes of objects using just three images. The project page is available at https://gsisaoki.github.io/SPARSE2DGS/

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

34 extracted references · 33 canonical work pages

[1]

There have been MVS-based methods th at use image correspondence through handcrafted features [2] as well as those leveraging deep learning techniques [3]

INTRODUCTION 3D reconstruction using multiple images captured from diff erent viewpoints, such as Multi-View Stereo (MVS), is a fundament al technique in computer vision [1], and has been applied to dig ital archiving, VR/AR, etc. There have been MVS-based methods th at use image correspondence through handcrafted features [2] as well as those leveraging ...

work page
[2]

Sparse2DGS: Sparse-View Surface Reconstruction using 2D Gaussian Splatting with Dense Point Cloud

RELATED WORK This section gives an overview of MVS, neural surface recons truc- tion, and Gaussian splatting related to this study. Multi-View Stereo — MVS reconstructs a dense 3D point cloud by estimating depth maps for each viewpoint using correspon dence between images and integrating them. The object surface is r econ- structed from the 3D point cloud...

work page Pith review arXiv 2025
[3]

We propose a 3D reconstruction method only from three input ima ges, which employs 2DGS [10] with DUSt3R [12] and COLMAP MVS [2]

METHOD As discussed above, to perform 3D reconstruction from a small num- ber of images by taking advantage of the speed and accuracy of GS, it is necessary to obtain dense and highly accurate 3D points from a small number of viewpoints for initialization of Gaussian s. We propose a 3D reconstruction method only from three input ima ges, which employs 2DG...

work page
[4]

EXPERIMENTS In this section, we evaluate the performance of the proposed method through the experiments on 3D reconstruction from three inp ut im- ages. 4.1. Datasets and Metrics In this experiment, we use the DTU dataset [13], which is a mul ti- view image dataset for 3D reconstruction. DTU consists of 15 scenes, each of which contains 49 or 64 images, g...

work page
[5]

Unlike the MV S method, which requires ground-truth depth maps during trai ning, NeuS can reconstruct object surfaces only from RGB images

minimizes the error between the rendered image and the gr ound- truth image, optimizes the SDF-based radiance ﬁelds, and re con- structs the object surface from the SDF values. Unlike the MV S method, which requires ground-truth depth maps during trai ning, NeuS can reconstruct object surfaces only from RGB images. On the other hand, it requires a long tr...

work page
[6]

reconstructs object surfaces with high accuracy by ﬁne- tuning trained models for scenes with only a few viewpoint images. R eTR

work page
[7]

and UFORecon [8] achieve more accurate reconstruction t han SparseNeuS [6] by rendering using Transformer as well as fea tures extracted from images. However, ReTR [7] and UFORecon [8] re - quire a large number of multi-view images and ground-truth d epth maps corresponding to each viewpoint for pre-training, and take sev- eral days of pre-training. Gauss...

work page
[8]

UFORe- con: Generalizable sparse-view surface reconstruction from ar- bitrary and unfavorable sets,

Y . Na, W. J. Kim, K. B. Han, S. Ha, and S.-E. Y oon, “UFORe- con: Generalizable sparse-view surface reconstruction from ar- bitrary and unfavorable sets,” IEEE Conf. Comput. Vis. Pattern Recog., pp. 5094–5104, June 2024

work page 2024

Show all 34 references

[9]

In Sparse2DGS, we utilize DUSt3R [12] and COLMAP MVS [2] to obtain the dense 3D point clouds for initializing 2D Gauss ians in 2DGS

CONCLUSION We have presented a 2DGS-based method, referred to asSparse2DGS, that requires only three input images for 3D reconstruction . In Sparse2DGS, we utilize DUSt3R [12] and COLMAP MVS [2] to obtain the dense 3D point clouds for initializing 2D Gauss ians in 2DGS. Throug...
[10]

for more details. After optimizing the radiance ﬁelds, we render the depth map for each viewpoint and integrate them to reconstruct the mes h model using Truncated Signed Distance Function (TSDF) integrati on [22]
[11]

ACKNOWLEDGMENT This work was supported in part by JSPS KAKENHI 23H00463 and 25K03131
[12]

Szeliski, Computer Vision: Algorithms and Applications , Springer-V erlag New Y ork Inc., 2022

R. Szeliski, Computer Vision: Algorithms and Applications , Springer-V erlag New Y ork Inc., 2022

2022
[13]

Pix- elwise view selection for unstructured multi-view stereo,

J. L. Sch¨ onberger, E. Zheng, M. Pollefeys, and J. Frahm, “Pix- elwise view selection for unstructured multi-view stereo, ” Eur . Conf. Comput. Vis., pp. 501–518, Oct. 2016

2016
[14]

MVSNet: Depth inference for unstructured multi-view stereo,

Y . Yao, Z. Luo, S. Li, T. Fang, and L. Quan, “MVSNet: Depth inference for unstructured multi-view stereo,” Eur . Conf. Com- put. Vis., pp. 785––801, Oct. 2018

2018
[15]

NeRF: Respresenting scenes as neural radiance ﬁelds for view synthesis,

B. Mildenhall, P . P . Srinivasan, M. Tancik, J. T. Barron, R. Ra- mamoorthi, and R. Ng, “NeRF: Respresenting scenes as neural radiance ﬁelds for view synthesis,” Eur . Conf. Comput. Vis., vol. 12346, pp. 405–421, Nov. 2020

2020
[16]

NeuS: Learning neural implicit surfaces by volume renderi ng for multi-view reconstruction,

P . Wang, L. Liu, Y . Liu, C. Theobalt, T. Komura, and W. Wang, “NeuS: Learning neural implicit surfaces by volume renderi ng for multi-view reconstruction,” Adv. Neural Inform. Process. Syst., vol. 34, pp. 27171–27183, Dec. 2021

2021
[17]

Sparseneus: Fast generalizable neural surface reconstru ction from sparse views,

X. Long, C. Lin, P . Wang, T. Komura, and W. Wang, “Sparseneus: Fast generalizable neural surface reconstru ction from sparse views,” Eur . Conf. Comput. Vis., pp. 210–227, Oct. 2022

2022
[18]

ReTR: Modeling render- ing via transformer for generalizable neural surface reconstruc- tion,

Y . Liang, H. He, and Y .-C. Chen, “ReTR: Modeling render- ing via transformer for generalizable neural surface reconstruc- tion,” Adv. Neural Inform. Process. Syst. , pp. 62332–62351, Dec. 2023

2023
[19]

3D gaussian splatting for real-time radiance ﬁeld rendering,

B. Kerbl, G. Kopanas, T. Leimk¨ uhler, and G. Drettakis, “ 3D gaussian splatting for real-time radiance ﬁeld rendering,” ACM Trans. Graph., vol. 42, no. 4, July 2023

2023
[20]

2D Gaus- sian splatting for geometrically accurate radiance ﬁelds,

B. Huang, Z. Y u, A. Chen, A. Geiger, and S. Gao, “2D Gaus- sian splatting for geometrically accurate radiance ﬁelds,” Proc. SIGGRAPH, pp. 1–11, July 2024

2024
[21]

SuGaR: Surface-aligned gau ssian splatting for efﬁcient 3D mesh reconstruction and high-qua lity mesh rendering,

A. Gu´ edon and V . Lepetit, “SuGaR: Surface-aligned gau ssian splatting for efﬁcient 3D mesh reconstruction and high-qua lity mesh rendering,” IEEE Conf. Comput. Vis. Pattern Recog., pp. 5354–5363, June 2024

2024
[22]

DUSt3R: Geometric 3D vision made easy,

Shuzhe Wang, Vincent Leroy, Y ohann Cabon, Boris Chidlovskii, and Jerome Revaud, “DUSt3R: Geometric 3D vision made easy,” IEEE Conf. Comput. Vis. Pattern Recog. , pp. 20697–20709, June 2024

2024
[23]

Large scale multi-view stereopsis evaluation,

R. Jensen, A. Dahl, G. V ogiatzis, E. Tola, and H. Aanæs, “Large scale multi-view stereopsis evaluation,” IEEE Conf. Comput. Vis. Pattern Recog., pp. 406–413, June 2014

2014
[24]

Screened Poisson surface reco n- struction,

M. Kazhdan and H. Hoppe, “Screened Poisson surface reco n- struction,” ACM Trans. Graph., vol. 32, no. 3, pp. 1–13, July 2013

2013
[25]

Structure-from-m otion revisited,

J. L. Sch¨ onberger and J.-M. Frahm, “Structure-from-m otion revisited,” IEEE Conf. Comput. Vis. Pattern Recog., pp. 4104– 4113, June 2016

2016
[26]

Cascade cost volume for high-resolution multi-view stereo and ster eo matching,

X. Gu, Z. Fan, S. Zhu, Z. Dai, F. Tan, and P . Tan, “Cascade cost volume for high-resolution multi-view stereo and ster eo matching,” IEEE Conf. Comput. Vis. Pattern Recog., pp. 2495– 2504, June 2020

2020
[27]

Visibility- aware multi-view stereo network,

J. Zhang, Y . Yao, S. Li, Z. Luo, and T. Fang, “Visibility- aware multi-view stereo network,” Brit. Mach. Vis. Conf. , pp. 1–12, Sept. 2020

2020
[28]

RC-MVSNet: Unsupervised multi-view stereo with neural rendering,

D. Chang, A. Boˇ ziˇ c, T. Zhang, Q. Yan, Y . Chen, S. S¨ usstrunk, and M. Nießner, “RC-MVSNet: Unsupervised multi-view stereo with neural rendering,” Eur . Conf. Comput. Vis. , vol. 13691, pp. 665––680, Oct. 2022

2022
[29]

GS2Mesh: Surface re- construction from Gaussian splatting via novel stereo view s,

Y . Wolf, A. Bracha, and R. Kimmel, “GS2Mesh: Surface re- construction from Gaussian splatting via novel stereo view s,” Eur . Conf. Comput. Vis., vol. 15147, pp. 207––224, Sept. 2024

2024
[30]

Open3D: A modern li- brary for 3D data processing,

Q.-Y . Zhou, J. Park, and V . Koltun, “Open3D: A modern li- brary for 3D data processing,” CoRR, vol. abs/1801.09847, Jan. 2018

2018 arXiv
[31]

A method for registration of 3 -D shapes,

P . J. Besl and N. D. McKay, “A method for registration of 3 -D shapes,” IEEE Trans. Pattern Anal. Mach. Intell. , vol. 14, no. 2, pp. 239–256, Feb. 1992

1992
[32]

Probabilistic vis- ibility for multi-view stereo,

C. Hern´ andez, G. V ogiatzis, and R. Cipolla, “Probabilistic vis- ibility for multi-view stereo,” IEEE Conf. Comput. Vis. Pattern Recog., pp. 1–8, June 2007

2007
[33]

SPARF: Neural radiance ﬁelds from sparse and noisy poses,

P . Truong, M.-J. Rakotosaona, F. Manhardt, and F. Tomba ri, “SPARF: Neural radiance ﬁelds from sparse and noisy poses,” IEEE Conf. Comput. Vis. Pattern Recog., pp. 4190–4200, June 2023

2023
[34]

Adam: A method for stochastic opti- mization,

D. Kingma and J. Ba, “Adam: A method for stochastic opti- mization,” Int. Conf. Learn. Represent., 2014

2014

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.

[1] [1]

There have been MVS-based methods th at use image correspondence through handcrafted features [2] as well as those leveraging deep learning techniques [3]

INTRODUCTION 3D reconstruction using multiple images captured from diff erent viewpoints, such as Multi-View Stereo (MVS), is a fundament al technique in computer vision [1], and has been applied to dig ital archiving, VR/AR, etc. There have been MVS-based methods th at use image correspondence through handcrafted features [2] as well as those leveraging ...

work page

[2] [2]

Sparse2DGS: Sparse-View Surface Reconstruction using 2D Gaussian Splatting with Dense Point Cloud

RELATED WORK This section gives an overview of MVS, neural surface recons truc- tion, and Gaussian splatting related to this study. Multi-View Stereo — MVS reconstructs a dense 3D point cloud by estimating depth maps for each viewpoint using correspon dence between images and integrating them. The object surface is r econ- structed from the 3D point cloud...

work page Pith review arXiv 2025

[3] [3]

We propose a 3D reconstruction method only from three input ima ges, which employs 2DGS [10] with DUSt3R [12] and COLMAP MVS [2]

METHOD As discussed above, to perform 3D reconstruction from a small num- ber of images by taking advantage of the speed and accuracy of GS, it is necessary to obtain dense and highly accurate 3D points from a small number of viewpoints for initialization of Gaussian s. We propose a 3D reconstruction method only from three input ima ges, which employs 2DG...

work page

[4] [4]

EXPERIMENTS In this section, we evaluate the performance of the proposed method through the experiments on 3D reconstruction from three inp ut im- ages. 4.1. Datasets and Metrics In this experiment, we use the DTU dataset [13], which is a mul ti- view image dataset for 3D reconstruction. DTU consists of 15 scenes, each of which contains 49 or 64 images, g...

work page

[5] [5]

Unlike the MV S method, which requires ground-truth depth maps during trai ning, NeuS can reconstruct object surfaces only from RGB images

minimizes the error between the rendered image and the gr ound- truth image, optimizes the SDF-based radiance ﬁelds, and re con- structs the object surface from the SDF values. Unlike the MV S method, which requires ground-truth depth maps during trai ning, NeuS can reconstruct object surfaces only from RGB images. On the other hand, it requires a long tr...

work page

[6] [6]

reconstructs object surfaces with high accuracy by ﬁne- tuning trained models for scenes with only a few viewpoint images. R eTR

work page

[7] [7]

and UFORecon [8] achieve more accurate reconstruction t han SparseNeuS [6] by rendering using Transformer as well as fea tures extracted from images. However, ReTR [7] and UFORecon [8] re - quire a large number of multi-view images and ground-truth d epth maps corresponding to each viewpoint for pre-training, and take sev- eral days of pre-training. Gauss...

work page

[8] [8]

UFORe- con: Generalizable sparse-view surface reconstruction from ar- bitrary and unfavorable sets,

Y . Na, W. J. Kim, K. B. Han, S. Ha, and S.-E. Y oon, “UFORe- con: Generalizable sparse-view surface reconstruction from ar- bitrary and unfavorable sets,” IEEE Conf. Comput. Vis. Pattern Recog., pp. 5094–5104, June 2024

work page 2024