Pith. sign in

REVIEW 3 major objections 7 minor 37 references

EGSRAL: An Enhanced 3D Gaussian Splatting based Renderer with Automated Labeling for Large-Scale Driving Scene

T0 review · 3 major / 7 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read EGSRAL claims a single 3D Gaussian Splatting renderer can synthesize photorealistic novel views of driving scenes and simultaneously generate the 2D/3D detection labels those views need, using only images and existing dataset annotations.

desk verdict A solid incremental contribution: the adaptor is genuinely new, rendering gains are consistent, but the 'images-only' claim is overstated and the adaptor's key assumption deserves a sharper test. read the letter →

arxiv 2412.15550 v1 pith:HPKSOOTC submitted 2024-12-20 cs.CV

Yixiong Huo , Guangfeng Jiang , Hongyang Wei , Ji Liu , Song Zhang , Han Liu , Xingliang Huang , Mingjie Lu
show 4 more authors
Jinzhang Peng Dong Li Lu Tian Emad Barsoum
This is my paper · ORCID
classification cs.CV
keywords 3DGaussianSplattingnovelviewsynthesisautonomousdrivingautolabelingdataaugmentationobjectdetectionscenereconstructioncoordinatetransformation
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

EGSRAL aims to prove that a single 3D Gaussian Splatting renderer can synthesize photorealistic novel views of large-scale driving scenes and, at the same time, generate the 2D and 3D detection annotations those views need, using only the input images and the dataset's existing annotations. This matters because prior driving-scene renderers require extra supervision—depth maps, LiDAR, object trajectories, or per-object models—and the frames they synthesize lack labels, so they cannot directly feed downstream detectors. The paper reports the highest rendering quality among the methods it compares against, with PSNR 29.04 on the nuScenes scenes used by DrivingGaussian, and shows that augmenting a detector's training data with its auto-labeled novel views improves both 2D and 3D detection accuracy. The load-bearing novelty is a trainable adaptor that learns the coordinate transformation between the dataset's world frame and the structure-from-motion frame, letting existing boxes be re-projected onto any novel pose.

What carries the argument

The machinery that carries the argument is the adaptor trained under three losses enforcing the same projective geometry in two coordinate systems: a pose loss aligning known original-world to SfM pose pairs, a 3D relative-position loss that leverages the similarity of real and SfM intrinsics to pin down scale, and a pixel-projection loss that keeps novel poses consistent when neighboring frames are projected onto their image planes. Supporting it are the deformation enhancement and opacity enhancement modules, which use the Gaussian's state attribute and time encoding to refine dynamic-object modeling, and the grouping strategy, which labels each Gaussian with a group id so that rendering uses only nearby primitives.

What would settle it

Measure, per nuScenes scene, the pixel reprojection error between the SfM-estimated intrinsics and the dataset's reported intrinsics on known 3D points; if that error exceeds a few pixels, the projection-consistency constraint used to train the adaptor is violated and auto-labeled boxes should drift. A direct experiment: deliberately perturb the SfM intrinsics by a fixed amount, retrain the adaptor, and check whether downstream 3D detection AP falls in proportion to the perturbation.

Watch

Extended reading notes

Core claim

The central claim is that an enhanced 3D Gaussian Splatting model, trained from images alone, can serve as a labeled data factory for autonomous driving: every novel view it renders comes with correct 3D bounding boxes in the dataset's original coordinate system. On the rendering side, the paper augments the Deformable 3D GS baseline with a per-Gaussian state attribute that tells the deformation network whether a primitive is dynamic or static, a deformation enhancement module and an opacity enhancement module that refine the deformation field and opacity, and a grouping strategy that assigns each Gaussian a group id so rendering excludes distant primitives outside the current view's field. On the labeling side, it introduces an adaptor—an 8-layer MLP that maps camera poses from the original world coordinate system to the SfM coordinate system—trained with three constraints: pose matching on known frame pairs, 3D relative-pose consistency, and pixel projection consistency for novel poses. The paper argues that this single-model design not only beats composite approaches like DrivingGaussian on rendering metrics but also produces annotations accurate enough to improve 2D and 3D detectors when used as data augmentation.

Load-bearing premise

The adaptor's training assumes the camera intrinsics estimated by structure-from-motion are close enough to the dataset's real intrinsics that the same 3D points project to the same pixels in both coordinate systems; if that similarity fails, the novel-pose labels will be misaligned.

Editorial extensions

If this is right

  • Synthetic frames for autonomous driving can be produced on demand with labels, removing the need to annotate every rendered image by hand.
  • A single model replaces the composite foreground/background pipelines of earlier driving 3D GS methods, cutting training and rendering overhead.
  • Downstream 2D and 3D detectors improve when auto-labeled novel views are added to their training data, suggesting synthetic data from the renderer can substitute for additional real labeled frames.
  • The grouping strategy extends 3D Gaussian Splatting to long, large-scale sequences by bounding each view's field of view, which should carry over to other outdoor reconstruction tasks.
  • Because the method requires only images plus existing 3D annotations and poses, it applies to any driving dataset that already carries such annotations.

Reading between the lines

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

  • Beyond the paper: if the adaptor's projection-consistency training generalizes, the same approach could replace hand-tuned coordinate alignment in other reconstruction pipelines that need to fuse dataset annotations with SfM geometry.
  • Beyond the paper: a testable extension is to feed the adaptor camera poses from fisheye or non-pinhole cameras; the paper's assumption of similar intrinsics would break, and the projection constraints would need a different formulation.
  • Beyond the paper: the reported detection gains compare auto-labeled images against the baseline dataset; isolating labeling error from rendering quality would require comparing against manually labeled versions of the same novel views.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 7 minor

Summary. The paper proposes EGSRAL, an enhanced 3D Gaussian Splatting renderer for large-scale driving scenes that combines a deformation enhancement module, an opacity enhancement module, and a grouping strategy, together with an adaptor network that maps camera poses from the original world coordinate system (OWCS) to the SfM coordinate system (EWCS) so that existing 3D annotations can be transformed to novel views. The adaptor is trained with three losses: pose matching on existing pairs, projection consistency, and a 3D coordinate constraint for randomly sampled nearby poses. Experiments on KITTI and nuScenes report improved rendering metrics over prior 3D GS methods, and detection experiments show that augmenting training data with synthesized labeled views improves Co-DETR and MonoLSS performance.

Significance. If the rendering claims hold, the proposed modules and grouping strategy are simple and effective improvements for dynamic driving scenes, and single-model training without per-object models is a practical advantage. The auto-labeling adaptor is the most novel component and has the potential to turn 3D GS renderers into data generators for perception tasks. The paper ships a code link, reports per-scene results, and includes ablations for module contributions and group counts. However, the central auto-labeling claim is validated only indirectly, as detailed below; the significance of the contribution depends on whether the adaptor truly generalizes to novel poses.

major comments (3)
  1. [Section 3.3, Eq. (11); Section 8.3, Table 15] The L3D loss for novel poses assumes that the real and SfM-estimated camera intrinsics are similar enough to treat the 3D positions of the following camera poses in the camera coordinate system as identical (text before Eq. (11)). This assumption is load-bearing because it defines the pseudo-ground truth for the novel-pose constraint, yet no experiment directly measures adaptor pose error or label accuracy on the RPT-sampled novel poses that are used for auto-labeling. Table 15 evaluates the adaptor against Umeyama-based matrix estimation on a held-out subset of existing poses transformed to OWCS, which does not test generalization to nearby novel poses; the detection gains in Tables 4, 5, 16, and 17 cannot isolate label correctness from image diversity or model regularization. Please add a direct evaluation of novel-pose prediction accuracy (e.g., comparing RPT-sampled poses to SfM-reconstructed poses for a held-out frame, or measuring 3D box center error/IoU on manually labeled novel views) and clarify whether the L3D weight w2 is scheduled as the text suggests or fixed at 0.1 throughout.
  2. [Section 7.1 (NuScenes-D); Abstract] The abstract and introduction claim the method 'relies solely on training images without extra annotations,' but the nuScenes-D benchmark explicitly initializes 3D Gaussians from LiDAR point clouds ('we utilize LiDAR point clouds to initialize the 3D Gaussians'). This contradiction affects a headline claim and the fairness of the comparison in Table 2, where DrivingGaussian also uses LiDAR. Please either revise the claim to state that the method requires only images and can optionally use LiDAR when available, or repeat the nuScenes-D experiment with SfM initialization and report both configurations.
  3. [Section 4.2; Section 8.3] The adaptor training protocol is not specified clearly: the text in Section 4.2 can be read as training one adaptor per scene (17 scenes, each with ~230 pose pairs, 1000 epochs each), which would be a significant practical limitation for auto-labeling at scale, whereas a single shared adaptor would make the method more generally applicable. The 34-scene extension in Section 8.3 should state whether additional adaptors are trained for the new scenes and how the per-scene SfM is obtained. Please specify the protocol, the total training cost, and whether the adaptor is applied to scenes outside its training distribution.
minor comments (7)
  1. [Section 3.2, Eq. (3)] The notation 'd ∈ R^{d×1}' uses d for both the state attribute and its dimensionality; rename the dimension, e.g., D, to avoid confusion.
  2. [Table 1] The 'Supervision' column labels 3D GS, Mip-splatting, Deformable 3D GS, and EGSRAL as 'RGB', but Section 3.1 states that the method takes SfM point clouds as input; the column should read 'RGB + SfM points' or be annotated to distinguish annotation supervision from input modalities, for consistency with Table 2.
  3. [Section 3.3, Eq. (7)] There is a typo: 'f ramen+i' should be 'frame n+i'.
  4. [Tables 4 and 5] The column header 'Total amount' is ambiguous because the dataset type (sample/all) and the augmentation multiplier (1×/2×/3×) are conflated; please split these into separate columns or use clearer labels.
  5. [Section 8.3] The statement that 'the SfM method used in 3D GS is monocular' is confusing in the context of the nuScenes-D experiment, which uses synchronized 6-camera input; clarify which SfM configuration is used for each dataset.
  6. [Appendix Section 8.4 (Limitations)] The limitations paragraph only mentions dependence on SfM point cloud quality; it should also acknowledge the adaptor's reliance on the intrinsics-similarity assumption and the lack of direct label-accuracy evaluation.
  7. [References] The reference list contains duplicate entries for Zong et al. 2023 (2023a and 2023b with different title capitalization); please unify.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: rendering and detection are evaluated against held-out data, and the adaptor's self-supervised losses do not use the labels it is said to predict.

full rationale

I walked the claimed derivation chain. The rendering contribution trains a 3D GS model with SfM-initialized Gaussians and posed images, using photometric losses against real training frames; evaluation (Tables 1-3) is on held-out frames of KITTI and nuScenes, so the PSNR/LPIPS/SSIM claims are not constructed from their inputs. The adaptor is a learned OWCS-to-EWCS pose mapper. Its first loss Lp (Eq. 6) is a supervised regression to existing SfM pose pairs; the novel-pose losses Lproj and L3D (Eqs. 10-11) compute pseudo-ground-truth pixel and 3D coordinates from the OWCS camera poses and the known dataset intrinsics, then compare them with the same poses projected through the adaptor's predicted novel pose. The target is therefore a function of the OWCS geometry, not of the adaptor's own output, so this is self-supervised geometric consistency rather than a prediction that reduces to a fitted input. At inference, annotations are produced by applying the same RPT affine transformation to the existing OWCS boxes that is applied to the sampled pose; this is annotation propagation through a known transform, and the adaptor only supplies the corresponding EWCS pose for rendering. The claimed downstream value is measured by training Co-DETR and MonoLSS on the augmented data and testing on real nuScenes validation scenes (Tables 4-5, 16-17), so the detection gain is an external benchmark result, not a restatement of the augmentation recipe. The load-bearing approximation that SfM and real intrinsics are similar enough to share camera-coordinate positions during early adaptor training (Sec. 3.3, before Eq. 11) is an empirical assumption and a limitation, and the paper itself acknowledges SfM-dependence in Sec. 8.4; however, this is an accuracy/robustness risk, not a circularity, because the adaptor is still evaluated on held-out pose/box accuracy in Table 15. No load-bearing self-citation or imported uniqueness theorem appears. Hyperparameters may have been tuned on the same benchmarks, which can inflate expected performance, but that is an overfitting concern outside the definition of circularity used here.

Assumptions & free parameters 5 free parameters · 6 assumptions · 1 invented entities

The central claims rest on learned networks rather than fitted constants, but several hand-chosen hyperparameters and geometric assumptions are load-bearing. The free parameters are dominated by grouping and adaptor settings; the axioms are the usual SfM and rigid-scene assumptions plus the paper-specific grouping heuristic.

free parameters (5)
  • Group number Ng = 8 (tuned on KITTI City)
    Chosen by ablation on KITTI City; Table 13 shows 8 groups gives PSNR 23.60 vs 22.64 for 4, 23.10 for 6, 23.38 for 10. This tuning uses the evaluation test set.
  • Adaptor camera pose count N = 15
    Selected in Table 14 by comparing AP3D on the nuScenes detection test set (10, 15, 20); 15 gives the best AP3d 19.78.
  • Adaptor loss weights w1, w2, w3 = 50, 0.1, 1
    Reported in Section 4.2; hand-chosen, not ablated.
  • Group valid distance threshold d = not reported
    Algorithm 1 takes a valid distance d to assign Gaussian group ids; the paper never gives its value or how it is set.
  • Overlap training image count No = not reported
    Algorithm 1 uses No images from the previous group for overlap training; the value is not specified.
assumptions (6)
  • domain assumption SfM point cloud and camera poses are sufficiently accurate for 3D GS initialization and coordinate alignment.
    Used throughout Section 3.1 and 3.3; the renderer is initialized from SfM points and the adaptor is trained against SfM poses. The paper's own limitations section acknowledges reconstruction quality depends on SfM point cloud accuracy.
  • domain assumption SfM-estimated camera intrinsics equal the real camera intrinsics, at least during initial adaptor training.
    Section 3.3: 'Given the similarity between real camera intrinsic parameters and those estimated by SfM, it is possible to assume identical position information in the camera coordinate system during the initial stages of adaptor training.' The projection consistency constraints depend on this.
  • standard math The projection consistency rule: corresponding 3D points in the two coordinate systems project to the same pixel coordinates under the novel camera pose.
    Used to construct Lproj (Eq. 10) and L3D (Eq. 11); this is a geometric identity if poses and intrinsics are correct.
  • domain assumption Applying the same affine random transformation to a camera pose and to the dataset's 3D annotations yields correct labels for the novel view.
    Section 3.3 'Annotation Generation': Pnov is sampled by affine random transformation and annsnov is obtained by applying the same transformation; assumes rigid scene and correct pose-annotation correspondence.
  • ad hoc to paper Grouping Gaussian primitives by fixed image intervals and a distance threshold removes only occluded or irrelevant distant Gaussians.
    Section 3.2 and Algorithm 1; the paper motivates GPS by 'unreasonable fields of view' but provides no geometric guarantee that the mask excludes exactly the irrelevant primitives.
  • domain assumption A unified deformation field with a learned per-primitive state attribute can represent both static background and dynamic objects in driving scenes.
    Section 3.2, Eq. 2; the baseline Deformable 3D GS uses a single deformation field, and the paper augments it with state attribute d instead of separating foreground and background models like DrivingGaussian.
invented entities (1)
  • Per-primitive state attribute d
    purpose: Learned latent vector for each Gaussian indicating static or dynamic status; used to condition deformation and opacity adjustments (Eqs. 2-4).
    No external handle; it is an internal latent variable trained with the rendering loss, so it cannot be checked independently of the paper's own renderer.

how reviews work

0 comments
Cite this review

Pith. "Pith review of EGSRAL: An Enhanced 3D Gaussian Splatting based Renderer with Automated Labeling for Large-Scale Driving Scene." pith.science (2026). https://pith.science/paper/HPKSOOTC

@misc{pith2026241215550,
  author       = {Pith},
  title        = {Pith review of: EGSRAL: An Enhanced 3D Gaussian Splatting based Renderer with Automated Labeling for Large-Scale Driving Scene},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/HPKSOOTC}},
  note         = {Machine review of arXiv:2412.15550}
}
read the original abstract

3D Gaussian Splatting (3D GS) has gained popularity due to its faster rendering speed and high-quality novel view synthesis. Some researchers have explored using 3D GS for reconstructing driving scenes. However, these methods often rely on various data types, such as depth maps, 3D boxes, and trajectories of moving objects. Additionally, the lack of annotations for synthesized images limits their direct application in downstream tasks. To address these issues, we propose EGSRAL, a 3D GS-based method that relies solely on training images without extra annotations. EGSRAL enhances 3D GS's capability to model both dynamic objects and static backgrounds and introduces a novel adaptor for auto labeling, generating corresponding annotations based on existing annotations. We also propose a grouping strategy for vanilla 3D GS to address perspective issues in rendering large-scale, complex scenes. Our method achieves state-of-the-art performance on multiple datasets without any extra annotation. For example, the PSNR metric reaches 29.04 on the nuScenes dataset. Moreover, our automated labeling can significantly improve the performance of 2D/3D detection tasks. Code is available at https://github.com/jiangxb98/EGSRAL.

Figures

Figures reproduced from arXiv: 2412.15550 by the authors.

Figure 1
Figure 1. Illustration of overall EGSRAL. The EGSRAL framework begins by aligning the input image, followed by initial [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Illustration of the grouping strategy. Opacity Enhancement Module (OEM). As seen in the im￾age rendering formula (Equation 1), the rendering of the cur￾rent pixel depends on the color and opacity of the Gaussian primitive. To enhance the capacity of opacity prediction, we initialize the Gaussian primitive by initializing opacity to a trainable parameter σ ′ ∈ R16×1 and introduce a lightweight network to accelerate o… view at source ↗
Figure 4
Figure 4. Qualitative comparison of novel view synthesis on [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗
Figures from the paper (6 more)
Figure 5
Figure 5. Figure 5: Visualizing 2D/3D auto labeling on nuScenes. [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 6
Figure 6. Figure 6: Field of view comparison of large-scale scenes. [PITH_FULL_IMAGE:figures/full_fig_p010_6.png]
Figure 7
Figure 7. Figure 7: 3D boxes projection quality comparison. Metric Matrix Adaptor AP (%) 21.71 72.53 AD (m) 1.867 0.605 [PITH_FULL_IMAGE:figures/full_fig_p011_7.png]
Figure 8
Figure 8. Figure 8: Qualitative comparison of novel view synthesis on the KITTI City dataset. [PITH_FULL_IMAGE:figures/full_fig_p012_8.png]
Figure 9
Figure 9. Figure 9: 3D auto labeling result. results are shown in [PITH_FULL_IMAGE:figures/full_fig_p012_9.png]
Figure 10
Figure 10. Figure 10: 2D auto labeling result [PITH_FULL_IMAGE:figures/full_fig_p012_10.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

37 extracted references · 34 canonical work pages

  1. [1]

    , " * write output.state after.block = add.period write newline

    ENTRY address archivePrefix author booktitle chapter edition editor eid eprint howpublished institution isbn journal key month note number organization pages publisher school series title type volume year label extra.label sort.label short.list INTEGERS output.state before.all mid.sentence after.sentence after.block FUNCTION init.state.consts #0 'before.a...

  2. [2]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in capitalize " " * FUNCT...

  3. [3]

    Agisoft. 2019. Agisoft: Metashape software]. retrieved 20.05.2019 (2019)

    work page 2019

  4. [4]

    H.; Vora, S.; Liong, V

    Caesar, H.; Bankiti, V.; Lang, A. H.; Vora, S.; Liong, V. E.; Xu, Q.; Krishnan, A.; Pan, Y.; Baldan, G.; and Beijbom, O. 2020. nuscenes: A multimodal dataset for autonomous driving. In CVPR, 11621--11631

    2020

  5. [5]

    Cao, A.; and Johnson, J. 2023. Hexplane: A fast representation for dynamic scenes. In CVPR, 130--141

    work page 2023

  6. [6]

    Chang, D.; Bo z i c , A.; Zhang, T.; Yan, Q.; Chen, Y.; S \"u sstrunk, S.; and Nie ner, M. 2022. RC-MVSNet: unsupervised multi-view stereo with neural rendering. In ECCV, 665--680. Springer

    work page 2022

  7. [7]

    Chen, S.-C. 2019. Multimedia for autonomous driving. IEEE MultiMedia, 5--8

    work page 2019

  8. [8]

    Fu, X.; Zhang, S.; Chen, T.; Lu, Y.; Zhu, L.; Zhou, X.; Geiger, A.; and Liao, Y. 2022. Panoptic nerf: 3d-to-2d label transfer for panoptic urban scene segmentation. In 3DV, 1--11. IEEE

    work page 2022

Show all 37 references
  1. [9]

    Ge, Y.; Behl, H.; Xu, J.; Gunasekar, S.; Joshi, N.; Song, Y.; Wang, X.; Itti, L.; and Vineet, V. 2022. Neural-sim: Learning to generate training data with nerf. In ECCV, 477--493. Springer

    2022

  2. [10]

    Geiger, A.; Lenz, P.; and Urtasun, R. 2012. Are we ready for autonomous driving? the kitti vision benchmark suite. In CVPR, 3354--3361. IEEE

    2012

  3. [11]

    Girshick, R. 2015. Fast r-cnn. In ICCV, 1440--1448

    2015

  4. [12]

    Isola, P.; Zhu, J.-Y.; Zhou, T.; and Efros, A. A. 2017. Image-to-image translation with conditional adversarial networks. In CVPR, 1125--1134

    2017

  5. [13]

    Kerbl, B.; Kopanas, G.; Leimk \"u hler, T.; and Drettakis, G. 2023. 3D Gaussian Splatting for Real-Time Radiance Field Rendering. ACM Trans. Graph., 139--1

    2023

  6. [14]

    J.; Tagliasacchi, A.; Dellaert, F.; and Funkhouser, T

    Kundu, A.; Genova, K.; Yin, X.; Fathi, A.; Pantofaru, C.; Guibas, L. J.; Tagliasacchi, A.; Dellaert, F.; and Funkhouser, T. 2022. Panoptic neural fields: A semantic object-aware neural scene representation. In CVPR, 12871--12881

    2022

  7. [15]

    Li, Z.; Jia, J.; and Shi, Y. 2024. MonoLSS: Learnable Sample Selection For Monocular 3D Detection. In 3DV, 1125--1135. IEEE

    2024

  8. [16]

    Li, Z.; Li, L.; and Zhu, J. 2023. Read: Large-scale neural scene rendering for autonomous driving. In AAAI, 1522--1529

    2023

  9. [17]

    Li, Z.; Wu, C.; Zhang, L.; and Zhu, J. 2023. DGNR: Density-Guided Neural Point Rendering of Large Driving Scenes

    2023

  10. [18]

    P.; Tancik, M.; Barron, J

    Mildenhall, B.; Srinivasan, P. P.; Tancik, M.; Barron, J. T.; Ramamoorthi, R.; and Ng, R. 2021. Nerf: Representing scenes as neural radiance fields for view synthesis. Communications of the ACM, 99--106

    2021

  11. [19]

    L.; and Frahm, J.-M

    Schonberger, J. L.; and Frahm, J.-M. 2016. Structure-from-motion revisited. In CVPR, 4104--4113

    2016

  12. [20]

    Sun, J.; Jiao, H.; Li, G.; Zhang, Z.; Zhao, L.; and Xing, W. 2024. 3dgstream: On-the-fly training of 3d gaussians for efficient streaming of photo-realistic free-viewpoint videos. In CVPR, 20675--20685

    2024

  13. [21]

    Tong, W.; Xie, J.; Li, T.; Deng, H.; Geng, X.; Zhou, R.; Yang, D.; Dai, B.; Lu, L.; and Li, H. 2023. 3D Data Augmentation for Driving Scenes on Camera

    2023

  14. [22]

    Tosi, F.; Tonioni, A.; De Gregorio, D.; and Poggi, M. 2023. NeRF-Supervised Deep Stereo. In CVPR, 855--866

    2023

  15. [23]

    Umeyama, S. 1991. Least-squares estimation of transformation parameters between two point patterns. IEEE Transactions on Pattern Analysis & Machine Intelligence, 376--380

    1991

  16. [24]

    Wang, T.-C.; Liu, M.-Y.; Zhu, J.-Y.; Tao, A.; Kautz, J.; and Catanzaro, B. 2018. High-resolution image synthesis and semantic manipulation with conditional gans. In CVPR, 8798--8807

    2018

  17. [25]

    Wu, G.; Yi, T.; Fang, J.; Xie, L.; Zhang, X.; Wei, W.; Liu, W.; Tian, Q.; and Wang, X. 2024. 4d gaussian splatting for real-time dynamic scene rendering. In CVPR, 20310--20320

    2024

  18. [26]

    Wu, Z.; Liu, T.; Luo, L.; Zhong, Z.; Chen, J.; Xiao, H.; Hou, C.; Lou, H.; Chen, Y.; Yang, R.; et al. 2023. Mars: An instance-aware, modular and realistic simulator for autonomous driving

    2023

  19. [27]

    Xie, Z.; Zhang, J.; Li, W.; Zhang, F.; and Zhang, L. 2023. S-nerf: Neural radiance fields for street views

    2023

  20. [28]

    Xu, C.; Wu, B.; Hou, J.; Tsai, S.; Li, R.; Wang, J.; Zhan, W.; He, Z.; Vajda, P.; Keutzer, K.; et al. 2023. Nerf-det: Learning geometry-aware volumetric representation for multi-view 3d object detection. In ICCV, 23320--23330

    2023

  21. [29]

    Yan, Y.; Lin, H.; Zhou, C.; Wang, W.; Sun, H.; Zhan, K.; Lang, X.; Zhou, X.; and Peng, S. 2024. Street gaussians for modeling dynamic urban scenes

    2024

  22. [30]

    Yang, Z.; Gao, X.; Zhou, W.; Jiao, S.; Zhang, Y.; and Jin, X. 2024. Deformable 3d gaussians for high-fidelity monocular dynamic scene reconstruction. In CVPR, 20331--20341

    2024

  23. [31]

    Yifan, W.; Serena, F.; Wu, S.; \"O ztireli, C.; and Sorkine-Hornung, O. 2019. Differentiable surface splatting for point-based geometry processing. TOG, 1--14

    2019

  24. [32]

    Yu, Z.; Chen, A.; Huang, B.; Sattler, T.; and Geiger, A. 2024. Mip-Splatting: Alias-free 3D Gaussian Splatting. In CVPR, 19447--19456

    2024

  25. [33]

    Zhang, J.; Zhang, F.; Kuang, S.; and Zhang, L. 2023. NeRF-LiDAR: Generating Realistic LiDAR Point Clouds with Neural Radiance Fields

    2023

  26. [34]

    Zhi, S.; Laidlow, T.; Leutenegger, S.; and Davison, A. J. 2021. In-place scene labelling and understanding with implicit scene representation. In ICCV, 15838--15847

    2021

  27. [35]

    Zhou, X.; Lin, Z.; Shan, X.; Wang, Y.; Sun, D.; and Yang, M.-H. 2024. Drivinggaussian: Composite gaussian splatting for surrounding dynamic autonomous driving scenes. In CVPR, 21634--21643

    2024

  28. [36]

    Zong, Z.; Song, G.; and Liu, Y. 2023 a . DETRs with Collaborative Hybrid Assignments Training. In ICCV, 6748--6758

    2023

  29. [37]

    Zong, Z.; Song, G.; and Liu, Y. 2023 b . Detrs with collaborative hybrid assignments training. In Proceedings of the IEEE/CVF international conference on computer vision, 6748--6758

    2023

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.