Pith. sign in

REVIEW 4 major objections 6 minor 113 references

Dynamic urban scenes can be rendered from RGB plus 2D priors alone

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

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

T0 review · deepseek-v4-flash

2026-08-04 09:45 UTC pith:5VPSXR4N

load-bearing objection UGSDF reports strong rendering numbers and a genuinely new SDF+3DGS integration for dynamic urban scenes, but the without-LiDAR claim rests on an unvalidated depth prior and no geometric evaluation. the 4 major comments →

arxiv 2510.13381 v1 pith:5VPSXR4N submitted 2025-10-15 cs.CV cs.GR

Leveraging 2D Priors and SDF Guidance for Dynamic Urban Scene Rendering

Siddharth Tourani , Jayaram Reddy , Akash Kumbar , Satyajit Tourani , Nishant Goyal , Madhava Krishna , N. Dinesh Reddy , Muhammad Haris Khan This is my paper
classification cs.CV cs.GR
keywords dynamic scene rendering3D Gaussian Splattingsigned distance functionurban scenesnovel view synthesismonocular metric depthpoint trackingscene editing
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

This paper claims that photorealistic novel views of moving objects in urban scenes can be reconstructed from ordinary camera images plus two 2D priors — per-frame metric depth and 2D point tracks — without LiDAR, 3D bounding boxes, object tracklets, or body templates. The method pairs a signed distance function with 3D Gaussian splatting so each representation corrects the other: the SDF guides where Gaussians are placed and pruned, and the Gaussians' rendered depth focuses the SDF's ray sampling. Reported metrics on two standard driving datasets are at or above prior systems that use LiDAR and 3D motion annotations. If correct, this would let autonomous-driving simulators and AR applications build editable, moving-object scenes from camera footage alone.

Core claim

The central claim is that a dual representation — a neural signed distance function and a set of 3D Gaussians, jointly optimized for each dynamic object — transfers geometric accuracy to the Gaussians and rendering fidelity to the SDF, and that this coupling is strong enough to substitute for explicit 3D supervision. Motion is modeled by warping all frames to a canonical object frame using 2D pixel tracks, with Gaussian trajectories expressed as sparse combinations of learnable basis motions. Densification and pruning of Gaussians are decided by querying the SDF's zero-level set, while the SDF's ray sampling range is narrowed using the Gaussians' rasterized depth. On two standard driving ben

What carries the argument

The coupled SDF–Gaussian object representation. Each moving object is stored as a canonical set of 3D Gaussians plus a deformation-code-conditioned SDF neural field; the SDF's zero-level set acts as a geometric oracle deciding when and where to densify or prune Gaussians, and the Gaussians' rendered depth maps are fed back to shrink the SDF's per-ray sampling interval. A sparse basis-trajectory model for Gaussian motion ties the two representations together over time.

Load-bearing premise

The entire LiDAR-free pipeline rests on the monocular metric depth estimator being accurate and scale-consistent for each moving object; if that depth is biased, the canonical point cloud and every depth supervision term inherit the error, with no independent geometric checkpoint to correct it.

What would settle it

A controlled experiment: take a dynamic object sequence where monocular depth is known to fail (e.g., a textureless or motion-blurred object, or a domain far from the depth network's training data), run the LiDAR-free variant, and compare the reconstructed object geometry against a LiDAR scan of the same object. If novel-view PSNR stays high but geometry error grows relative to LiDAR, the SDF–Gaussian coupling is acting as a rendering prior rather than recovering true geometry; if both degrade, the depth prior is confirmed as load-bearing.

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

If this is right

  • If correct, LiDAR-free reconstruction pipelines for driving scenes become viable, cutting sensor cost and simplifying data collection for simulation and augmented reality.
  • The same object representation supports scene editing — removing, duplicating, or relocating dynamic objects — without retraining per edit, because each object is canonicalized into a shared frame.
  • Hard cases like non-rigid pedestrians and cyclists, which break template-free baselines, are handled by the dense SDF-guided Gaussian placement, narrowing the gap with template-based approaches.
  • The method extends to casually captured monocular footage, not just multi-camera driving rigs, so ordinary video could be turned into editable 4D scenes.

Where Pith is reading between the lines

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

  • The success of the 2D-prior substitution suggests the real bottleneck for dynamic urban scene modeling is the precision of monocular metric depth; improving that prior or adding a learned correction should transfer directly to rendering quality.
  • Because both the Gaussians and the SDF are supervised by the same depth source, a systematic depth bias might be entrenched rather than corrected — a testable extension would add sparse LiDAR-only consistency points and measure how much bias remains.
  • The ablation finding that denser representations help thin objects like pedestrians indicates an adaptive-resolution variant could allocate densification budget based on object thinness.
  • The learned sparse motion basis could enable motion editing or re-timing of objects in the rendered scene, an application the paper does not explore.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 6 minor

Summary. The paper proposes UGSDF, a hybrid representation that couples a signed distance function (SDF) deformation network with 3D Gaussian Splatting (3DGS) for modeling dynamic objects in urban scenes. The method uses per-object masks, 2D point tracks from CoTracker, and metric depth maps from UniDepth to build a canonical object scaffold, and it optionally accepts LiDAR. Motion is represented as sparse combinations of learnable basis trajectories; the SDF and the Gaussians are trained jointly, with the SDF guiding Gaussian densification/pruning and the rendered Gaussians guiding SDF ray sampling. Experiments on the Waymo NOTR subset and KITTI report state-of-the-art or near-state-of-the-art rendering metrics, including a LiDAR-free variant, and ablations show that SDF-guided densification is important. The paper also demonstrates scene editing and qualitative results on casually captured iPhone sequences.

Significance. If the claims hold, the method is a practically valuable step toward LiDAR-free, annotation-light reconstruction and novel-view synthesis of dynamic urban scenes, with potential applications in simulation and editing. The combination of SDF and 3DGS for dynamic urban objects is novel, and the paper is honest about some limitations. The authors provide detailed supplementary materials, including hyperparameters, runtime, a tracking evaluation, and qualitative comparisons. The central empirical finding that SDF-guided Gaussian distribution improves rendering is supported by the ablation in Table 4. However, the headline claims of 'geometric accuracy' and 'state-of-the-art without LiDAR' are currently under-supported: the evaluation is almost entirely in rendering metrics, the LiDAR-free pipeline depends on an unvalidated monocular depth prior, and the KITTI comparison protocol is not fully specified.

major comments (4)
  1. [Section 5, Tables 1–3; Abstract] The abstract and introduction claim enhanced 'geometric accuracy' and a 'more precise representation' from the SDF-Gaussian coupling, but the evaluation is entirely in rendering metrics (PSNR/SSIM/LPIPS) plus qualitative depth figures. No quantitative geometry metric (e.g., depth error, Chamfer distance against LiDAR on dynamic objects) is reported. Since the method's central mechanism — SDF-guided densification/pruning and Gaussian-guided SDF ray sampling — is about geometry, the geometric-accuracy claim is unsupported. Please add quantitative depth/surface evaluation, at minimum for the LiDAR variant, and report the w/o-LiDAR geometry error against LiDAR for reference.
  2. [Sec. 4.1, Eq. (6), Sec. A Eqs. (10)–(11)] The LiDAR-free pipeline's entire geometric scaffold and supervision come from UniDepth metric depth: P_o is lifted from UniDepth maps and warped with CoTracker, Eq. (6) supervises Gaussian depth with UniDepth D_t, and the SDF losses use the same D_t as ground truth. There is no independent geometric checkpoint in the w/o-LiDAR variant, and the Limitations paragraph notes sensitivity to CoTracker but not to UniDepth. If UniDepth is systematically biased on dynamic objects (scale drift, motion blur, low texture), the error propagates through canonicalization, densification (Sec. 4.5) and SDF losses. Please validate the prior: compare UniDepth depth against LiDAR on the dynamic-object masks used, or ablate by substituting LiDAR depth in the same pipeline, and report the effect on both rendering and geometry.
  3. [Sec. 5, Table 2; Supp. Table 6] The KITTI comparison is not shown to be protocol-matched. The text says 'MOT 21 sequences of the city category,' while Supp. Table 6 lists only 14 raw sequences. It is unclear whether the baseline numbers in Table 2 were re-run under exactly the same sequence split, training/test frame selection (every 8th frame), and background/sky modeling as UGSDF, or were copied from the original papers. Since the KITTI gains over OmniRe are the basis for the SOTA claim, please state the exact protocol and either re-run baselines under it or clearly mark numbers from the literature.
  4. [Tables 1–4] All rendering numbers are single-run point estimates with no standard deviations or number of seeds. Several margins are small (e.g., Waymo scene reconstruction PSNR 33.88 vs. OmniRe 33.79; Table 1), so sampling or optimization variance could change the ranking. Please report at least 3 seeds or otherwise characterize variance for the main tables.
minor comments (6)
  1. [Table 4 vs. Table 3] Table 4 'Ours (Full)' reports Vehicle SSIM 0.916, while Table 3 'UGSDF (Ours)' reports Vehicle SSIM 0.906 for the same setting; also the row label 'GRS4S' should be 'GPS4S' to match the text.
  2. [Sec. 5.2] The ablation enumeration contains a duplicated '(3)' before 'A default design choice...' — the list should be renumbered.
  3. [Fig. 7 caption] The caption contains a typo: 'varies by mehod' should be 'varies by method.'
  4. [Table 1] The input symbol '-' for S3Gaussians is undefined; please specify what inputs S3Gaussians uses (e.g., LiDAR, masks, tracklets) for a fair comparison.
  5. [Sec. 4.5 / Supp. F] The main text mentions several hyperparameters (τ_s, τ_n, τ_pr, γ, B, ε) only in the supplementary; a compact summary in the main text would help reproducibility.
  6. [Eq. (2)] The notation for the rasterizer is inconsistent: the text defines R, but Eq. (2) uses R_o(Go) without prior definition.

Circularity Check

0 steps flagged

No significant circularity: the w/o-LiDAR result is an empirical system comparison anchored on external 2D priors, not a derivation whose output is equivalent to its input.

full rationale

The paper's central claim is that its dual SDF/3DGS pipeline achieves state-of-the-art rendering without LiDAR using RGB, poses, masks, CoTracker tracks, and UniDepth metric depth. This is an empirical benchmark claim, not a derivation whose conclusion is assumed in its inputs. UniDepth is used to build the canonical object scaffold (Sec. 4.1), to supervise rendered depth in Eq. (6), and to define SDF truncation/free-space bounds in Sec. A (Eqs. 10-11). That makes the LiDAR-free variant dependent on the external depth prior's accuracy for dynamic objects, which is a genuine robustness/correctness concern and is partly acknowledged in the Limitations paragraph (sensitivity to CoTracker tracks), but it is not circularity: the method is fitting to its stated inputs, not renaming a fitted parameter as a prediction. The Sec. 4.5 loop in which Gaussian-rendered images are fed through UniDepth to guide SDF ray sampling (Eq. 8) and SDF values guide Gaussian densification/pruning (Eq. 7) is a mutual bootstrapping optimization; each representation's output is used as a training signal or sampling prior for the other, and the final rendered output is not identical to the input by construction. No load-bearing claim is justified by self-citation: the only self-citations [75,78,79] appear in the concluding future-work sentence about motion segmentation and are not used to support the method's correctness. No uniqueness theorem is imported from the authors' prior work, and no ansatz is smuggled in via self-citation. The method is evaluated against external baselines on Waymo NOTR and KITTI with standard PSNR/SSIM/LPIPS metrics. I therefore find no significant circularity; score 0.

Axiom & Free-Parameter Ledger

6 free parameters · 6 axioms · 0 invented entities

The central claim rests on external pretrained 2D priors (UniDepth, CoTracker, SAM2) and hand-set thresholds. These are not fitted to the benchmark target, but they are not independently verified for the test scenes. No new physical entities are introduced.

free parameters (6)
  • SDF proximity threshold tau_s = 0.01
    Used in Sec 4.5 to decide which voxel grids are near the surface for densification; chosen by hand.
  • minimum-Gaussian threshold tau_n = 0.02
    Densification trigger in Sec 4.5; hand-set.
  • pruning threshold tau_pr = 0.02
    SDF-guided pruning threshold in Eq. (7); hand-set.
  • sampling range scale gamma = 3
    Controls SDF ray-sampling range around Gaussian depth in Eq. (8); hand-set.
  • number of basis trajectories B
    Hyperparameter in Eq. (3)-(4); value not reported in main text.
  • SDF truncation threshold epsilon
    Defines near-surface vs free-space points in the supplementary loss definitions; value not reported.
axioms (6)
  • domain assumption UniDepth monocular metric depth is accurate and scale-consistent enough to initialize object geometry
    Sec 4.1 lifts pixels using D_t^c from UniDepth; this is the sole 3D scaffold when LiDAR is absent.
  • domain assumption CoTracker 2D tracks correspond to reliable 3D correspondences across frames and cameras
    Sec 4.1 warps point clouds to canonical frame using 2D tracks; Limitations admits tracks are sometimes inaccurate.
  • domain assumption SAM2 masks seeded by the first-frame ground-truth mask are accurate enough for object supervision
    Section 4 'Mask Generation' initializes from GT annotations and tracks with SAM2.
  • domain assumption Object motion can be represented as a linear combination of B learnable basis trajectories
    Eq. (3)-(4), inherited from prior works; if false, non-rigid articulation is under-modeled.
  • standard math Volume-rendering SDF opacity conversion (NeuS) holds for this scene configuration
    Eq. (1) from NeuS [87], used unmodified.
  • domain assumption Camera intrinsics and extrinsics are known and correct
    Inputs listed in Sec 4; no bundle adjustment is mentioned.

pith-pipeline@v1.3.0-alltime-deepseek · 20727 in / 13780 out tokens · 111528 ms · 2026-08-04T09:45:10.646237+00:00 · methodology

0 comments
read the original abstract

Dynamic scene rendering and reconstruction play a crucial role in computer vision and augmented reality. Recent methods based on 3D Gaussian Splatting (3DGS), have enabled accurate modeling of dynamic urban scenes, but for urban scenes they require both camera and LiDAR data, ground-truth 3D segmentations and motion data in the form of tracklets or pre-defined object templates such as SMPL. In this work, we explore whether a combination of 2D object agnostic priors in the form of depth and point tracking coupled with a signed distance function (SDF) representation for dynamic objects can be used to relax some of these requirements. We present a novel approach that integrates Signed Distance Functions (SDFs) with 3D Gaussian Splatting (3DGS) to create a more robust object representation by harnessing the strengths of both methods. Our unified optimization framework enhances the geometric accuracy of 3D Gaussian splatting and improves deformation modeling within the SDF, resulting in a more adaptable and precise representation. We demonstrate that our method achieves state-of-the-art performance in rendering metrics even without LiDAR data on urban scenes. When incorporating LiDAR, our approach improved further in reconstructing and generating novel views across diverse object categories, without ground-truth 3D motion annotation. Additionally, our method enables various scene editing tasks, including scene decomposition, and scene composition.

Figures

Figures reproduced from arXiv: 2510.13381 by Akash Kumbar, Jayaram Reddy, Madhava Krishna, Muhammad Haris Khan, N. Dinesh Reddy, Nishant Goyal, Satyajit Tourani, Siddharth Tourani.

Figure 1
Figure 1. Figure 1: We propose Urban Gaussians via Signed Distance Functions (UGSDF) for dynamic object modeling and rendering in urban scenes. UGSDF maintains a Signed Distance Func￾tion (SDF) and 3D Gaussian Splatting (3DGS) representation to model and render a dynamic object. It can be used for object re￾moval, scene decomposition, scene composition (inserted object indicated with green arrow) and other tasks related to si… view at source ↗
Figure 2
Figure 2. Figure 2: Overview UGSDF takes posed images, dynamic object masks, 2D tracking data, and depth maps (optionally LiDAR) as input, and outputs rendered dynamic scene. The initial model is constructed through canonicalization described in Section 4.1. A dynamic 3DGS motion representation described in Section 4.2. The object of interest is represented using a combined representation of Signed Distance Functions (SDFs) a… view at source ↗
Figure 3
Figure 3. Figure 3: SDF Deformation Network The network takes as input the observed point xi,t and outputs an SDF value Si(t) via the φsdf MLP. linear combination of learnable basis trajectories. µ(t) = µo + XB j=1 cj (t)b µ j (t) (3) q(t) = qo + XB j=1 cj (t)b R j (t) (4) Here B the number of basis trajectories is a hyper￾parameter, b µ j (t) and b R j (t) are learnable basis trajectories for the means µo ∈ R 3 and quaternio… view at source ↗
Figure 4
Figure 4. Figure 4: SDF Guidance for Gaussian Primitive Distribution Densification and Pruning of Gaussians is done by querying the SDF network. Points far away from the estimated SDF zero-level set are pruned, while sparse regions close to the zero-level set are chosen for densification. Dens. denotes densification. via UniDepth. We partition the region around the object into N3 cubic grids. The dimensions of the region arou… view at source ↗
Figure 5
Figure 5. Figure 5: Ablation Analysis Our ablation analysis confirms the need for dense representations of thin objects like pedestrians and cyclists (rows 2 and 3). Furthermore, removing SDF guidance for Gaussian primitive distribution adversely impacts rendering qual￾ity (rows 1 and 3). Datasets and Metrics We evaluate our approach on two benchmarks: the Nerf-On-The Road (NOTR) dataset [101] (a subset of the Waymo Open Data… view at source ↗
Figure 6
Figure 6. Figure 6: Qualitative comparison of baselines and our method on the NOTR Dataset with zoomed in regions We show results on some atypical objects observed in urban scenes namely: a tram and two cyclists, as well as a pedestrian. Some of these are tricky cases not always very well modeled by some or all of the baselines. In contrast UGSDF achieves a very high level of fidelity in novel view synthesis for these objects… view at source ↗
Figure 7
Figure 7. Figure 7: Image and Depth Rendering Results for the NOTR Dataset Our method is rendered without LiDAR and is compared to StreetGS and 4DGF. Even though the rendered images look similar, the depth achieved varies by mehod. Our method is able to capture the details (feet of the pedestrian) and smoothness of moving objects (cars) with greater accuracy. StreetGS cannot model pedestrians, hence it fails to render in the … view at source ↗
Figure 8
Figure 8. Figure 8: Qualitative Results on the IPhone Dataset We show the rendering results of Dynamic Gaussian Marbles [70], Shape of Motion [88] and our method on 3 sequences of the IPhone Dataset. The bounding boxes highlight regions where our method generates a more high-fidelity rendering of the scene. Citation numbers in the figure correspond to the main paper [PITH_FULL_IMAGE:figures/full_fig_p017_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Scene Editing We show original and edited pairs of images, with the region of interest highlighted in a green bounding box. (a) and (c) show the white car and van removed from the scene respectively. (b), (d) and (e) show duplicated cars and pedestrians in the scenes. (f) shows the rendered scene after removing all moving objects and the sky. Videos for the edited scenes are in the supp. video. ations for … view at source ↗

discussion (0)

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

Reference graph

Works this paper leans on

113 extracted references · 22 linked inside Pith

  1. [1]

    Particlenerf: A particle-based encoding for online neural ra- diance fields

    Jad Abou-Chakra, Feras Dayoub, and Niko S ¨underhauf. Particlenerf: A particle-based encoding for online neural ra- diance fields. InProceedings of the IEEE/CVF Winter Con- ference on Applications of Computer Vision, pages 5975– 5984, 2024. 4

    2024

  2. [2]

    Neural rgb-d surface reconstruction

    Dejan Azinovi ´c, Ricardo Martin-Brualla, Dan B Goldman, Matthias Nießner, and Justus Thies. Neural rgb-d surface reconstruction. InProceedings of the IEEE/CVF Confer- ence on Computer Vision and Pattern Recognition (CVPR), pages 6290–6301, 2022. 14

    2022

  3. [3]

    Mip-nerf: A multiscale representation for anti-aliasing neu- ral radiance fields

    Jonathan T Barron, Ben Mildenhall, Matthew Tancik, Peter Hedman, Ricardo Martin-Brualla, and Pratul P Srinivasan. Mip-nerf: A multiscale representation for anti-aliasing neu- ral radiance fields. InProceedings of the IEEE/CVF In- ternational Conference on Computer Vision, pages 5855– 5864, 2021. 2

    2021

  4. [4]

    Mip-nerf 360: Unbounded anti-aliased neural radiance fields

    Jonathan T Barron, Ben Mildenhall, Dor Verbin, Pratul P Srinivasan, and Peter Hedman. Mip-nerf 360: Unbounded anti-aliased neural radiance fields. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 5470–5479, 2022. 2

    2022

  5. [5]

    Dm-nerf: 3d scene geometry decomposition and manipulation from 2d images

    Wang Bing, Lu Chen, and Bo Yang. Dm-nerf: 3d scene geometry decomposition and manipulation from 2d images. arXiv preprint arXiv:2208.07227, 2022. 2

    Pith/arXiv arXiv 2022

  6. [6]

    Neural surface reconstruction of dy- namic scenes with monocular rgb-d camera

    Hongrui Cai, Wanquan Feng, Xuetao Feng, Yan Wang, and Juyong Zhang. Neural surface reconstruction of dy- namic scenes with monocular rgb-d camera. InThirty- sixth Conference on Neural Information Processing Sys- tems (NeurIPS), 2022. 2

    2022

  7. [7]

    Objectfusion: Multi-modal 3d object detection with object-centric fusion

    Qi Cai, Yingwei Pan, Ting Yao, Chong-Wah Ngo, and Tao Mei. Objectfusion: Multi-modal 3d object detection with object-centric fusion. InProceedings of the IEEE/CVF In- ternational Conference on Computer Vision, pages 18067– 18076, 2023. 1

    2023

  8. [8]

    Emerging properties in self-supervised vision transformers

    Mathilde Caron, Hugo Touvron, Ishan Misra, Herv ´e J´egou, Julien Mairal, Piotr Bojanowski, and Armand Joulin. Emerging properties in self-supervised vision transformers. InProceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), pages 9650–9660, 2021. 2

    2021

  9. [9]

    Segment any 3d gaussians.arXiv preprint arXiv:2312.00860, 2023

    Jiazhong Cen, Jiemin Fang, Chen Yang, Lingxi Xie, Xi- aopeng Zhang, Wei Shen, and Qi Tian. Segment any 3d gaussians.arXiv preprint arXiv:2312.00860, 2023. 2

    Pith/arXiv arXiv 2023

  10. [10]

    Segment anything in 3d with nerfs

    Jiazhong Cen, Zanwei Zhou, Jiemin Fang, Chen Yang, Wei Shen, Lingxi Xie, Dongsheng Jiang, Xiaopeng Zhang, and Qi Tian. Segment anything in 3d with nerfs. InNeurIPS,

  11. [11]

    Neusg: Neural implicit surface reconstruction with 3d gaussian splatting guidance, 2023

    Hanlin Chen, Chen Li, and Gim Hee Lee. Neusg: Neural implicit surface reconstruction with 3d gaussian splatting guidance, 2023. 2

    2023

  12. [12]

    Futr3d: A unified sensor fusion framework for 3d detection

    Xuanyao Chen, Tianyuan Zhang, Yue Wang, Yilun Wang, and Hang Zhao. Futr3d: A unified sensor fusion framework for 3d detection. Inproceedings of the IEEE/CVF con- ference on computer vision and pattern recognition, pages 172–181, 2023. 1

    2023

  13. [13]

    Periodic vibration gaussian: Dynamic urban scene reconstruction and real-time rendering.arXiv preprint arXiv:2311.18561, 2023

    Yurui Chen, Chun Gu, Junzhe Jiang, Xiatian Zhu, and Li Zhang. Periodic vibration gaussian: Dynamic urban scene reconstruction and real-time rendering.arXiv preprint arXiv:2311.18561, 2023. 1, 6, 7

    Pith/arXiv arXiv 2023

  14. [14]

    Om- nire: Omni urban scene reconstruction.arXiv preprint arXiv:2408.16760, 2024

    Ziyu Chen, Jiawei Yang, Jiahui Huang, Riccardo de Lu- tio, Janick Martinez Esturo, Boris Ivanovic, Or Litany, Zan Gojcic, Sanja Fidler, Marco Pavone, et al. Om- nire: Omni urban scene reconstruction.arXiv preprint arXiv:2408.16760, 2024. 1, 2, 6, 7, 15

    Pith/arXiv arXiv 2024

  15. [15]

    Gaus- sianpro: 3d gaussian splatting with progressive propaga- tion, 2024

    Kai Cheng, Xiaoxiao Long, Kaizhi Yang, Yao Yao, Wei Yin, Yuexin Ma, Wenping Wang, and Xuejin Chen. Gaus- sianpro: 3d gaussian splatting with progressive propaga- tion, 2024. 3, 5

    2024

  16. [16]

    Carla: An open urban driv- ing simulator

    Alexey Dosovitskiy, German Ros, Felipe Codevilla, Anto- nio Lopez, and Vladlen Koltun. Carla: An open urban driv- ing simulator. InConference on robot learning, pages 1–16. PMLR, 2017. 1

    2017

  17. [17]

    Dynamic 3d gaussian fields for urban areas

    Tobias Fischer, Jonas Kulhanek, Samuel Rota Bul `o, Lorenzo Porzi, Marc Pollefeys, and Peter Kontschieder. Dynamic 3d gaussian fields for urban areas. InThe Thirty- eighth Annual Conference on Neural Information Process- ing Systems, 2024. 6, 15

    2024

  18. [18]

    Multi-level neural scene graphs for dynamic urban environments

    Tobias Fischer, Lorenzo Porzi, Samuel Rota Bulo, Marc Pollefeys, and Peter Kontschieder. Multi-level neural scene graphs for dynamic urban environments. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 21125–21135, 2024. 1, 2, 15

    2024

  19. [19]

    Plenoxels: Radiance fields without neural networks

    Sara Fridovich-Keil, Alex Yu, Matthew Tancik, Qinhong Chen, Benjamin Recht, and Angjoo Kanazawa. Plenoxels: Radiance fields without neural networks. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 5501–5510, 2022. 2

    2022

  20. [20]

    Panoptic nerf: 3d-to-2d label transfer for panoptic urban scene segmentation

    Xiao Fu, Shangzhan Zhang, Tianrun Chen, Yichong Lu, Lanyun Zhu, Xiaowei Zhou, Andreas Geiger, and Yiyi Liao. Panoptic nerf: 3d-to-2d label transfer for panoptic urban scene segmentation. InInternational Conference on 3D Vision (3DV), 2022. 2

    2022

  21. [21]

    Monocular dynamic view synthe- sis: A reality check

    Hang Gao, Ruilong Li, Shubham Tulsiani, Bryan Russell, and Angjoo Kanazawa. Monocular dynamic view synthe- sis: A reality check. InNeurIPS, 2022. 6

    2022

  22. [22]

    Vision meets robotics: The kitti dataset.The Inter- national Journal of Robotics Research, 32(11):1231–1237,

    Andreas Geiger, Philip Lenz, Christoph Stiller, and Raquel Urtasun. Vision meets robotics: The kitti dataset.The Inter- national Journal of Robotics Research, 32(11):1231–1237,

  23. [23]

    Narayanan

    Rahul Goel, Dhawal Sirikonda, Saurabh Saini, and P.J. Narayanan. Interactive Segmentation of Radiance Fields. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2023. 2

    2023

  24. [24]

    Humans in 4d: Re- constructing and tracking humans with transformers

    Shubham Goel, Georgios Pavlakos, Jathushan Rajasegaran, Angjoo Kanazawa, and Jitendra Malik. Humans in 4d: Re- constructing and tracking humans with transformers. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 14783–14794, 2023. 7

    2023

  25. [25]

    Implicit geometric regularization for learn- ing shapes

    Amos Gropp, Lior Yariv, Niv Haim, Matan Atzmon, and Yaron Lipman. Implicit geometric regularization for learn- ing shapes. InProceedings of the 37th International Con- ference on Machine Learning, pages 3789–3799. PMLR,

  26. [26]

    Sugar: Surface- aligned gaussian splatting for efficient 3d mesh reconstruc- tion and high-quality mesh rendering

    Antoine Gu ´edon and Vincent Lepetit. Sugar: Surface- aligned gaussian splatting for efficient 3d mesh reconstruc- tion and high-quality mesh rendering. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 5354–5363, 2024. 2, 5

    2024

  27. [27]

    Streetsurf: Extending multi-view im- plicit surface reconstruction to street views.arXiv preprint arXiv:2306.04988, 2023

    Jianfei Guo, Nianchen Deng, Xinyang Li, Yeqi Bai, Bo- tian Shi, Chiyu Wang, Chenjing Ding, Dongliang Wang, and Yikang Li. Streetsurf: Extending multi-view im- plicit surface reconstruction to street views.arXiv preprint arXiv:2306.04988, 2023. 1

    Pith/arXiv arXiv 2023

  28. [28]

    Semantic any- thing in 3d gaussians.arXiv preprint arXiv:2401.17857,

    Xu Hu, Yuxi Wang, Lue Fan, Junsong Fan, Junran Peng, Zhen Lei, Qing Li, and Zhaoxiang Zhang. Semantic any- thing in 3d gaussians.arXiv preprint arXiv:2401.17857,

    Pith/arXiv arXiv

  29. [29]

    2d gaussian splatting for geometrically ac- curate radiance fields

    Binbin Huang, Zehao Yu, Anpei Chen, Andreas Geiger, and Shenghua Gao. 2d gaussian splatting for geometrically ac- curate radiance fields. InSIGGRAPH 2024 Conference Pa- pers. Association for Computing Machinery, 2024. 2

    2024

  30. [30]

    S3 gaussians: Self-supervised street gaussians for autonomous driving.arXiv preprint arXiv:2405.20323, 2024

    Nan Huang, Xiaobao Wei, Wenzhao Zheng, Pengju An, Ming Lu, Wei Zhan, Masayoshi Tomizuka, Kurt Keutzer, and Shanghang Zhang. S3 gaussians: Self-supervised street gaussians for autonomous driving.arXiv preprint arXiv:2405.20323, 2024. 1, 2, 6, 15

    Pith/arXiv arXiv 2024

  31. [31]

    Sc-gs: Sparse-controlled gaussian splatting for editable dynamic scenes.arXiv preprint arXiv:2312.14937, CVPR 2024

    Yi-Hua Huang, Yang-Tian Sun, Ziyi Yang, Xiaoyang Lyu, Yan-Pei Cao, and Xiaojuan Qi. Sc-gs: Sparse-controlled gaussian splatting for editable dynamic scenes.arXiv preprint arXiv:2312.14937, CVPR 2024. 2

    Pith/arXiv arXiv 2024

  32. [32]

    Seg- ment any 4d gaussians.arXiv preprint arXiv:2407.04504,

    Shengxiang Ji, Guanjun Wu, Jiemin Fang, Jiazhong Cen, Taoran Yi, Wenyu Liu, Qi Tian, and Xinggang Wang. Seg- ment any 4d gaussians.arXiv preprint arXiv:2407.04504,

    Pith/arXiv arXiv

  33. [33]

    Co- tracker: It is better to track together.arXiv preprint arXiv:2307.07635, 2023

    Nikita Karaev, Ignacio Rocco, Benjamin Graham, Natalia Neverova, Andrea Vedaldi, and Christian Rupprecht. Co- tracker: It is better to track together.arXiv preprint arXiv:2307.07635, 2023. 2, 3, 4, 6, 7, 8, 15, 16

    Pith/arXiv arXiv 2023

  34. [34]

    3d gaussian splatting for real-time radiance field rendering.ACM Trans

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

  35. [35]

    3d gaussian splatting for real-time radiance field rendering.ACM Transactions on Graphics, 42(4), 2023

    Bernhard Kerbl, Georgios Kopanas, Thomas Leimk ¨uhler, and George Drettakis. 3d gaussian splatting for real-time radiance field rendering.ACM Transactions on Graphics, 42(4), 2023. 4, 5, 16

    2023

  36. [36]

    Lerf: Language embed- ded radiance fields

    Justin Kerr, Chung Min Kim, Ken Goldberg, Angjoo Kanazawa, and Matthew Tancik. Lerf: Language embed- ded radiance fields. InProceedings of the IEEE/CVF In- ternational Conference on Computer Vision (ICCV), pages 19729–19739, 2023. 2

    2023

  37. [37]

    Garfield: Group anything with radiance fields

    Chung Min Kim, Mingxuan Wu, Justin Kerr, Ken Gold- berg, Matthew Tancik, and Angjoo Kanazawa. Garfield: Group anything with radiance fields. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 21530–21539, 2024. 2

    2024

  38. [38]

    Adam: A method for stochastic optimization.arXiv preprint arXiv:1412.6980,

    Diederik P Kingma and Jimmy Ba. Adam: A method for stochastic optimization.arXiv preprint arXiv:1412.6980,

    Pith/arXiv arXiv

  39. [39]

    Segment anything

    Alexander Kirillov, Eric Mintun, Nikhila Ravi, Hanzi Mao, Chloe Rolland, Laura Gustafson, Tete Xiao, Spencer Whitehead, Alexander C Berg, Wan-Yen Lo, et al. Segment anything. InProceedings of the IEEE/CVF International Conference on Computer Vision, pages 4015–4026, 2023. 2

    2023

  40. [40]

    Decomposing nerf for editing via feature field distil- lation

    Sosuke Kobayashi, Eiichi Matsumoto, and Vincent Sitz- mann. Decomposing nerf for editing via feature field distil- lation. InProceedings of the 36th International Conference on Neural Information Processing Systems, Red Hook, NY , USA, 2024. Curran Associates Inc. 2

    2024

  41. [41]

    Panoptic neu- ral fields: A semantic object-aware neural scene representa- tion

    Abhijit Kundu, Kyle Genova, Xiaoqi Yin, Alireza Fathi, Caroline Pantofaru, Leonidas J Guibas, Andrea Tagliasac- chi, Frank Dellaert, and Thomas Funkhouser. Panoptic neu- ral fields: A semantic object-aware neural scene representa- tion. InProceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition, pages 12871–12881,

  42. [42]

    Mosca: Dynamic gaussian fusion from casual videos via 4d motion scaffolds.arXiv preprint arXiv:2405.17421, 2024

    Jiahui Lei, Yijia Weng, Adam Harley, Leonidas Guibas, and Kostas Daniilidis. Mosca: Dynamic gaussian fusion from casual videos via 4d motion scaffolds.arXiv preprint arXiv:2405.17421, 2024. 2

    Pith/arXiv arXiv 2024

  43. [43]

    Language-driven semantic seg- mentation

    Boyi Li, Kilian Q Weinberger, Serge Belongie, Vladlen Koltun, and Rene Ranftl. Language-driven semantic seg- mentation. InInternational Conference on Learning Rep- resentations, 2022. 2

    2022

  44. [44]

    Neural 3d video synthesis from multi-view video

    Tianye Li, Mira Slavcheva, Michael Zollhoefer, Simon Green, Christoph Lassner, Changil Kim, Tanner Schmidt, Steven Lovegrove, Michael Goesele, Richard Newcombe, et al. Neural 3d video synthesis from multi-view video. In Proceedings of the IEEE/CVF Conference on Computer Vi- sion and Pattern Recognition, pages 5521–5531, 2022. 2

    2022

  45. [45]

    Neural scene flow fields for space-time view syn- thesis of dynamic scenes

    Zhengqi Li, Simon Niklaus, Noah Snavely, and Oliver Wang. Neural scene flow fields for space-time view syn- thesis of dynamic scenes. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 6498–6508, 2021. 2

    2021

  46. [46]

    Neuralangelo: High-fidelity neural surface reconstruc- tion

    Zhaoshuo Li, Thomas M ¨uller, Alex Evans, Russell H Tay- lor, Mathias Unberath, Ming-Yu Liu, and Chen-Hsuan Lin. Neuralangelo: High-fidelity neural surface reconstruc- tion. InIEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2023. 2

    2023

  47. [47]

    Dynibar: Neural dynamic image-based rendering

    Zhengqi Li, Qianqian Wang, Forrester Cole, Richard Tucker, and Noah Snavely. Dynibar: Neural dynamic image-based rendering. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 4273–4284, 2023. 4

    2023

  48. [48]

    Modgs: Dy- namic gaussian splatting from causually-captured monocu- lar videos.arXiv preprint arXiv:2406.00434, 2024

    Qingming Liu, Yuan Liu, Jiepeng Wang, Xianqiang Lv, Peng Wang, Wenping Wang, and Junhui Hou. Modgs: Dy- namic gaussian splatting from causually-captured monocu- lar videos.arXiv preprint arXiv:2406.00434, 2024. 2

    Pith/arXiv arXiv 2024

  49. [49]

    Dynamic 3d gaussians: Tracking by per- sistent dynamic view synthesis

    Jonathon Luiten, Georgios Kopanas, Bastian Leibe, and Deva Ramanan. Dynamic 3d gaussians: Tracking by per- sistent dynamic view synthesis. In2024 International Con- ference on 3D Vision (3DV), pages 800–809. IEEE, 2024. 2

    2024

  50. [50]

    3dgsr: Implicit surface reconstruction with 3d gaussian splatting, 2024

    Xiaoyang Lyu, Yang-Tian Sun, Yi-Hua Huang, Xiuzhe Wu, Ziyi Yang, Yilun Chen, Jiangmiao Pang, and Xiaojuan Qi. 3dgsr: Implicit surface reconstruction with 3d gaussian splatting, 2024. 2 10

    2024

  51. [51]

    Nerf in the wild: Neural radiance fields for uncon- strained photo collections

    Ricardo Martin-Brualla, Noha Radwan, Mehdi SM Sajjadi, Jonathan T Barron, Alexey Dosovitskiy, and Daniel Duck- worth. Nerf in the wild: Neural radiance fields for uncon- strained photo collections. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 7210–7219, 2021. 1

    2021

  52. [52]

    Nerf: Representing scenes as neural radiance fields for view syn- thesis.Communications of the ACM, 65(1):99–106, 2021

    Ben Mildenhall, Pratul P Srinivasan, Matthew Tancik, Jonathan T Barron, Ravi Ramamoorthi, and Ren Ng. Nerf: Representing scenes as neural radiance fields for view syn- thesis.Communications of the ACM, 65(1):99–106, 2021. 1, 2

    2021

  53. [53]

    Dynamicsurf: Dynamic neural rgb-d surface reconstruction with an opti- mizable feature grid

    Mirgahney Mohamed and Lourdes Agapito. Dynamicsurf: Dynamic neural rgb-d surface reconstruction with an opti- mizable feature grid. In2024 International Conference on 3D Vision (3DV), pages 820–830. IEEE, 2024. 4

    2024

  54. [54]

    Instant neural graphics primitives with a multiresolution hash encoding.ACM Trans

    Thomas M ¨uller, Alex Evans, Christoph Schied, and Alexander Keller. Instant neural graphics primitives with a multiresolution hash encoding.ACM Trans. Graph., 41 (4):102:1–102:15, 2022. 2, 4

    2022

  55. [55]

    Unisurf: Unifying neural implicit surfaces and radiance fields for multi-view reconstruction

    Michael Oechsle, Songyou Peng, and Andreas Geiger. Unisurf: Unifying neural implicit surfaces and radiance fields for multi-view reconstruction. InInternational Con- ference on Computer Vision (ICCV), 2021. 2

    2021

  56. [56]

    isdf: Real-time neural signed distance fields for robot perception

    Joseph Ortiz, Alexander Clegg, Jing Dong, Edgar Su- car, David Novotny, Michael Zollhoefer, and Mustafa Mukadam. isdf: Real-time neural signed distance fields for robot perception. InRobotics: Science and Systems, 2022. 5, 14

    2022

  57. [57]

    Neural scene graphs for dynamic scenes

    Julian Ost, Fahim Mannan, Nils Thuerey, Julian Knodt, and Felix Heide. Neural scene graphs for dynamic scenes. In Proceedings of the IEEE/CVF Conference on Computer Vi- sion and Pattern Recognition, pages 2856–2865, 2021. 1, 2

    2021

  58. [58]

    Simpletrack: Understanding and rethinking 3d multi-object tracking

    Ziqi Pang, Zhichao Li, and Naiyan Wang. Simpletrack: Understanding and rethinking 3d multi-object tracking. In European Conference on Computer Vision, pages 680–696. Springer, 2022. 15

    2022

  59. [59]

    Nerfies: Deformable neural radiance fields

    Keunhong Park, Utkarsh Sinha, Jonathan T Barron, Sofien Bouaziz, Dan B Goldman, Steven M Seitz, and Ricardo Martin-Brualla. Nerfies: Deformable neural radiance fields. InProceedings of the IEEE/CVF International Conference on Computer Vision, pages 5865–5874, 2021. 2

    2021

  60. [60]

    Hypernerf: A higher- dimensional representation for topologically varying neural radiance fields.arXiv preprint arXiv:2106.13228, 2021

    Keunhong Park, Utkarsh Sinha, Peter Hedman, Jonathan T Barron, Sofien Bouaziz, Dan B Goldman, Ricardo Martin- Brualla, and Steven M Seitz. Hypernerf: A higher- dimensional representation for topologically varying neural radiance fields.arXiv preprint arXiv:2106.13228, 2021. 2, 4

    Pith/arXiv arXiv 2021

  61. [61]

    Unidepth: Universal monocular metric depth estimation

    Luigi Piccinelli, Yung-Hsu Yang, Christos Sakaridis, Mat- tia Segu, Siyuan Li, Luc Van Gool, and Fisher Yu. Unidepth: Universal monocular metric depth estimation. In Proceedings of the IEEE/CVF Conference on Computer Vi- sion and Pattern Recognition, pages 10106–10116, 2024. 2, 4, 6, 7, 15

    2024

  62. [62]

    D-nerf: Neural radiance fields for dynamic scenes

    Albert Pumarola, Enric Corona, Gerard Pons-Moll, and Francesc Moreno-Noguer. D-nerf: Neural radiance fields for dynamic scenes. InProceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition, pages 10318–10327, 2021. 2

    2021

  63. [63]

    Langsplat: 3d language gaussian splat- ting.arXiv preprint arXiv:2312.16084, 2023

    Minghan Qin, Wanhua Li, Jiawei Zhou, Haoqian Wang, and Hanspeter Pfister. Langsplat: 3d language gaussian splat- ting.arXiv preprint arXiv:2312.16084, 2023. 2

    Pith/arXiv arXiv 2023

  64. [64]

    Learning transferable visual models from natural language supervision

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, Gretchen Krueger, and Ilya Sutskever. Learning transferable visual models from natural language supervision. InProceedings of the 38th International Conference on Machine Learning, pages 8748–8763. PMLR, 2021. 2

    2021

  65. [65]

    Sam 2: Segment anything in images and videos.arXiv preprint arXiv:2408.00714, 2024

    Nikhila Ravi, Valentin Gabeur, Yuan-Ting Hu, Ronghang Hu, Chaitanya Ryali, Tengyu Ma, Haitham Khedr, Ro- man R ¨adle, Chloe Rolland, Laura Gustafson, et al. Sam 2: Segment anything in images and videos.arXiv preprint arXiv:2408.00714, 2024. 3, 5, 6, 16

    Pith/arXiv arXiv 2024

  66. [66]

    Srinivasan, Dor Verbin, Richard Szeliski, Ben Mildenhall, Jonathan T

    Christian Reiser, Stephan Garbin, Pratul P. Srinivasan, Dor Verbin, Richard Szeliski, Ben Mildenhall, Jonathan T. Bar- ron, Peter Hedman, and Andreas Geiger. Binary opacity grids: Capturing fine geometric detail for mesh-based view synthesis.SIGGRAPH, 2024. 2

    2024

  67. [67]

    Urban radiance fields

    Konstantinos Rematas, Andrew Liu, Pratul P Srini- vasan, Jonathan T Barron, Andrea Tagliasacchi, Thomas Funkhouser, and Vittorio Ferrari. Urban radiance fields. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 12932–12942, 2022. 1

    2022

  68. [68]

    Gm self-driving unit cruise to pay 15 million usd fine over crash disclosure, 2024

    Reuters. Gm self-driving unit cruise to pay 15 million usd fine over crash disclosure, 2024. 1

    2024

  69. [69]

    Panoptic lifting for 3d scene understanding with neural fields

    Yawar Siddiqui, Lorenzo Porzi, Samuel Rota Bul `o, Nor- man M ¨uller, Matthias Nießner, Angela Dai, and Peter Kontschieder. Panoptic lifting for 3d scene understanding with neural fields. InProceedings of the IEEE/CVF Confer- ence on Computer Vision and Pattern Recognition (CVPR), pages 9043–9052, 2023. 2

    2023

  70. [70]

    Dynamic gaussian marbles for novel view synthe- sis of casual monocular videos

    Colton Stearns, Adam W Harley, Mikaela Uy, Florian Du- bost, Federico Tombari, Gordon Wetzstein, and Leonidas Guibas. Dynamic gaussian marbles for novel view synthe- sis of casual monocular videos. InSIGGRAPH Asia 2024 Conference Papers, pages 1–11, 2024. 15, 17

    2024

  71. [71]

    Scalability in perception for autonomous driving: Waymo open dataset

    Pei Sun, Henrik Kretzschmar, Xerxes Dotiwalla, Aure- lien Chouard, Vijaysai Patnaik, Paul Tsui, James Guo, Yin Zhou, Yuning Chai, Benjamin Caine, et al. Scalability in perception for autonomous driving: Waymo open dataset. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 2446–2454, 2020. 2, 6, 15

    2020

  72. [72]

    Block-nerf: Scalable large scene neural view synthesis

    Matthew Tancik, Vincent Casser, Xinchen Yan, Sabeek Pradhan, Ben Mildenhall, Pratul P Srinivasan, Jonathan T Barron, and Henrik Kretzschmar. Block-nerf: Scalable large scene neural view synthesis. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 8248–8258, 2022. 1

    2022

  73. [73]

    Motion planning for autonomous 11 driving: The state of the art and future perspectives

    Siyu Teng, Xuemin Hu, Peng Deng, Bai Li, Yuchen Li, Yunfeng Ai, Dongsheng Yang, Lingxi Li, Zhe Xuanyuan, Fenghua Zhu, et al. Motion planning for autonomous 11 driving: The state of the art and future perspectives. IEEE Transactions on Intelligent Vehicles, 8(6):3692–3711,

  74. [74]

    Neurad: Neural rendering for autonomous driving

    Adam Tonderski, Carl Lindstr ¨om, Georg Hess, William Ljungbergh, Lennart Svensson, and Christoffer Petersson. Neurad: Neural rendering for autonomous driving. InPro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 14895–14904, 2024. 1, 2

    2024

  75. [75]

    Using in-frame shear constraints for monocular motion segmentation of rigid bodies.Journal of Intelligent & Robotic Systems, 82 (2):237–255, 2016

    Siddharth Tourani and K Madhava Krishna. Using in-frame shear constraints for monocular motion segmentation of rigid bodies.Journal of Intelligent & Robotic Systems, 82 (2):237–255, 2016. 8

    2016

  76. [76]

    Mplp++: Fast, parallel dual block-coordinate ascent for dense graphical models

    Siddharth Tourani, Alexander Shekhovtsov, Carsten Rother, and Bogdan Savchynskyy. Mplp++: Fast, parallel dual block-coordinate ascent for dense graphical models. InProceedings of the European Conference on Computer Vision (ECCV), pages 251–267, 2018. 2

    2018

  77. [77]

    Taxonomy of dual block-coordinate ascent methods for discrete energy min- imization

    Siddharth Tourani, Alexander Shekhovtsov, Carsten Rother, and Bogdan Savchynskyy. Taxonomy of dual block-coordinate ascent methods for discrete energy min- imization. InInternational conference on artificial intelli- gence and statistics, pages 2775–2785. PMLR, 2020. 2

    2020

  78. [78]

    Discrete cycle- consistency based unsupervised deep graph matching

    Siddharth Tourani, Muhammad Haris Khan, Carsten Rother, and Bogdan Savchynskyy. Discrete cycle- consistency based unsupervised deep graph matching. In Proceedings of the AAAI Conference on Artificial Intelli- gence, pages 5252–5260, 2024. 8

    2024

  79. [79]

    Leveraging cycle-consistent anchor points for self-supervised rgb-d registration

    Siddharth Tourani, Jayaram Reddy, Sarvesh Thakur, K Madhava Krishna, Muhammad Haris Khan, and N Di- nesh Reddy. Leveraging cycle-consistent anchor points for self-supervised rgb-d registration. In2024 IEEE Inter- national Conference on Robotics and Automation (ICRA), pages 2737–2744. IEEE, 2024. 8

    2024

  80. [80]

    Non-rigid neural radiance fields: Reconstruc- tion and novel view synthesis of a dynamic scene from monocular video

    Edgar Tretschk, Ayush Tewari, Vladislav Golyanik, Michael Zollh ¨ofer, Christoph Lassner, and Christian Theobalt. Non-rigid neural radiance fields: Reconstruc- tion and novel view synthesis of a dynamic scene from monocular video. InProceedings of the IEEE/CVF In- ternational Conference on Computer Vision (ICCV), pages 12959–12970, 2021. 2

    2021

Showing first 80 references.