SLAM&Render: A Benchmark for the Intersection Between Neural Rendering, Gaussian Splatting and SLAM

Gaetano Meli; Javier Civera; Kirill Safronov; Samuel Cerezo; Tom\'as Berriel Martins

arxiv: 2504.13713 · v6 · submitted 2025-04-18 · 💻 cs.RO · cs.CV

SLAM&Render: A Benchmark for the Intersection Between Neural Rendering, Gaussian Splatting and SLAM

Samuel Cerezo , Gaetano Meli , Tom\'as Berriel Martins , Kirill Safronov , Javier Civera This is my paper

Pith reviewed 2026-05-22 18:55 UTC · model grok-4.3

classification 💻 cs.RO cs.CV

keywords SLAMNeural RenderingGaussian SplattingBenchmark DatasetRobot KinematicsRGB-DNovel View SynthesisLighting Conditions

0 comments

The pith

SLAM&Render supplies robot-recorded multi-modal sequences to test combined SLAM and neural rendering methods under controlled variations.

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

The paper introduces SLAM&Render, a dataset of 40 sequences recorded with a robot manipulator that supplies time-synchronized RGB-D images, IMU readings, kinematic data, and ground-truth poses. It targets gaps in prior resources by supporting sequential SLAM operations, multi-modal inputs, and rendering tests for viewpoint and illumination generalization while making sensor motion exactly reproducible. Existing datasets often rely on handheld or drone sensors that hinder precise motion replication and lack separate training and test paths across lighting changes and object rearrangements. If the dataset captures the intended challenges, integrated SLAM-plus-rendering pipelines could be developed and compared more reliably for robotic use.

Core claim

Existing datasets for SLAM and novel view synthesis lack sequential processing, multi-modality, controlled generalization tests across viewpoints and lighting, and exact sensor motion reproduction, so SLAM&Render records 40 sequences with a robot manipulator across five setups of consumer and industrial objects under four lighting conditions, each with separate training and test trajectories in static scenes that include different levels of rearrangements and occlusions, plus time-synchronized RGB-D, IMU, kinematic, and ground-truth pose data.

What carries the argument

Robot kinematic data streams that enable exact reproduction of sensor trajectories and direct assessment of SLAM methods inside robotic control loops.

If this is right

SLAM methods can be evaluated for sequential mapping performance using the time-synchronized streams.
Neural rendering and Gaussian Splatting techniques can be measured for robustness to lighting and viewpoint changes with separate test trajectories.
Integrations of SLAM into robotic systems become testable through the released kinematic data.
Occlusion and rearrangement effects on reconstruction quality can be isolated across the four lighting conditions.

Where Pith is reading between the lines

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

The kinematic data could support closed-loop simulation of robot paths to test how well rendering affects localization feedback.
Industrial objects in the setups open direct comparisons between consumer-grade and factory-grade scene reconstruction under the same motion paths.
Future work could add slight scene motion to the static sequences to check whether the benchmark still ranks methods consistently.

Load-bearing premise

Laboratory setups with controlled lighting, static scenes, consumer and industrial objects, and robot-recorded trajectories capture the main real-world difficulties that SLAM and neural rendering methods must overcome.

What would settle it

If standard SLAM and Gaussian Splatting baselines achieve similar generalization and accuracy scores on SLAM&Render as on existing handheld datasets, the new benchmark would not have introduced meaningfully harder or more representative conditions.

Figures

Figures reproduced from arXiv: 2504.13713 by Gaetano Meli, Javier Civera, Kirill Safronov, Samuel Cerezo, Tom\'as Berriel Martins.

**Figure 1.** Figure 1: Illustration of the capture setup for our SLAM&Render dataset, recorded from a Intel RealSense at the end effector of a robotic arm that moves around a set of objects on a table. See the four different light conditions present in our dataset: a) natural, b) cold, c) warm, and d) dark SLAM&Render. Recorded with a robot manipulator, it ensures accurate reproductions of camera motions for different lighting c… view at source ↗

**Figure 2.** Figure 2: Illustration of the objects included in SLAM&Render: a-b) supermarket goods, c) industrial goods, d) – h) object arrangements of the five setups. where RB A ∈ SO (3) and p B A ∈ R 3 denote the rotation and translation of A with respect to B, respectively. For any point p A ∈ R 3 , its coordinates in B are p˜ B = TB Ap˜ A, where p˜ A and p˜ B are the homogeneous vectors of p A and p B, respectively. 3.2 For… view at source ↗

**Figure 3.** Figure 3: Our SLAM&Render dataset contains train and test camera trajectories. From top left to bottom right: a) 2D representation of train trajectory, b) 3D representation of train trajectory, c) 2D representation of test trajectory, and d) 3D representation of test trajectory. Blue and yellow points represent start and end points of the trajectories, respectively. To minimize reflections from the surrounding envir… view at source ↗

**Figure 4.** Figure 4: Illustration of the reference frames involved in the data collection. The accelerometer data are provided in m/s 2 , the gyroscope data in rad/s, and the ground truth and flange positions in mm. All sequences are also available as ROS2 bag files. 4.3 Reference frames The reference frames shown in [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗

**Figure 6.** Figure 6: Corners and fiducial markers detected within the calibration procedure [PITH_FULL_IMAGE:figures/full_fig_p006_6.png] view at source ↗

**Figure 7.** Figure 7: Comparison of NVS for validation and test viewpoints. Although both models perform well when evaluated on validation viewpoints of the training trajectory, the independent test trajectory reveals overfitting [PITH_FULL_IMAGE:figures/full_fig_p007_7.png] view at source ↗

**Figure 8.** Figure 8: Camera tracking result for setup-4 under cold light using a) MonoGS and b) MonoGS with kinematic data as initial seed. Dashed line represent the ground-truth while colored line represent the camera trajectory. Incorporating kinematic data allows drift correction in camera tracking. includes 40 sequences of synchronized RGB and depth images, IMU readings, robot kinematic data, intrinsic and extrinsic camera… view at source ↗

read the original abstract

Models and methods originally developed for Novel View Synthesis and Scene Rendering, such as Neural Radiance Fields (NeRF) and Gaussian Splatting, are increasingly being adopted as representations in Simultaneous Localization and Mapping (SLAM). However, existing datasets fail to include the specific challenges of both fields, such as sequential operations and, in many settings, multi-modality in SLAM or generalization across viewpoints and illumination conditions in neural rendering. Additionally, the data are often collected using sensors which are handheld or mounted on drones or mobile robots, which complicates the accurate reproduction of sensor motions. To bridge these gaps, we introduce SLAM&Render, a novel dataset designed to benchmark methods in the intersection between SLAM, Novel View Rendering and Gaussian Splatting. Recorded with a robot manipulator, it uniquely includes 40 sequences with time-synchronized RGB-D images, IMU readings, robot kinematic data, and ground-truth pose streams. By releasing robot kinematic data, the dataset also enables the assessment of recent integrations of SLAM paradigms within robotic applications. The dataset features five setups with consumer and industrial objects under four controlled lighting conditions, each with separate training and test trajectories. All sequences are static with different levels of object rearrangements and occlusions. Our experimental results, obtained with several baselines from the literature, validate SLAM&Render as a relevant benchmark for this emerging research area.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

A dataset paper with robot kinematics and controlled splits that fills some gaps in SLAM-rendering benchmarks, though the validation stays thin and the lab conditions look limited.

read the letter

This paper's core offering is SLAM&Render, a dataset of 40 sequences recorded on a robot manipulator. It supplies time-synchronized RGB-D, IMU, kinematic data, and ground-truth poses across five setups, four lighting conditions, and separate train/test trajectories. Scenes stay static but vary in object rearrangements and occlusions, using both consumer and industrial items. The robot mount lets users reproduce exact motions, which prior handheld or drone datasets make difficult. That combination is the main practical addition for work at the SLAM and neural rendering intersection, especially when people want to test multi-modal SLAM or viewpoint and illumination generalization in Gaussian Splatting or NeRF-style methods. Releasing the kinematic streams also supports direct robotic integration checks. The abstract notes that several baselines were run and that they support the benchmark's relevance, which at least shows the data can be used with existing pipelines. The collection itself appears straightforward and reproducible on the hardware side. The main weaknesses sit in the evaluation and scope. No quantitative numbers or adaptation details appear in the provided text, so the claim that baselines validate the benchmark rests on an assertion rather than shown results. The setups use only four discrete lighting levels and fully static scenes; that may not expose continuous illumination drift, moving objects, or unstructured environments that dominate field use. If the goal is to test real generalization, these controlled conditions leave a gap. Readers working on hybrid SLAM-rendering pipelines or robotic mapping would find the data useful for controlled experiments. Dataset users who need precise motion replay and multi-condition splits get the clearest value. The work is coherent enough on its own terms to deserve referee time rather than a desk reject, though reviewers will probably ask for fuller results and discussion of how well the conditions match target applications. I would send it to peer review.

Referee Report

2 major / 2 minor

Summary. The paper introduces SLAM&Render, a benchmark dataset for methods at the intersection of SLAM, novel view synthesis, and Gaussian Splatting. It comprises 40 sequences recorded with a robot manipulator, providing time-synchronized RGB-D images, IMU readings, robot kinematic data, and ground-truth poses. The data cover five setups with consumer and industrial objects under four controlled lighting conditions, static scenes with object rearrangements and occlusions, and separate training/test trajectories to evaluate sequential/multi-modal SLAM and viewpoint/illumination generalization.

Significance. If the dataset design and baseline results hold, this work would provide a useful controlled benchmark for integrated SLAM and neural rendering pipelines, with particular value in the release of robot kinematic data for accurate motion reproduction and assessment in robotic contexts. The multi-modal synchronization and structured variations in lighting and scene configuration are explicit strengths that could support systematic evaluation where prior datasets fall short.

major comments (2)

[Dataset] Dataset section: the central claim that the dataset bridges gaps by enabling assessment of generalization across viewpoints and illumination conditions rests on five setups and four discrete lighting conditions with static scenes; however, the absence of continuous illumination variation, dynamic elements, or unstructured environments means the chosen conditions may not expose the failure modes that dominate practical SLAM and rendering deployments, weakening the generalization argument.
[Experiments] Experiments section: while the abstract states that baselines validate the benchmark's relevance, the manuscript provides insufficient detail on quantitative metrics, error analysis, or how baselines were adapted to incorporate the robot kinematic data and multi-modal inputs; this leaves the utility claim only partially supported and requires explicit results to be load-bearing.

minor comments (2)

[Abstract] Abstract: the validation statement would be strengthened by briefly noting one or two key quantitative outcomes from the baselines rather than a general reference.
[Related Work] Related work: add a table comparing SLAM&Render to prior datasets on dimensions such as kinematic data availability, lighting variation, and multi-modality to clarify the claimed gaps.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the detailed and constructive feedback on our manuscript. We address each major comment below and indicate the revisions we will incorporate to strengthen the presentation of the benchmark and its evaluation.

read point-by-point responses

Referee: [Dataset] Dataset section: the central claim that the dataset bridges gaps by enabling assessment of generalization across viewpoints and illumination conditions rests on five setups and four discrete lighting conditions with static scenes; however, the absence of continuous illumination variation, dynamic elements, or unstructured environments means the chosen conditions may not expose the failure modes that dominate practical SLAM and rendering deployments, weakening the generalization argument.

Authors: We appreciate this observation on the scope of our controlled variations. SLAM&Render is intentionally designed around static scenes with discrete lighting conditions, object rearrangements, and occlusions to enable systematic, reproducible testing of viewpoint and illumination generalization using precise robot kinematic ground truth. This structured approach fills a specific gap left by prior datasets that lack synchronized multi-modal data and controlled factors. We agree that the manuscript should more explicitly delineate the intended scope of these generalization tests. In the revision we will update the Dataset and Introduction sections to clarify the design rationale and add a limitations paragraph discussing the absence of continuous illumination changes, dynamic elements, and fully unstructured environments. revision: yes
Referee: [Experiments] Experiments section: while the abstract states that baselines validate the benchmark's relevance, the manuscript provides insufficient detail on quantitative metrics, error analysis, or how baselines were adapted to incorporate the robot kinematic data and multi-modal inputs; this leaves the utility claim only partially supported and requires explicit results to be load-bearing.

Authors: We agree that additional detail is required to make the experimental validation fully load-bearing. The manuscript reports results from several literature baselines, yet we acknowledge the need for greater transparency. In the revised version we will expand the Experiments section with explicit quantitative metrics (including ATE/RPE for SLAM trajectories and PSNR/SSIM/LPIPS for rendering), a breakdown of error sources, and step-by-step descriptions of how the robot kinematic data was used for motion reproduction and how multi-modal (RGB-D + IMU) inputs were integrated into each baseline. These additions will directly support the utility claim. revision: yes

Circularity Check

0 steps flagged

No circularity: dataset paper relies on physical collection and external baselines

full rationale

The paper introduces SLAM&Render as a benchmark dataset collected with a robot manipulator, featuring 40 sequences of time-synchronized RGB-D, IMU, kinematic, and ground-truth pose data across five setups, four lighting conditions, and varying object rearrangements/occlusions. Its central claims concern the dataset's coverage of sequential/multi-modal SLAM challenges and viewpoint/illumination generalization for neural rendering, achieved directly via hardware recording and release of robot kinematics rather than any derivation, prediction, or first-principles result. No equations, fitted parameters, or self-referential definitions appear in the provided text. Validation consists of running literature baselines on the new data, which constitutes external testing rather than circular reduction. The contribution is therefore self-contained against external benchmarks with no load-bearing steps that collapse to inputs by construction.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The paper contributes a new physical dataset rather than deriving new equations or entities; it relies on standard robotics sensor models and static-scene assumptions common to the field.

axioms (1)

domain assumption Static scenes with controlled rearrangements and occlusions adequately test generalization for SLAM and rendering methods
Invoked in the description of sequence features and experimental validation.

pith-pipeline@v0.9.0 · 5794 in / 1242 out tokens · 39633 ms · 2026-05-22T18:55:41.311144+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

IndisputableMonolith/Foundation/RealityFromDistinction.lean reality_from_one_distinction unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

we introduce SLAM&Render, a novel dataset designed to benchmark methods in the intersection between SLAM, Novel View Rendering and Gaussian Splatting

What do these tags mean?

matches: The paper's claim is directly supported by a theorem in the formal canon.
supports: The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends: The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses: The paper appears to rely on the theorem as machinery.
contradicts: The paper's claim conflicts with a theorem or certificate in the canon.
unclear: Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

12 extracted references · 12 canonical work pages · 1 internal anchor

[1]

Bonarini A, Burgard W, Fontana G, Matteucci M, Sorrenti DG, Tardos JD et al

Barron JT, Mildenhall B, Verbin D, Srinivasan PP and Hedman P (2022) Mip-nerf 360: Unbounded anti-aliased neural radiance fields.CVPR. Bonarini A, Burgard W, Fontana G, Matteucci M, Sorrenti DG, Tardos JD et al. (2006) Rawseeds: Robotics advancement through web-publishing of sensorial and elaborated extensive data sets. In:In proceedings of IROS, volume

work page 2022
[2]

Burri M, Nikolic J, Gohl P, Schneider T, Rehder J, Omari S, Achtelik MW and Siegwart R (2016) The euroc micro aerial vehicle datasets.The International Journal of Robotics Research35(10): 1157–1163. Cadena C, Carlone L, Carrillo H, Latif Y , Scaramuzza D, Neira J, Reid I and Leonard JJ (2016) Past, present, and future of simultaneous localization and mapp...

work page 2016
[3]

In:2015 international conference on advanced robotics (ICAR)

Calli B, Singh A, Walsman A, Srinivasa S, Abbeel P and Dollar AM (2015) The ycb object and model set: Towards common benchmarks for manipulation research. In:2015 international conference on advanced robotics (ICAR). IEEE, pp. 510–517. Campos C, Elvira R, Rodr ´ıguez JJG, Montiel JM and Tard ´os JD (2021) Orb-slam3: An accurate open-source library for vis...

work page 2015
[4]

In:Proceedings of the IEEE conference on computer vision and pattern recognition

Jensen R, Dahl A, V ogiatzis G, Tola E and Aanæs H (2014) Large scale multi-view stereopsis evaluation. In:Proceedings of the IEEE conference on computer vision and pattern recognition. pp. 406–413. Keetha N, Karhade J, Jatavallabhula KM, Yang G, Scherer S, Ramanan D and Luiten J (2024) Splatam: Splat track & map 3d gaussians for dense rgb-d slam. In:Proc...

work page 2014
[5]

Robotics11(1):

Macario Barros A, Michel M, Moline Y , Corre G and Carrel F (2022) A comprehensive survey of visual slam algorithms. Robotics11(1):

work page 2022
[6]

In:Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

Martin-Brualla R, Radwan N, Sajjadi MS, Barron JT, Dosovitskiy A and Duckworth D (2021) Nerf in the wild: Neural radiance fields for unconstrained photo collections. In:Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 7210–7219. Martins TB and Civera J (2024) Feature splatting for better novel view synthesis with low ...

work page 2021
[7]

The Replica Dataset: A Digital Replica of Indoor Spaces

Matsuki H, Murai R, Kelly PH and Davison AJ (2024) Gaussian splatting slam. In:Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 18039– 18048. Mildenhall B, Srinivasan PP, Tancik M, Barron JT, Ramamoorthi R and Ng R (2021) Nerf: Representing scenes as neural radiance fields for view synthesis.Communications of the ACM6...

work page internal anchor Pith review Pith/arXiv arXiv 2024
[8]

16558–16569

pp. 16558–16569. Tosi F, Zhang Y , Gong Z, Sandstr¨om E, Mattoccia S, Oswald MR and Poggi M (2024) How nerfs and 3d gaussian splatting are reshaping slam: a survey.arXiv preprint arXiv:2402.132554:

work page arXiv 2024
[9]

Bovik, Hamid R

Wang Z, Bovik A, Sheikh H and Simoncelli E (2004) Image quality assessment: from error visibility to structural similarity.IEEE Transactions on Image Processing13(4): 600–612. DOI: 10.1109/TIP.2003.819861. Ye S, Dong ZH, Hu Y , Wen YH and Liu YJ (2024) Gaussian in the dark: Real-time view synthesis from inconsistent dark images using gaussian splatting. I...

work page doi:10.1109/tip.2003.819861 2004
[10]

Wiley Online Library, p. e15213. Yeshwanth C, Liu YC, Nießner M and Dai A (2023) Scannet++: A high-fidelity dataset of 3d indoor scenes. In:Proceedings of the IEEE/CVF International Conference on Computer Vision. pp. 12–22. Zhang J and Singh S (2014) Loam: Lidar odometry and mapping in real-time. In:Robotics: Science and Systems (RSS). p

work page 2023
[11]

In:Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR)

Zhang R, Isola P, Efros AA, Shechtman E and Wang O (2018) The unreasonable effectiveness of deep features as a perceptual metric. In:Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR). p

work page 2018
[12]

In:Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

Zhu Z, Peng S, Larsson V , Xu W, Bao H, Cui Z, Oswald MR and Pollefeys M (2022) Nice-slam: Neural implicit scalable encoding for slam. In:Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 12786–12796. Prepared usingsagej.cls

work page 2022

[1] [1]

Bonarini A, Burgard W, Fontana G, Matteucci M, Sorrenti DG, Tardos JD et al

Barron JT, Mildenhall B, Verbin D, Srinivasan PP and Hedman P (2022) Mip-nerf 360: Unbounded anti-aliased neural radiance fields.CVPR. Bonarini A, Burgard W, Fontana G, Matteucci M, Sorrenti DG, Tardos JD et al. (2006) Rawseeds: Robotics advancement through web-publishing of sensorial and elaborated extensive data sets. In:In proceedings of IROS, volume

work page 2022

[2] [2]

Burri M, Nikolic J, Gohl P, Schneider T, Rehder J, Omari S, Achtelik MW and Siegwart R (2016) The euroc micro aerial vehicle datasets.The International Journal of Robotics Research35(10): 1157–1163. Cadena C, Carlone L, Carrillo H, Latif Y , Scaramuzza D, Neira J, Reid I and Leonard JJ (2016) Past, present, and future of simultaneous localization and mapp...

work page 2016

[3] [3]

In:2015 international conference on advanced robotics (ICAR)

Calli B, Singh A, Walsman A, Srinivasa S, Abbeel P and Dollar AM (2015) The ycb object and model set: Towards common benchmarks for manipulation research. In:2015 international conference on advanced robotics (ICAR). IEEE, pp. 510–517. Campos C, Elvira R, Rodr ´ıguez JJG, Montiel JM and Tard ´os JD (2021) Orb-slam3: An accurate open-source library for vis...

work page 2015

[4] [4]

In:Proceedings of the IEEE conference on computer vision and pattern recognition

Jensen R, Dahl A, V ogiatzis G, Tola E and Aanæs H (2014) Large scale multi-view stereopsis evaluation. In:Proceedings of the IEEE conference on computer vision and pattern recognition. pp. 406–413. Keetha N, Karhade J, Jatavallabhula KM, Yang G, Scherer S, Ramanan D and Luiten J (2024) Splatam: Splat track & map 3d gaussians for dense rgb-d slam. In:Proc...

work page 2014

[5] [5]

Robotics11(1):

Macario Barros A, Michel M, Moline Y , Corre G and Carrel F (2022) A comprehensive survey of visual slam algorithms. Robotics11(1):

work page 2022

[6] [6]

In:Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

Martin-Brualla R, Radwan N, Sajjadi MS, Barron JT, Dosovitskiy A and Duckworth D (2021) Nerf in the wild: Neural radiance fields for unconstrained photo collections. In:Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 7210–7219. Martins TB and Civera J (2024) Feature splatting for better novel view synthesis with low ...

work page 2021

[7] [7]

The Replica Dataset: A Digital Replica of Indoor Spaces

Matsuki H, Murai R, Kelly PH and Davison AJ (2024) Gaussian splatting slam. In:Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 18039– 18048. Mildenhall B, Srinivasan PP, Tancik M, Barron JT, Ramamoorthi R and Ng R (2021) Nerf: Representing scenes as neural radiance fields for view synthesis.Communications of the ACM6...

work page internal anchor Pith review Pith/arXiv arXiv 2024

[8] [8]

16558–16569

pp. 16558–16569. Tosi F, Zhang Y , Gong Z, Sandstr¨om E, Mattoccia S, Oswald MR and Poggi M (2024) How nerfs and 3d gaussian splatting are reshaping slam: a survey.arXiv preprint arXiv:2402.132554:

work page arXiv 2024

[9] [9]

Bovik, Hamid R

Wang Z, Bovik A, Sheikh H and Simoncelli E (2004) Image quality assessment: from error visibility to structural similarity.IEEE Transactions on Image Processing13(4): 600–612. DOI: 10.1109/TIP.2003.819861. Ye S, Dong ZH, Hu Y , Wen YH and Liu YJ (2024) Gaussian in the dark: Real-time view synthesis from inconsistent dark images using gaussian splatting. I...

work page doi:10.1109/tip.2003.819861 2004

[10] [10]

Wiley Online Library, p. e15213. Yeshwanth C, Liu YC, Nießner M and Dai A (2023) Scannet++: A high-fidelity dataset of 3d indoor scenes. In:Proceedings of the IEEE/CVF International Conference on Computer Vision. pp. 12–22. Zhang J and Singh S (2014) Loam: Lidar odometry and mapping in real-time. In:Robotics: Science and Systems (RSS). p

work page 2023

[11] [11]

In:Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR)

Zhang R, Isola P, Efros AA, Shechtman E and Wang O (2018) The unreasonable effectiveness of deep features as a perceptual metric. In:Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR). p

work page 2018

[12] [12]

In:Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

Zhu Z, Peng S, Larsson V , Xu W, Bao H, Cui Z, Oswald MR and Pollefeys M (2022) Nice-slam: Neural implicit scalable encoding for slam. In:Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 12786–12796. Prepared usingsagej.cls

work page 2022