Pith. sign in

REVIEW 4 major objections 5 minor 2 cited by

VIN-NBV: A View Introspection Network for Next-Best-View Selection

T0 review · 4 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read The paper claims that scoring candidate viewpoints by predicted reconstruction improvement, rather than by coverage, gives roughly 30% better reconstruction quality under the same greedy strategy, and roughly 40% better than RL-based…

desk verdict A genuinely new objective for next-best-view selection, but the headline claims against published baselines outrun the evidence. read the letter →

arxiv 2505.06219 v3 pith:2LSSNCVR submitted 2025-05-09 cs.CV cs.RO

Noah Frahm , Dongxu Zhao , Andrea Dunn Beltran , Ron Alterovitz , Jan-Michael Frahm , Junier Oliva , Roni Sengupta This is my paper · ORCID
classification cs.CVcs.RO
keywords NextBestViewactive3DreconstructionplanningqualityChamferdistanceimitationlearningordinalregressionRGB-Dacquisition
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The Next-Best-View problem asks which viewpoint an agent should capture next to build the best 3D reconstruction under a limited budget. The paper argues that the usual proxy for choosing views, maximizing coverage of unseen surface, is the wrong objective, because occluded or geometrically complex regions are exactly where an extra view helps most even when it covers little new area. It introduces the View Introspection Network (VIN), a lightweight network that predicts the Relative Reconstruction Improvement (RRI) of a candidate viewpoint directly from already captured RGB-D images and camera parameters, without acquiring that view. Wrapped in a simple greedy sampler, VIN-NBV, this quality-targeting score improves Chamfer distance by about 30% over a coverage score with the same greedy strategy, and by about 40% over RL policies such as GenNBV and ScanRL on the OmniObject3D houses benchmark. If correct, this means a small imitation-learned predictor can beat more complex learned planners, and that reconstruction quality itself is a learnable acquisition target.

What carries the argument

The load-bearing object is the RRI fitness score produced by VIN. VIN projects the current point-cloud reconstruction into a 512 by 512 by 5 feature grid as seen from the candidate camera, carrying surface normals, point visibility counts, and depth, plus per-pixel variances and a two-element coverage feature Fempty that distinguishes empty pixels inside the reconstructed hull (holes) from empty pixels outside it (unseen geometry). A convolutional encoder and MLP map this to an ordinal class over 15 RRI bins. Training uses stage-dependent z-score normalization of RRI so early captures, which have larger absolute gains, are scored on the same scale as later incremental ones, and uses the CORAL ranking-aware classification loss so that large misordering between far-apart bins is penalized most. The network's job is to score any query view without rendering it, which is what makes the greedy policy cheap.

What would settle it

Run VIN-NBV against Cov-NBV on objects with unknown and off-center bounding boxes using only noisy, estimated depth from real sensors, and compare final reconstructions; if the roughly 30% Chamfer-distance advantage disappears or reverses outside the object-centric, noise-free benchmark, the claimed generality and direct-quality advantage are not established.

Watch

Extended reading notes

Core claim

The central claim is that a policy trained to optimize 3D reconstruction quality directly, rather than coverage, selects next views that build markedly better reconstructions under the same resource budget. The paper defines the Relative Reconstruction Improvement of a query view q as RRI(q) = (CD(Rbase,RGT) - CD(Rbase∪q,RGT))/CD(Rbase,RGT), i.e. how much adding that view would reduce Chamfer distance to the ground truth. The View Introspection Network is trained by imitation to predict this oracle RRI from the existing reconstruction, the base camera parameters, and the query camera parameters alone; the acquisition policy greedily renders the candidate with the highest predicted RRI. The empirical claim is that this RRI fitness criterion yields roughly 30% lower reconstruction error than a coverage criterion run through the identical greedy sampling loop, and roughly 40% lower error than the coverage-based RL methods ScanRL and GenNBV at 20 captures, with transfer to unseen object categories and to time- and collision-constrained settings.

Load-bearing premise

The policy's candidate views are sampled from hemispherical shells centered on the object, so the agent must know where the object is and roughly how large it is; the 'no prior scene knowledge' claim depends on that object-centric setup, and the evaluation also assumes noise-free depth and ground-truth Chamfer distances.

Editorial extensions

If this is right

  • Using VIN's predicted RRI instead of a coverage score in the same greedy sampling loop reduces Chamfer distance by about 30% at 20 captures.
  • Under a strict 15-second motion budget, VIN-NBV beats the coverage baseline by about 25%, and it retains gains when straight-line paths must avoid collisions.
  • The policy transfers to unseen categories such as dinosaurs, toy animals, toy motorcycles, and trucks, with the largest early-stage gains on shapes with self-occlusion.
  • The coverage feature Fempty matters mainly in later acquisition stages, suggesting coverage remains a useful signal once most of the scene is visible.
  • The remaining gap to an oracle that knows ground-truth RRI is concentrated early in acquisition, where the biggest quality gains are still on the table.

Reading between the lines

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

  • The RRI idea is not tied to point clouds: the same imitation objective could score candidate views for radiance-field or 3D Gaussian reconstructions, replacing point-cloud projection with a differentiable renderer; the paper does not test this.
  • The stage-wise z-score normalization of improvement scores is a transferable trick: any learned utility that shrinks as information accumulates can be normalized per stage before training, which likely stabilizes active-learning and reward-learning pipelines.
  • Because the evaluation uses noise-free ground-truth depth and ground-truth Chamfer distance, real deployment would need to estimate both; testing with monocular depth and no object centering is the natural next experiment.
  • A scalar RRI over arbitrary camera poses could also be used as a planning cost for continuous trajectory optimization, not just for discrete view selection.
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

4 major / 5 minor

Summary. The paper proposes VIN-NBV, a next-best-view (NBV) selection method that trains a lightweight neural network, the View Introspection Network (VIN), to predict the Relative Reconstruction Improvement (RRI) of a candidate viewpoint, defined via Chamfer distance against ground truth. The VIN is trained with imitation learning on oracle RRI values computed by reconstructing the scene with each candidate view, and the policy greedily selects the candidate with the highest predicted RRI. Experiments on OmniObject3D houses compare VIN-NBV with a coverage-based greedy baseline (Cov-NBV), as well as with reinforcement-learning baselines Scan-RL and GenNBV, reporting a ~30% gain over the coverage criterion and ~40% over the RL baselines. Additional experiments show generalization to dinosaurs, motorcycles, animals, and trucks, and an ablation of the coverage feature F_empty.

Significance. If the central claims hold, the paper makes a useful contribution by showing that directly optimizing a predicted reconstruction-quality criterion can substantially outperform coverage-based view selection under the same greedy policy. The internal comparison against Cov-NBV is the most credible part of the paper: both methods use the same sampling and greedy selection procedure, and the reported Chamfer distance curves show a consistent and early gap in favor of VIN-NBV. The imitation-learning formulation, with training labels derived from external ground-truth Chamfer distance, is not circular and is a sound way to supervise a view-utility predictor. The paper also demonstrates generalization to unseen object categories, which is a meaningful strength. However, the headline comparison against Scan-RL and GenNBV rests on quoted numbers from prior papers rather than commensurable re-runs, and the claims are currently stronger than the evidence.

major comments (4)
  1. [§4, Table 1 and Section 4.1] The claim that VIN-NBV outperforms Scan-RL and GenNBV by ~40% is not established by the evidence presented. Table 1 mixes VIN-NBV and Cov-NBV numbers computed by the authors with GenNBV and ScanRL numbers quoted from their respective papers, as the text explicitly states in Section 4 ('Since the model weights of GenNBV are unavailable, we compare with their reported results in the paper directly'). The evaluation conditions are not the same: the candidate views in this paper are sampled from 120 viewpoints in 3 hemispherical shells around the object, the initial two views are chosen in a specific way, and Appendix 7.2 states that house 27 is excluded because its scale factor cannot be computed. The paper provides no per-object values, object counts, confidence intervals, or error bars for Table 1, so it is impossible to determine whether the difference between 0.20 cm and 0.33 cm is significant. To support the abstract claim, the authors should either re-run GenNBV and ScanRL under their own protocol (if weights or implementations become available) or clearly restrict the claim to an approximate comparison and remove the unqualified '~40%' statement.
  2. [Table 2, Section 4.4] Table 2 directly contradicts the unqualified statement that VIN-NBV 'outperforms deep reinforcement learning methods, Scan-RL and GenNBV, by ~40%.' On the dinosaurs category, VIN-NBV reports a Chamfer distance of 0.04, which is worse than GenNBV's 0.03, and the text itself concedes that VIN-NBV is 'slightly behind GenNBV on dinosaurs.' The abstract and the conclusions should be revised to report the per-category results honestly, for example by presenting the improvement as category-dependent and noting the dinosaur exception, rather than asserting a single universal margin.
  3. [§4, sampling protocol] The claim that VIN-NBV 'operates without prior scene knowledge' is not supported by the experimental protocol. Section 4 states that 'for all sampling-based acquisition policies ... we uniformly render 120 viewpoints in 3 hemispherical shells around the object.' This requires the agent to know that there is a single object, that it is centered in the coordinate frame, and that the hemispherical shells at the chosen radii cover the object's extent. In a real deployment without prior scene knowledge, the agent would not know where to place these shells or whether the candidate views are even feasible. The paper should either demonstrate the method under an object-agnostic sampling strategy (for example, sampling over a bounding volume estimated from the initial captures) or explicitly qualify the 'without prior scene knowledge' claim to mean 'without a prior 3D model or dense scan.'
  4. [Figures 4 and 8, Section 4.1] The ~30% improvement over Cov-NBV, which is the strongest internally controlled result, is reported as a single average curve without error bars, per-object variance, or statistical significance testing. Figure 4 shows that the gap is largest in early acquisition stages and nearly vanishes by 20 captures, so the headline gain depends on the averaging procedure and the chosen number of captures. The authors should report the distribution over objects (for example, per-object Chamfer distances at 20 captures, or standard errors over the test set) and, ideally, run multiple initial-view randomizations to support the claim that the improvement is consistent rather than driven by a few objects.
minor comments (5)
  1. [Abstract and Section 1] The abstract says 'outperforms ... by ~40%' while the conclusion says 'reducing reconstruction error by up to 40%'; these are different claims, and the manuscript should use one consistent, precisely qualified statement.
  2. [Equation (4)] Equation (4) uses the notation '[RRI(q) = M_phi(...)' with an unclosed bracket and an inconsistent use of the hat symbol; this should be typeset consistently as \hat{RRI}(q).
  3. [Section 4.2] The assumption of a drone traveling at 4 mph and taking straight-line paths is stated without justification or sensitivity analysis; adding a sentence on how this choice affects the time-limited results would improve reproducibility.
  4. [Appendix 7.2] The exclusion of house 27 is only mentioned in the appendix, but it affects the comparability of Table 1 with the original GenNBV and ScanRL results; this should be stated prominently in Section 4.1.
  5. [Figure 7] The 'collision' variants in Figure 7 are not defined in the caption or the text; the caption should explain what 'Cov-NBV collision' and 'VIN-NBV collision' mean and how the collision constraint is applied.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation: VIN is trained on externally computed oracle RRI and evaluated on held-out categories; the abstract's comparison to published baselines is a data-comparability concern, not a circularity.

full rationale

The paper's claimed derivation is not circular. VIN's training target, Eq. (2), is the Oracle RRI computed by explicitly reconstructing the scene with each candidate query view and measuring Chamfer distance against ground truth (Section 3.4). The policy then selects the view maximizing VIN's predicted RRI (Algorithm 1), and evaluation measures final reconstruction Chamfer distance (Section 4). Alignment between the training objective and the evaluation metric is direct optimization, not a self-referential reduction: the oracle labels come from an external rendering of ground-truth models, training is on Houses3K, and tests are on held-out OmniObject3D houses plus generalization classes. The concrete limitations disclosed in the manuscript — GenNBV weights are unavailable so published numbers are used directly, house 27 is excluded because its scale factor cannot be computed, ground-truth depth is used, and the sampling shells require the object to be centered with roughly known scale — weaken the external validity and comparability of the headline ~40% claim, but none of these makes a prediction equal to its input by construction. No load-bearing self-citation chain or ansatz-smuggling-via-citation was found; the cited prior works are used as baselines and dataset protocols, not to justify the RRI criterion. Therefore the circularity score is 0.

Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

The VIN's learned weights are standard trainable parameters, but the paper also relies on data-derived normalization statistics and a scale factor to make comparisons. The core assumptions are the validity of Chamfer distance as the quality metric, the transferability of the oracle RRI training signal, and the object-centric sampling setup. No new physical entities are postulated.

free parameters (2)
  • Stage-wise RRI z-score statistics = means and standard deviations per capture stage (numerical values not reported)
    Used to normalize oracle RRI into 15 ordinal classes during training (Sec. 3.4). These statistics are estimated from the training set and are part of the training pipeline, though they do not directly enter the policy at test time.
  • Chamfer distance scale factor for OmniObject3D houses = derived from GenNBV's resized houses (not reported)
    Authors rescale their Chamfer distance values to match GenNBV's reported scale (Sec. 7.2). This factor is computed from the provided resized models and affects all house comparisons, but is an auxiliary calibration rather than a model parameter.
assumptions (4)
  • domain assumption Chamfer distance to the ground-truth point cloud is a valid and sufficient measure of reconstruction quality.
    The main metric and the training target (RRI) are both defined via Chamfer distance (Eq. 1, 2); if this metric does not capture reconstruction quality, the central comparison is weakened.
  • domain assumption The oracle RRI, computed by exhaustively simulating each query view with ground-truth reconstruction, is an accurate and transferable training signal.
    VIN is trained via imitation learning to predict oracle RRI (Sec. 3.4); the assumption that this offline-computed target generalizes to unseen objects is load-bearing for the policy's performance.
  • ad hoc to paper Candidate views are sampled from three hemispherical shells centered on the object, so the object's location and rough scale are known at acquisition time.
    The policy evaluates a fixed set of 120 sampled query views around the object (Sec. 4). This setup is specific to the paper's object-centric evaluation and is not justified for arbitrary unknown scenes.
  • domain assumption Back-projecting RGB-D images into point clouds provides a sufficient reconstruction for featurization and policy decisions.
    The VIN operates on a point cloud reconstructed from base RGB-D views (Sec. 3.3); the quality of this reconstruction limits the quality of the predicted RRI.

how reviews work

0 comments
Cite this review

Pith. "Pith review of VIN-NBV: A View Introspection Network for Next-Best-View Selection." pith.science (2026). https://pith.science/paper/2LSSNCVR

@misc{pith2026250506219,
  author       = {Pith},
  title        = {Pith review of: VIN-NBV: A View Introspection Network for Next-Best-View Selection},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/2LSSNCVR}},
  note         = {Machine review of arXiv:2505.06219}
}
read the original abstract

Next Best View (NBV) algorithms aim to maximize 3D scene acquisition quality using minimal resources, e.g. number of acquisitions, time taken, or distance traversed. Prior methods often rely on coverage maximization as a proxy for reconstruction quality, but for complex scenes with occlusions and finer details, this is not always sufficient and leads to poor reconstructions. Our key insight is to train an acquisition policy that directly optimizes for reconstruction quality rather than just coverage. To achieve this, we introduce the View Introspection Network (VIN): a lightweight neural network that predicts the Relative Reconstruction Improvement (RRI) of a potential next viewpoint without making any new acquisitions. We use this network to power a simple, yet effective, sequential samplingbased greedy NBV policy. Our approach, VIN-NBV, generalizes to unseen object categories, operates without prior scene knowledge, is adaptable to resource constraints, and can handle occlusions. We show that our RRI fitness criterion leads to a ~30% gain in reconstruction quality over a coverage-based criterion using the same greedy strategy. Furthermore, VIN-NBV also outperforms deep reinforcement learning methods, Scan-RL and GenNBV, by ~40%.

Figures

Figures reproduced from arXiv: 2505.06219 by the authors.

Figure 1
Figure 1. Our approach, VIN-NBV (blue), is an NBV policy that selects best next views by maximizing a predicted reconstruction quality [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. (a) VIN-NBV reconstructs the 3D scene from prior RGB-D captures, samples candidate viewpoints, and selects the one with [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. We compare reconstruction quality of our VIN-NBV [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (10 more)
Figure 4
Figure 4. Figure 4: Evaluation with Limited Acquisitions. Average re￾construction error (in cm) on OmniObject3D [41] houses under constraint on number of acquisitions. VIN-NBV consistently out￾performs coverage-based NBV policies. we calculate the metrics between the reconstructed and the…
Figure 5
Figure 5. Figure 5: (a) mean coverage percentage and (b) F1 score across all OmniObject3D house objects at each acquisition step. 20 30 40 50 60 70 80 90 0.2 0.3 0.4 0.5 0.6 0.7 0.8 Average Final Chamfer Distance Average Final Chamfer Distance vs Time Cov-NBV VIN-NBV (Ours) Oracle-NBV [P…
Figure 6
Figure 6. Figure 6: Evaluation with Time-Limited Agent Motion. Av￾erage reconstruction error (in cm) on OmniObject3D [41] houses under constraint on time traversed. VIN-NBV consistently outper￾forms coverage based NBV policies. achieves large gains over Cov-NBV early on, converging to a s…
Figure 8
Figure 8. Figure 8: Generalization Across Object Categories. We graph the average Chamfer Distance at each capture stage in three additional object classes namely dinosaurs, toy motorcycles, and toy animals. We compare VIN-NBV to a coverage baseline for 20 captures. 0s 15s 30s 45s 60s 75s…
Figure 9
Figure 9. Figure 9: We visualize reconstruction obtained by VIN-NBV and Cov-NBV for different constraints on time traversed. For smaller time [PITH_FULL_IMAGE:figures/full_fig_p008_9.png]
Figure 10
Figure 10. Figure 10: Average Chamfer Distance across all truck objects from [PITH_FULL_IMAGE:figures/full_fig_p011_10.png]
Figure 11
Figure 11. Figure 11: Comparison of VIN-NBV and Cov-NBV after 10 acquisitions on motorcycles and houses from OmniObject3D[ [PITH_FULL_IMAGE:figures/full_fig_p012_11.png]
Figure 12
Figure 12. Figure 12: Comparison of VIN-NBV and Cov-NBV after 10 acquisitions on animals, house, and motorcycles from OmniObject3D[ [PITH_FULL_IMAGE:figures/full_fig_p013_12.png]
Figure 13
Figure 13. Figure 13: Comparison of VIN-NBV and Cov-NBV after 10 acquisitions on dinosaurs and animals from OmniObject3D[ [PITH_FULL_IMAGE:figures/full_fig_p014_13.png]
Figure 14
Figure 14. Figure 14: Comparison of VIN-NBV and Cov-NBV after 10 acquisitions on motorcycles and houses from OmniObject3D[ [PITH_FULL_IMAGE:figures/full_fig_p015_14.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 2 Pith papers

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

  1. DA-NBV: A Direction-Aware Next-Best-View Planner for Efficient 3D Reconstruction of Ships at Sea

    cs.RO 2026-08 conditional novelty 5.0 of 10

    By tracking viewing directions per voxel and using a learned position-advantage field, DA-NBV lifts reconstruction coverage by about 3 percentage points and cuts Chamfer distance by 43% over strong baselines in simula...

  2. Scene-SAM3D: Multi-View Scene Asset Generation Without Fine-Tuning

    cs.CV 2026-07 conditional novelty 5.0 of 10

    A training-free pipeline combining occupancy-based view selection, early-stop sign-consensus velocity fusion, and rigid Gaussian layout refinement improves SAM3D-based 3D scene generation on Replica and ScanNet++.

Reference graph

Works this paper leans on

49 extracted references · 39 canonical work pages · cited by 2 Pith papers

  1. [1]

    https : / / projects

    Robot rescuers to help save lives after disas- ters. https : / / projects . research - and - innovation . ec . europa . eu / en / horizon - magazine / robot - rescuers - help - save - lives-after-disasters. Written 19 March 2014. 2

    work page 2014

  2. [2]

    Building rome in a day

    Sameer Agarwal, Yasutaka Furukawa, Noah Snavely, Ian Si- mon, Brian Curless, Steven M Seitz, and Richard Szeliski. Building rome in a day. Communications of the ACM , 54 (10):105–112, 2011. 1

    work page 2011

  3. [3]

    Patchmatch stereo-stereo matching with slanted support win- dows

    Michael Bleyer, Christoph Rhemann, and Carsten Rother. Patchmatch stereo-stereo matching with slanted support win- dows. In Bmvc, pages 1–11, 2011. 1

    work page 2011

  4. [4]

    Rank consistent ordinal regression for neural networks with appli- cation to age estimation

    Wenzhi Cao, Vahid Mirjalili, and Sebastian Raschka. Rank consistent ordinal regression for neural networks with appli- cation to age estimation. Pattern Recognition Letters, 140: 325–331, 2020. 5, 11

    work page 2020

  5. [5]

    Gennbv: Generalizable next-best-view policy for active 3d reconstruction

    Xiao Chen, Quanyi Li, Tai Wang, Tianfan Xue, and Jiang- miao Pang. Gennbv: Generalizable next-best-view policy for active 3d reconstruction. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 16436–16445, 2024. 2, 3, 5, 6, 8, 11

    work page 2024

  6. [6]

    Connolly

    C. Connolly. The determination of next best views. In Pro- ceedings. 1985 IEEE International Conference on Robotics and Automation, pages 432–435, 1985. 5

    work page 1985

  7. [7]

    A reinforcement learning approach to the view planning problem

    Mustafa Devrim Kaba, Mustafa Gokhan Uzunbas, and Ser Nam Lim. A reinforcement learning approach to the view planning problem. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition , pages 6933– 6941, 2017. 2

    work page 2017

  8. [8]

    PyTorch Lightning, 2019

    William Falcon and The PyTorch Lightning team. PyTorch Lightning, 2019. 11

    work page 2019

Show all 49 references
  1. [9]

    Build- ing rome on a cloudless day

    Jan-Michael Frahm, Pierre Fite-Georgel, David Gallup, Tim Johnson, Rahul Raguram, Changchang Wu, Yi-Hung Jen, Enrique Dunn, Brian Clipp, Svetlana Lazebnik, et al. Build- ing rome on a cloudless day. In Computer Vision–ECCV 2010: 11th European Conference on Computer Vision, Her...

    2010

  2. [10]

    Accurate, dense, and robust multiview stereopsis

    Yasutaka Furukawa and Jean Ponce. Accurate, dense, and robust multiview stereopsis. IEEE transactions on pattern analysis and machine intelligence, 32(8):1362–1376, 2009

    2009

  3. [11]

    Towards internet-scale multi-view stereo

    Yasutaka Furukawa, Brian Curless, Steven M Seitz, and Richard Szeliski. Towards internet-scale multi-view stereo. In 2010 IEEE computer society conference on computer vi- sion and pattern recognition, pages 1434–1441. IEEE, 2010. 2

    2010

  4. [12]

    Multi-view stereo for commu- nity photo collections

    Michael Goesele, Noah Snavely, Brian Curless, Hugues Hoppe, and Steven M Seitz. Multi-view stereo for commu- nity photo collections. In 2007 IEEE 11th international con- ference on computer vision, pages 1–8. IEEE, 2007. 1

    2007

  5. [13]

    Scone: Surface coverage optimization in unknown environ- ments by volumetric integration

    Antoine Gu ´edon, Pascal Monasse, and Vincent Lepetit. Scone: Surface coverage optimization in unknown environ- ments by volumetric integration. Advances in Neural Infor- mation Processing Systems, 35:20731–20743, 2022. 2

    2022

  6. [14]

    MACARONS: Mapping And Coverage Antic- ipation with RGB ONline Self-supervision

    Antoine Gu ˜A©don, Tom Monnier, Pascal Monasse, and Vin- cent Lepetit. MACARONS: Mapping And Coverage Antic- ipation with RGB ONline Self-supervision. In CVPR, 2023. 2

    2023

  7. [15]

    Next-best-view plan- ning for surface reconstruction of large-scale 3d environ- ments with multiple uavs

    Guillaume Hardouin, Julien Moras, Fabio Morbidi, Julien Marzat, and El Mustapha Mouaddib. Next-best-view plan- ning for surface reconstruction of large-scale 3d environ- ments with multiple uavs. In 2020 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS...

    2020

  8. [16]

    Learn-to-score: Ef- ficient 3d scene exploration by predicting view utility

    Benjamin Hepp, Debadeepta Dey, Sudipta N Sinha, Ashish Kapoor, Neel Joshi, and Otmar Hilliges. Learn-to-score: Ef- ficient 3d scene exploration by predicting view utility. In Proceedings of the European conference on computer vision (ECCV), pages 437–452, 2018. 2

    2018

  9. [17]

    Plan3d: Viewpoint and trajectory optimization for aerial multi-view stereo reconstruction

    Benjamin Hepp, Matthias Nießner, and Otmar Hilliges. Plan3d: Viewpoint and trajectory optimization for aerial multi-view stereo reconstruction. ACM Transactions on Graphics (TOG), 38(1):1–17, 2018. 2

    2018

  10. [18]

    Image se- lection for improved multi-view stereo

    Alexander Hornung, Boyi Zeng, and Leif Kobbelt. Image se- lection for improved multi-view stereo. In 2008 IEEE Con- ference on Computer Vision and Pattern Recognition, pages 1–8. IEEE, 2008. 2

    2008

  11. [19]

    An information gain formulation for active vol- umetric 3d reconstruction

    Stefan Isler, Reza Sabzevari, Jeffrey Delmerico, and Davide Scaramuzza. An information gain formulation for active vol- umetric 3d reconstruction. In 2016 IEEE International Con- ference on Robotics and Automation (ICRA) , pages 3477– 3484, 2016. 2, 5

    2016

  12. [20]

    Fisherrf: Active view selection and uncertainty quantification for radiance fields using fisher information

    Wen Jiang, Boshu Lei, and Kostas Daniilidis. Fisherrf: Active view selection and uncertainty quantification for radiance fields using fisher information. arXiv preprint arXiv:2311.17874, 2023. 2

    2023 arXiv

  13. [21]

    View planning for 3d shape reconstruction of buildings with unmanned aerial vehicles

    Wei Jing, Joseph Polden, Pey Yuen Tao, Wei Lin, and Kenji Shimada. View planning for 3d shape reconstruction of buildings with unmanned aerial vehicles. In 2016 14th Inter- national Conference on Control, Automation, Robotics and Vision (ICARCV), pages 1–6. IEEE, 2016. 2

    2016

  14. [22]

    So-nerf: Active view planning for nerf using surrogate objectives

    Keifer Lee, Shubham Gupta, Sunglyoung Kim, Bhargav Makwana, Chao Chen, and Chen Feng. So-nerf: Active view planning for nerf using surrogate objectives. arXiv preprint arXiv:2312.03266, 2023. 2

    2023 arXiv

  15. [23]

    Uncertainty guided pol- icy for active robotic 3d reconstruction using neural radiance fields, 2022

    Soomin Lee, Le Chen, Jiahao Wang, Alexander Liniger, Suryansh Kumar, and Fisher Yu. Uncertainty guided pol- icy for active robotic 3d reconstruction using neural radiance fields, 2022. 2

    2022

  16. [24]

    Sgdr: Stochastic gradient descent with warm restarts, 2017

    Ilya Loshchilov and Frank Hutter. Sgdr: Stochastic gradient descent with warm restarts, 2017. 11 9

    2017

  17. [25]

    Decoupled weight decay regularization, 2019

    Ilya Loshchilov and Frank Hutter. Decoupled weight decay regularization, 2019. 11

    2019

  18. [26]

    Occlusions as a guide for planning the next view

    Jasna Maver and Ruzena Bajcsy. Occlusions as a guide for planning the next view. IEEE transactions on pattern analy- sis and machine intelligence, 15(5):417–433, 1993. 2

    1993

  19. [27]

    Ac- tivenerf: Learning where to see with uncertainty estimation

    Xuran Pan, Zihang Lai, Shiji Song, and Gao Huang. Ac- tivenerf: Learning where to see with uncertainty estimation. In European Conference on Computer Vision , pages 230–

  20. [28]

    Pytorch: An im- perative style, high-performance deep learning library, 2019

    Adam Paszke, Sam Gross, Francisco Massa, Adam Lerer, James Bradbury, Gregory Chanan, Trevor Killeen, Zeming Lin, Natalia Gimelshein, Luca Antiga, Alban Desmaison, Andreas K ¨opf, Edward Yang, Zach DeVito, Martin Raison, Alykhan Tejani, Sasank Chilamkurthy, Benoit Steiner, Lu F...

    2019

  21. [29]

    Next-best view policy for 3d reconstruction

    Daryl Peralta, Joel Casimiro, Aldrin Michael Nilles, Jus- tine Aletta Aguilar, Rowel Atienza, and Rhandley Cajote. Next-best view policy for 3d reconstruction. In Computer Vision–ECCV 2020 Workshops: Glasgow, UK, August 23– 28, 2020, Proceedings, Part IV 16, pages 558–573. Springer,

    2020

  22. [30]

    A solution to the next best view problem for automated surface acquisition

    Richard Pito. A solution to the next best view problem for automated surface acquisition. IEEE Transactions on pattern analysis and machine intelligence, 21(10):1016–1030, 1999. 6

    1999

  23. [31]

    Sukhatme

    Christian Potthast and Gaurav S. Sukhatme. A probabilis- tic framework for next best view estimation in a cluttered environment. Journal of Visual Communication and Image Representation, 25(1):148–164, 2014. Visual Understanding and Applications with RGB-D Cameras. 2

    2014

  24. [32]

    My3dgen: Building lightweight personalized 3d gen- erative model

    Luchao Qi, Jiaye Wu, Shengze Wang, and Soumyadip Sen- gupta. My3dgen: Building lightweight personalized 3d gen- erative model. arXiv preprint arXiv:2307.05468, 2023. 6

    2023 arXiv

  25. [33]

    Neurar: Neural uncertainty for autonomous 3d reconstruction with implicit neural representations

    Yunlong Ran, Jing Zeng, Shibo He, Jiming Chen, Lincheng Li, Yingfeng Chen, Gimhee Lee, and Qi Ye. Neurar: Neural uncertainty for autonomous 3d reconstruction with implicit neural representations. IEEE Robotics and Automation Let- ters, 8(2):1125–1132, 2023. 2

    2023

  26. [34]

    Accelerating 3d deep learning with pytorch3d

    Nikhila Ravi, Jeremy Reizenstein, David Novotny, Tay- lor Gordon, Wan-Yen Lo, Justin Johnson, and Georgia Gkioxari. Accelerating 3d deep learning with pytorch3d. arXiv:2007.08501, 2020. 11

    2007 arXiv

  27. [35]

    Submodular trajectory optimization for aerial 3d scanning

    Mike Roberts, Debadeepta Dey, Anh Truong, Sudipta Sinha, Shital Shah, Ashish Kapoor, Pat Hanrahan, and Neel Joshi. Submodular trajectory optimization for aerial 3d scanning. In Proceedings of the IEEE International Conference on Computer Vision, pages 5324–5333, 2017. 2

    2017

  28. [36]

    Pixelwise view selection for unstructured multi-view stereo

    Johannes L Sch ¨onberger, Enliang Zheng, Jan-Michael Frahm, and Marc Pollefeys. Pixelwise view selection for unstructured multi-view stereo. In Computer Vision–ECCV 2016: 14th European Conference, Amsterdam, The Nether- lands, October 11-14, 2016, Proceedings, Part III 14, pag...

    2016

  29. [37]

    Uncertainty- driven active vision for implicit scene reconstruction

    Edward J Smith, Michal Drozdzal, Derek Nowrouzezahrai, David Meger, and Adriana Romero-Soriano. Uncertainty- driven active vision for implicit scene reconstruction. arXiv preprint arXiv:2210.00978, 2022. 2

    2022 arXiv

  30. [38]

    Wide- baseline stereo from multiple views: a probabilistic account

    Christoph Strecha, Rik Fransens, and Luc Van Gool. Wide- baseline stereo from multiple views: a probabilistic account. In Proceedings of the 2004 IEEE Computer Society Con- ference on Computer Vision and Pattern Recognition, 2004. CVPR 2004., pages I–I. IEEE, 2004. 1

    2004

  31. [39]

    Learning view selection for 3d scenes

    Yifan Sun, Qixing Huang, Dun-Yu Hsiao, Li Guan, and Gang Hua. Learning view selection for 3d scenes. In Pro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 14464–14473, 2021. 2

    2021

  32. [40]

    Quality-driven poisson-guided autoscanning

    Shihao Wu, Wei Sun, Pinxin Long, Hui Huang, Daniel Cohen-Or, Minglun Gong, Oliver Deussen, and Baoquan Chen. Quality-driven poisson-guided autoscanning. ACM Trans. Graph., 33(6), 2014. 6

    2014

  33. [41]

    Omniobject3d: Large-vocabulary 3d object dataset for realistic perception, reconstruction and generation

    Tong Wu, Jiarui Zhang, Xiao Fu, Yuxin Wang, Jiawei Ren, Liang Pan, Wayne Wu, Lei Yang, Jiaqi Wang, Chen Qian, et al. Omniobject3d: Large-vocabulary 3d object dataset for realistic perception, reconstruction and generation. In Pro- ceedings of the IEEE/CVF Conference on Compute...

    2023

  34. [42]

    Sampling-based path planning for high-quality aerial 3d re- construction of urban scenes

    Feihu Yan, Enyong Xia, Zhaoxin Li, and Zhong Zhou. Sampling-based path planning for high-quality aerial 3d re- construction of urban scenes. Remote Sensing, 13(5):989,

  35. [43]

    Pc-nbv: A point cloud based deep network for efficient next best view plan- ning

    Rui Zeng, Wang Zhao, and Yong-Jin Liu. Pc-nbv: A point cloud based deep network for efficient next best view plan- ning. In 2020 IEEE/RSJ International Conference on Intel- ligent Robots and Systems (IROS) , pages 7050–7057, 2020. 2

    2020

  36. [44]

    Activermap: Radiance field for active mapping and planning, 2022

    Huangying Zhan, Jiyang Zheng, Yi Xu, Ian Reid, and Hamid Rezatofighi. Activermap: Radiance field for active mapping and planning, 2022. 5, 6

    2022

  37. [45]

    Continuous aerial path planning for 3d ur- ban scene reconstruction

    Han Zhang, Yucong Yao, Ke Xie, Chi-Wing Fu, Hao Zhang, and Hui Huang. Continuous aerial path planning for 3d ur- ban scene reconstruction. ACM Trans. Graph., 40(6):225–1,

  38. [46]

    Offsite aerial path planning for efficient urban scene reconstruction

    Xiaohui Zhou, Ke Xie, Kai Huang, Yilin Liu, Yang Zhou, Minglun Gong, and Hui Huang. Offsite aerial path planning for efficient urban scene reconstruction. ACM Transactions on Graphics (TOG), 39(6):1–16, 2020. 2 10

    2020

  39. [47]

    Overview Our appendix includes the following: • Section 7.2

    Appendix 7.1. Overview Our appendix includes the following: • Section 7.2. More details on model implementation and data setup. • Section 7.3. The chamfer distance graph for the truck cat- egory from OmniObject3D [41] and various visual results of our policy and the coverage b...

  40. [48]

    We use the open-source implementation of this loss and other necessary components provided by the authors on GitHub

    layer as its final layer so that a CORAL loss can be used during training. We use the open-source implementation of this loss and other necessary components provided by the authors on GitHub. Point Cloud Projection. To project our point clouds to query views, we use the Pytorc...

  41. [49]

    The resized objects were made available by the authors on the project GitHub and we used them to calculate the exact scale factor applied to the original OmniObject3D [41] houses

    resized the house objects from OmniObject3D [41] to better fit their problem context. The resized objects were made available by the authors on the project GitHub and we used them to calculate the exact scale factor applied to the original OmniObject3D [41] houses. We use this...

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.