Pith. sign in

REVIEW 4 major objections 6 minor 32 references

LeC$^2$O-NeRF: Learning Continuous and Compact Large-Scale Occupancy for Urban Scenes

T0 review · 4 major / 6 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read A 0.15M-parameter occupancy network can replace the discrete occupancy grid for large-scale NeRF, speeding up training while improving PSNR, SSIM, and LPIPS.

desk verdict Credible engineering result with a real compute-accounting gap: report GPU-hours and add error bars before this is ready. read the letter →

arxiv 2411.11374 v1 pith:V6RNMYQ5 submitted 2024-11-18 cs.CV cs.GR

Zhenxing Mi , Dan Xu This is my paper · ORCID
classification cs.CVcs.GR
keywords occupancynetworklarge-scaleNeRFemptyspaceskippingimbalancedlossdensityscenedecompositionurbanscenesvolumerendering
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

Large-scale neural radiance fields are slowed by the discrete occupancy grid used to skip empty space: grids have fixed resolution, need a tight bounding box, and cost memory to update. This paper claims that a small continuous MLP — an occupancy network with only 0.15M parameters — can learn where a scene is occupied through three self-supervised losses, and that this learned occupancy is more compact, accurate, and smooth than the grid. When frozen and used to guide sampling in existing methods such as Instant-NGP and Switch-NeRF, it speeds up training while improving PSNR, SSIM, and LPIPS. If correct, this would give large-scale NeRF a learned occupancy prior that is not limited by grid resolution or bounding-box regularity.

What carries the argument

LeC$^2$O-NeRF's central object is the occupancy network: a 4-layer MLP that returns normalized occupancy scores over $n$ scene sub-networks and one empty-space network, acting as a continuous, trainable replacement for the discrete occupancy grid. It is driven by three losses: the imbalanced occupancy loss $L_o = (n+v)\left(\frac{f_e p_e}{v} + \sum_{i=1}^n f_i p_i\right)$, which generalizes the balanced load loss of GShard by treating the empty-space network as $v$ virtual experts so that roughly $v/(n+v)$ of points are dispatched as unoccupied; the density loss $L_d = \sigma_e/\sigma_s$, computed with detached densities, which forces the mean density of empty-space points below that of scene points; and the standard volume-rendering loss. Together these let the MLP learn occupancy from the scene's own radiance field without external geometry supervision.

What would settle it

Train the same architecture on one large-scale scene but swap the density ratio $L_d$ for its inverse, so the loss prefers empty-space points to have higher density than scene points; if the occupancy network still converges to a useful surface-near binary occupancy, then the density loss is not the mechanism driving the dispatch, and the paper's explanation would need revision. A second check: evaluate occupancy accuracy against independent geometry from LiDAR or COLMAP instead of the depth maps of a fully-trained Switch-NeRF.

Watch

Extended reading notes

Core claim

The central claim is that occupancy for large-scale NeRF can be encoded as a continuous function learned by a compact MLP, rather than maintained as a discrete momentum-density grid. The occupancy network $O(x)$ outputs $n+1$ normalized values that dispatch each 3D point into one of $n$ scene sub-networks (occupied) or a tiny empty-space network (unoccupied). It is trained end-to-end with the radiance field using the rendering loss plus an imbalanced occupancy loss that controls the occupied/unoccupied ratio and a density loss $L_d = \sigma_e/\sigma_s$ that pushes unoccupied points to smaller density. The paper reports that this 0.15M-parameter network classifies occupancy more accurately than a 2.0M-parameter $128^3$ grid, retains only about 13–16% of points as occupied, and when frozen and used as a sampler it improves the training of Instant-NGP and Switch-NeRF at matched training time, including the occupancy-training time itself.

Load-bearing premise

The occupancy network trusts the density predictions of the very radiance field it is helping to train: if the density field is initially poor, the density loss can reinforce wrong routing decisions instead of correcting them.

Editorial extensions

If this is right

  • Existing grid-based accelerators, including the NerfAcc occupancy grid and Instant-NGP's multi-scale grids, can be replaced by a frozen 0.15M-parameter MLP that needs no per-scene grid-resolution tuning.
  • Occupancy converges quickly (within about 10k–40k training steps), so the network can be frozen early and then used purely as a sampler to accelerate the remaining NeRF training.
  • At aligned wall-clock training time, LeC2O-guided Instant-NGP and Switch-NeRF attain higher PSNR, SSIM, and lower LPIPS than the same methods with an occupancy grid, meaning the speed-up does not sacrifice accuracy.
  • Because the occupancy function is continuous, it represents irregular large-scale urban scenes without the bounding-box and resolution constraints of grids.

Reading between the lines

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

  • If the same occupancy network can be pretrained on a city and then transferred to a new scene, the per-scene 1.6–1.8h occupancy training could be skipped entirely, making the acceleration nearly free at inference time.
  • The trick of treating empty space as $v$ virtual experts in the load-balancing loss could transfer to other sparse 3D representations, such as pruning Gaussians in splatting or sampling in robotic occupancy mapping, wherever most of the volume is empty.
  • The density ratio $L_d$ provides a parameter-free, supervision-free diagnostic for how well an occupancy predictor separates surface from free space, potentially useful in settings where ground-truth occupancy is unavailable.
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 / 6 minor

Summary. This paper proposes LeC2O-NeRF, a method for learning a continuous, compact occupancy field as a small MLP (0.15M parameters) for large-scale NeRF. The occupancy network dispatches each 3D point either to one of n scene sub-networks or to a tiny empty-space network; the model is trained end-to-end with a rendering loss plus an imbalanced occupancy loss (Eq. 1) and a density loss (Eq. 2) that uses detached densities. After pretraining, the frozen occupancy network is used to discard unoccupied points during point sampling in Instant-NGP and Switch-NeRF. Experiments on Mega-NeRF and Block-NeRF report occupancy classification metrics against a depth-derived ground truth and image reconstruction metrics with wall-clock training time aligned between methods.

Significance. If the claims hold, the paper provides a practical alternative to occupancy grids for large-scale NeRF: a compact continuous predictor that retains roughly 13--16% of points, improves recall substantially over the grid, and improves PSNR/SSIM/LPIPS on most scenes under wall-clock alignment. The core derivation of the imbalanced loss is sound, the ablations are informative, and the paper gives useful occupancy diagnostics (point proportions, density/alpha ratios, memory usage). The main limitation is that the speed claim is made on wall-clock time where GPU counts differ between occupancy pretraining and target NeRF training; until total GPU-hours are reported, the 'speed up without sacrificing accuracy' conclusion is not established. The occupancy ground-truth proxy is generated by the same model family used to train the occupancy network, which weakens the absolute accuracy claims.

major comments (4)
  1. [§4.3, §4.4, Table 3] The aligned-time comparison does not control for total GPU compute. Occupancy pretraining is reported as 1.6--1.8 h in §4.3 without specifying the number of GPUs, while §4.4 states that INGP is trained on 2 RTX 3090 GPUs and Switch-NeRF on 8. Since the occupancy network is fitted with 8 scene sub-networks and then reused for both target methods, it is plausible that pretraining uses 8 GPUs; under that assumption INGP+Ours costs roughly 8×1.7 + 2×(20−1.7) ≈ 50.2 GPU-hours versus 40 GPU-hours for vanilla INGP. The reported wall-clock tie therefore does not imply a compute tie. Please report GPU-hours for every configuration (including occupancy pretraining) or run all comparisons on identical GPU counts, and state the alignment protocol explicitly. Without this, the central claim that the method can speed up state-of-the-art NeRF methods without sacrificing accuracy is not supported.
  2. [§4.2, Table 1] The occupancy accuracy ground truth is derived from depth maps of a fully trained Switch-NeRF, the same model family that the occupancy network is trained with. A learned occupancy model trained jointly with a Switch-NeRF-like radiance field may be systematically favored by this reference, so the reported Accuracy/Precision/Recall/F1 improvements over the grid are not an independent measure of geometric quality. Please corroborate the occupancy evaluation with an external depth or LiDAR source, or at least report how the metrics change with a different reference model, and state this limitation explicitly in the paper.
  3. [§3.3, Eq. (2)] The density loss supervises occupancy with the model's own density predictions, detached, and this signal is assumed to be reliable. Early in training, when the radiance field is poor, these densities may be unreliable and the loss could reinforce routing errors rather than correct them. The paper should include an experiment in which Ld is applied only after a warm-up period or in which the detached densities come from a separate frozen reference, and should report the ratio σe/σs at very early steps; the current Figure 8d starts at 1k steps and does not show the noisiest regime.
  4. [Table 3, Residence row] The statement in §4.4 that the method consistently outperforms Switch, NGP, and Switch+Grid is contradicted on the Residence scene, where Switch+Ours has PSNR 22.10 versus 22.18 for Switch+Grid, even though SSIM and LPIPS are better. No error bars or multiple seeds are reported anywhere in the tables. Please report standard deviations over at least three seeds and either soften the consistency claim or identify the metric (e.g., LPIPS or an aggregate) on which consistency is claimed.
minor comments (6)
  1. [§3.3] The sentence 'Our empty space network O should secure more 3D points' should refer to the occupancy network, not the empty space network; the empty space network is the one whose fraction should increase.
  2. [§3.3, Eq. (2)] The ratio σe/σs can have a near-zero denominator during early training; add a small positive constant to the denominator for numerical stability and state this in the text.
  3. [§4.3] The sampling counts are ambiguous: the text says '512 points for each ray' for occupancy training and later 'usually sample 128 samples along a ray' for guided training; clarify which stage each number applies to.
  4. [Table 1] The 'Para. Number' row compares only the occupancy representation parameters; state whether the scene and empty sub-network parameters are excluded, and give the grid resolution (e.g., 128^3 cells) in the caption for context.
  5. [§4.4] There is a typo in 'Instan-NGP' in the paragraph describing Table 3; it should be 'Instant-NGP'.
  6. [Figure 5] The caption label 'GridOurs' should be written 'Grid / Ours' to avoid confusion between the two columns.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the rendering benchmarks are independent of the occupancy network's fitted parameters, and the occupancy ratio is an explicitly stated design target, not a hidden prediction.

full rationale

The paper's central claims are validated with held-out novel-view metrics (PSNR/SSIM/LPIPS) on Instant-NGP and Switch-NeRF, both of which are trained after the occupancy network is frozen and are not used to fit the occupancy network. Those external benchmarks make the main speed/accuracy claim independently testable. The occupancy-ratio numbers in Table 1 are indeed a direct consequence of the imbalanced occupancy loss: with n=8 and v=80, the loss targets an empty-space share of v/(n+v), about 91%, and the paper says so explicitly in Section 3.3: 'These values make the occupancy network dispatch about 85% points to the empty space network.' Reporting the resulting 13-16% occupied ratio is therefore reporting a designed constraint, not a prediction derived from hidden fitted inputs. The nontrivial finding is that render quality improves at this enforced sparsity, which is checked against held-out images and against the occupancy grid at matched time. The density loss Ld = sigma_e / sigma_s uses the same network's detached density outputs as a soft training signal; this is a self-supervised objective and can pose a training-stability risk if early densities are poor, but it is not a circular derivation because no reported result is defined as the minimizer of that loss. The occupancy-accuracy evaluation uses depth maps from a fully-trained Switch-NeRF as a geometric reference; this is a self-citation by the same authors, but the reference model was trained independently of the proposed occupancy network and is not used to fit the occupancy parameters, so it functions as an external benchmark rather than as an input-output identity. The skeptic's GPU-hour alignment concern is a legitimate fairness/correctness question about whether wall-clock comparisons hide extra parallel compute, but it is not a circularity of the derivation chain. No equation in the paper sets a reported result equal to its own input by construction, and no load-bearing premise is justified solely by an unverified self-citation.

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

The method relies on two domain priors (scene sparsity, small empty-space density) formalized as losses, plus hyperparameters v, n, and loss weights that were tuned by the authors. The learned modules (occupancy network and empty network) are new components, but they are not independently verified outside the paper's benchmarks.

free parameters (4)
  • virtual empty sub-network count v = 80 for Mega-NeRF, 40 for Block-NeRF
    Controls the target empty-to-occupied ratio: v/(n+v) = 80/88 ≈ 90.9%, chosen by hand to approximate the sparsity prior. The paper uses this to force ~85-90% of points to the empty network.
  • scene sub-network count n = 8
    Number of scene experts in the MoE architecture, chosen manually to balance capacity and memory.
  • loss weights wo, wd = wo=0.0005, wd=0.1 (Mega-NeRF), wd=0.005 (Block-NeRF)
    Weights for the imbalanced occupancy loss and density loss in the final loss Lf, tuned by the authors.
  • occupancy network width = 256
    Channel number for the main layers of the occupancy MLP, chosen by hand.
assumptions (4)
  • domain assumption Most 3D points in a large-scale scene are unoccupied (sparsity prior).
    Motivates the imbalanced occupancy loss and the choice v >> n; the paper states 'most of 3D scene points are unoccupied' in Sections 1 and 3.3.
  • domain assumption Density of unoccupied points is much smaller than that of occupied points.
    Used to design the density loss Ld in Eq. 2; the paper states 'the density of an unoccupied point is much smaller than that of an occupied point' in Sections 1 and 3.3.
  • standard math The balanced loss from GShard (Lepikhin et al., 2021) transfers to the occupancy-dispatch setting.
    Eq. 1 directly extends the balanced loss Lb = n * sum(f_i * p_i) to n scene experts plus v virtual empty experts; the derivation is straightforward but relies on the cited result.
  • domain assumption Depth maps predicted by a fully-trained Switch-NeRF provide a good reference for ground-truth occupancy.
    Section 4.2 states 'we use it as a good reference for evaluation'; the occupancy metrics are computed against this proxy rather than an independent sensor.
invented entities (2)
  • Occupancy network O (4-layer MLP)
    purpose: Classifies 3D points as occupied or unoccupied and dispatches them to either the scene network or the empty space network.
    Introduced in this paper; validated only through the paper's own rendering benchmarks and proxy occupancy metrics, with no external geometric ground truth.
  • Empty space network Ee (identity layer + prediction head)
    purpose: Encodes unoccupied points with minimal capacity, producing smooth and low densities.
    Architectural invention of this paper; ablations show it is important, but its behavior is not validated outside the paper's datasets.

how reviews work

0 comments
Cite this review

Pith. "Pith review of LeC$^2$O-NeRF: Learning Continuous and Compact Large-Scale Occupancy for Urban Scenes." pith.science (2026). https://pith.science/paper/V6RNMYQ5

@misc{pith2026241111374,
  author       = {Pith},
  title        = {Pith review of: LeC$^2$O-NeRF: Learning Continuous and Compact Large-Scale Occupancy for Urban Scenes},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/V6RNMYQ5}},
  note         = {Machine review of arXiv:2411.11374}
}
read the original abstract

In NeRF, a critical problem is to effectively estimate the occupancy to guide empty-space skipping and point sampling. Grid-based methods work well for small-scale scenes. However, on large-scale scenes, they are limited by predefined bounding boxes, grid resolutions, and high memory usage for grid updates, and thus struggle to speed up training for large-scale, irregularly bounded and complex urban scenes without sacrificing accuracy. In this paper, we propose to learn a continuous and compact large-scale occupancy network, which can classify 3D points as occupied or unoccupied points. We train this occupancy network end-to-end together with the radiance field in a self-supervised manner by three designs. First, we propose a novel imbalanced occupancy loss to regularize the occupancy network. It makes the occupancy network effectively control the ratio of unoccupied and occupied points, motivated by the prior that most of 3D scene points are unoccupied. Second, we design an imbalanced architecture containing a large scene network and a small empty space network to separately encode occupied and unoccupied points classified by the occupancy network. This imbalanced structure can effectively model the imbalanced nature of occupied and unoccupied regions. Third, we design an explicit density loss to guide the occupancy network, making the density of unoccupied points smaller. As far as we know, we are the first to learn a continuous and compact occupancy of large-scale NeRF by a network. In our experiments, our occupancy network can quickly learn more compact, accurate and smooth occupancy compared to the occupancy grid. With our learned occupancy as guidance for empty space skipping on challenging large-scale benchmarks, our method consistently obtains higher accuracy compared to the occupancy grid, and our method can speed up state-of-the-art NeRF methods without sacrificing accuracy.

Figures

Figures reproduced from arXiv: 2411.11374 by the authors.

Figure 1
Figure 1. Differences between the occupancy grid (Li et al., 2022; M [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Our proposed LeC2O-NeRF. The occupancy network predicts the occupancy of each point and dispatches them into different sub-networks. x1 and x2 go through the occupancy network and are dispatched to the empty space network and the scene network, according to their occupancy values. The occupancy can be trained end-to-end together with the NeRF network by multiplying occupancy values on the output of sub-networks. If … view at source ↗
Figure 3
Figure 3. (a) The computation of the imbalanced occupancy loss and the density loss from occu [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (6 more)
Figure 4
Figure 4. Figure 4: The visualization of our occupancy and the grid occupancy as point clouds. Our predicted [PITH_FULL_IMAGE:figures/full_fig_p009_4.png]
Figure 5
Figure 5. Figure 5: The rendered images of our occupancy network and the occupancy grid based on Switch [PITH_FULL_IMAGE:figures/full_fig_p009_5.png]
Figure 6
Figure 6. Figure 6: (a) The occupied points and images with the imbalanced and balanced networks. Our [PITH_FULL_IMAGE:figures/full_fig_p010_6.png]
Figure 7
Figure 7. Figure 7: The point clouds dispatched to the scene network and the empty space network at each [PITH_FULL_IMAGE:figures/full_fig_p011_7.png]
Figure 8
Figure 8. Figure 8: The statistics of the scene network S and the empty space network Ee at different training steps. (a) The portion of the points in S and Ee. (b) (c) The mean density values and alpha values of the points in S and Ee. (d) The density value ratio and alpha value ratio be…
Figure 9
Figure 9. Figure 9: Analysis of training time and accuracy: our method (S+Ours) demonstrates remarkable [PITH_FULL_IMAGE:figures/full_fig_p011_9.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

32 extracted references · 14 canonical work pages

  1. [1]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION format.date year duplicate empty "emp...

  2. [2]

    Barron, Ben Mildenhall, Matthew Tancik, Peter Hedman, Ricardo Martin-Brualla, and Pratul P

    Jonathan T. Barron, Ben Mildenhall, Matthew Tancik, Peter Hedman, Ricardo Martin-Brualla, and Pratul P. Srinivasan. Mip-nerf: A multiscale representation for anti-aliasing neural radiance fields. ICCV, 2021

    2021

  3. [3]

    Barron, Ben Mildenhall, Dor Verbin, Pratul P

    Jonathan T. Barron, Ben Mildenhall, Dor Verbin, Pratul P. Srinivasan, and Peter Hedman. Mip-nerf 360: Unbounded anti-aliased neural radiance fields. In CVPR, 2022

    2022

  4. [4]

    Chan, Connor Z

    Eric R. Chan, Connor Z. Lin, Matthew A. Chan, Koki Nagano, Boxiao Pan, Shalini De Mello, Orazio Gallo, Leonidas Guibas, Jonathan Tremblay, Sameh Khamis, Tero Karras, and Gordon Wetzstein. Efficient geometry-aware 3D generative adversarial networks. In CVPR, 2022

    2022

  5. [5]

    Tensorf: Tensorial radiance fields

    Anpei Chen, Zexiang Xu, Andreas Geiger, Jingyi Yu, and Hao Su. Tensorf: Tensorial radiance fields. In ECCV, 2022

    2022

  6. [6]

    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. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp.\ 5501--5510, June 2022

    work page 2022

  7. [7]

    Efficientnerf efficient neural radiance fields

    Tao Hu, Shu Liu, Yilun Chen, Tiancheng Shen, and Jiaya Jia. Efficientnerf efficient neural radiance fields. In CVPR, 2022

    work page 2022

  8. [8]

    Motion-oriented compositional neural radiance fields for monocular dynamic human modeling

    Kim Jaehyeok, Wee Dongyoon, and Dan Xu. Motion-oriented compositional neural radiance fields for monocular dynamic human modeling. In ECCV, 2024

    work page 2024

Show all 32 references
  1. [9]

    3d gaussian splatting for real-time radiance field rendering

    Bernhard Kerbl, Georgios Kopanas, Thomas Leimk \"u hler, and George Drettakis. 3d gaussian splatting for real-time radiance field rendering. ACM Transactions on Graphics, 42 0 (4), July 2023. URL https://repo-sam.inria.fr/fungraph/3d-gaussian-splatting/

    2023

  2. [10]

    Gshard: Scaling giant models with conditional computation and automatic sharding

    Dmitry Lepikhin, HyoukJoong Lee, Yuanzhong Xu, Dehao Chen, Orhan Firat, Yanping Huang, Maxim Krikun, Noam Shazeer, and Zhifeng Chen. Gshard: Scaling giant models with conditional computation and automatic sharding. In ICLR, 2021

    2021

  3. [11]

    Nerfacc: A general nerf accleration toolbox

    Ruilong Li, Matthew Tancik, and Angjoo Kanazawa. Nerfacc: A general nerf accleration toolbox. arXiv preprint arXiv:2210.04847, 2022

    2022 arXiv

  4. [12]

    Neural sparse voxel fields

    Lingjie Liu, Jiatao Gu, Kyaw Zaw Lin, Tat-Seng Chua, and Christian Theobalt. Neural sparse voxel fields. NeurIPS, 2020

    2020

  5. [13]

    Nerf in the wild: Neural radiance fields for unconstrained photo collections

    Ricardo Martin-Brualla, Noha Radwan, Mehdi SM Sajjadi, Jonathan T Barron, Alexey Dosovitskiy, and Daniel Duckworth. Nerf in the wild: Neural radiance fields for unconstrained photo collections. In CVPR, 2021

    2021

  6. [14]

    Switch-ne RF : Learning scene decomposition with mixture of experts for large-scale neural radiance fields

    Zhenxing MI and Dan Xu. Switch-ne RF : Learning scene decomposition with mixture of experts for large-scale neural radiance fields. In The Eleventh International Conference on Learning Representations, 2023. URL https://openreview.net/forum?id=PQ2zoIZqvm

    2023

  7. [15]

    Srinivasan, Matthew Tancik, Jonathan T

    Ben Mildenhall, Pratul P. Srinivasan, Matthew Tancik, Jonathan T. Barron, Ravi Ramamoorthi, and Ren Ng. Nerf: Representing scenes as neural radiance fields for view synthesis. In ECCV, 2020

    2020

  8. [16]

    Instant neural graphics primitives with a multiresolution hash encoding

    Thomas M\"uller, Alex Evans, Christoph Schied, and Alexander Keller. Instant neural graphics primitives with a multiresolution hash encoding. ACM Trans. Graph., 41 0 (4): 0 102:1--102:15, 2022

    2022

  9. [17]

    Implicit event-rgbd neural slam

    Delin Qu, Chi Yan, Dong Wang, Jie Yin, Qizhi Chen, Dan Xu, Yiting Zhang, Bin Zhao, and Xuelong Li. Implicit event-rgbd neural slam. In CVPR, 2024

    2024

  10. [18]

    Srinivasan, Jonathan T

    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. In CVPR, 2022

    2022

  11. [19]

    Sign-agnostic conet: Learning implicit surface reconstructions by sign-agnostic optimization of convolutional occupancy networks

    Jiapeng Tang, Jiabao Lei, Dan Xu, Feiying Ma, Kui Jia, and Lei Zhang. Sign-agnostic conet: Learning implicit surface reconstructions by sign-agnostic optimization of convolutional occupancy networks. In ICCV, 2021

    2021

  12. [20]

    Mega-nerf: Scalable construction of large-scale nerfs for virtual fly-throughs

    Haithem Turki, Deva Ramanan, and Mahadev Satyanarayanan. Mega-nerf: Scalable construction of large-scale nerfs for virtual fly-throughs. In CVPR, 2022

    2022

  13. [21]

    Learning unified decompositional and compositional nerf for editable novel view synthesis

    Yuxin Wang, Wayne Wu, and Dan Xu. Learning unified decompositional and compositional nerf for editable novel view synthesis. In ICCV, 2023

    2023

  14. [22]

    Image quality assessment: from error visibility to structural similarity

    Zhou Wang, Alan C Bovik, Hamid R Sheikh, and Eero P Simoncelli. Image quality assessment: from error visibility to structural similarity. IEEE transactions on image processing, 2004

    2004

  15. [23]

    Pygs: Large-scale scene representation with pyramidal 3d gaussian splatting

    Zipeng Wang and Dan Xu. Pygs: Large-scale scene representation with pyramidal 3d gaussian splatting. arXiv preprint arXiv:2405.16829, 2024

    2024 arXiv

  16. [24]

    Grid-guided neural radiance fields for large urban scenes

    Linning Xu, Yuanbo Xiangli, Sida Peng, Xingang Pan, Nanxuan Zhao, Christian Theobalt, Bo Dai, and Dahua Lin. Grid-guided neural radiance fields for large urban scenes. In CVPR, 2023

    2023

  17. [25]

    Point-nerf: Point-based neural radiance fields

    Qiangeng Xu, Zexiang Xu, Julien Philip, Sai Bi, Zhixin Shu, Kalyan Sunkavalli, and Ulrich Neumann. Point-nerf: Point-based neural radiance fields. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp.\ 5438--5448, 2022

    2022

  18. [26]

    The unreasonable effectiveness of deep features as a perceptual metric

    Richard Zhang, Phillip Isola, Alexei A Efros, Eli Shechtman, and Oliver Wang. The unreasonable effectiveness of deep features as a perceptual metric. In CVPR, 2018

    2018

  19. [27]

    Nerflets: Local radiance fields for efficient structure-aware 3d scene representation from 2d supervision

    Xiaoshuai Zhang, Abhijit Kundu, Thomas Funkhouser, Leonidas Guibas, Hao Su, and Kyle Genova. Nerflets: Local radiance fields for efficient structure-aware 3d scene representation from 2d supervision. CVPR, 2023 a

    2023

  20. [28]

    Efficient large-scale scene representation with a hybrid of high-resolution grid and plane features

    Yuqi Zhang, Guanying Chen, and Shuguang Cui. Efficient large-scale scene representation with a hybrid of high-resolution grid and plane features. arXiv preprint arXiv:2303.03003, 2023 b

    2023 arXiv

  21. [29]

    Cvt-xrf: Contrastive in-voxel transformer for 3d consistent radiance fields from sparse inputs

    Yingji Zhong, Lanqing Hong, Zhenguo Li, and Dan Xu. Cvt-xrf: Contrastive in-voxel transformer for 3d consistent radiance fields from sparse inputs. In CVPR, 2024

    2024

  22. [30]

    @esa (Ref

    \@ifxundefined[1] #1\@undefined \@firstoftwo \@secondoftwo \@ifnum[1] #1 \@firstoftwo \@secondoftwo \@ifx[1] #1 \@firstoftwo \@secondoftwo [2] @ #1 \@temptokena #2 #1 @ \@temptokena \@ifclassloaded agu2001 natbib The agu2001 class already includes natbib coding, so you should ...

  23. [31]

    \@lbibitem[] @bibitem@first@sw\@secondoftwo \@lbibitem[#1]#2 \@extra@b@citeb \@ifundefined br@#2\@extra@b@citeb \@namedef br@#2 \@nameuse br@#2\@extra@b@citeb \@ifundefined b@#2\@extra@b@citeb @num @parse #2 @tmp #1 NAT@b@open@#2 NAT@b@shut@#2 \@ifnum @merge>\@ne @bibitem@firs...

  24. [32]

    @open @close @open @close and [1] URL: #1 \@ifundefined chapter * \@mkboth \@ifxundefined @sectionbib * \@mkboth * \@mkboth\@gobbletwo \@ifclassloaded amsart * \@ifclassloaded amsbook * \@ifxundefined @heading @heading NAT@ctr thebibliography [1] @ \@biblabel @NAT@ctr \@bibset...

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.