Pith. sign in

REVIEW 4 major objections 5 minor 44 references

NeurNCD: Novel Class Discovery via Implicit Neural Representation

T0 review · 4 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read The paper claims that NeurNCD, built on an Embedding-NeRF implicit scene representation, outperforms explicit-map baselines in both known-class segmentation and novel-class discovery on NYUv2 and Replica without dense labels.

desk verdict Algorithm 1 as written zeroes out cosine similarities between distinct segments, so the described method cannot produce the reported clustering results—though the NeRF-for-NCD idea itself merits a second look. read the letter →

arxiv 2506.06412 v1 pith:7ZWQYVA5 submitted 2025-06-06 cs.LG cs.CV

Junming Wang , Yi Shi This is my paper
classification cs.LGcs.CV
keywords NovelclassdiscoveryImplicitneuralrepresentationradiancefieldsSemanticsegmentationVisualembeddingspaceFeaturemodulationMarkovclusteringOpen-worldperception
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

NeurNCD is a framework for discovering object classes that a segmentation model was never trained on, using an implicit neural representation instead of an explicit 3D segmentation map. The paper argues that explicit maps are discrete, noisy, and hole-prone, and that replacing them with an Embedding-NeRF, trained with KL divergence, produces globally consistent semantic embeddings and entropy. These features are attached to geometrically segmented regions and clustered with Markov clustering, yielding known-class segmentation and novel-class discovery without dense supervision or human interaction. The paper reports state-of-the-art results on NYUv2 and Replica, including a mean IoU of 51.29 on NYUv2 and a known-class mIoU of 81.3 on Replica with no annotations.

What carries the argument

The load-bearing component is Embedding-NeRF, an augmented neural radiance field whose multilayer perceptron maps a 3D position to color, density, and semantic embedding logits. It is trained with a KL-divergence embedding loss, which encourages the rendered embedding to match the pre-trained segmentation network's embedding whether the input is clean or noisy, plus a photometric loss; this produces globally consistent semantic embeddings and entropy across views. These features are coupled with geometric segmentation into convex sub-instance-level segments, then fused by feature modulation and grouped by Markov clustering, so that over-segmented geometric pieces of the same known or novel class are reunited.

What would settle it

Implement Algorithm 1 and Equation (12) on two disjoint segments from any depth frame: because the masks are disjoint, the cosine similarity is zero for every pair, the Markov transition matrix is degenerate, and the reported clusters cannot emerge, so a working implementation must be doing something other than what the paper describes.

Watch

Extended reading notes

Core claim

The central claim is that implicit neural representations can substitute for explicitly constructed 3D segmentation maps in open-world semantic segmentation. NeurNCD extends NeRF so that its MLP outputs radiance, volume density, and semantic embedding logits, and renders these embeddings along rays; the rendered embeddings are trained to match embeddings from a pre-trained RGB-D segmentation network via KL divergence, while a photometric loss maintains scene appearance. The resulting Embedding-NeRF produces a hole-free, low-noise semantic embedding field and an entropy field, which are queried by convex sub-instance-level segments from depth-based geometric segmentation, concatenated in a feature modulation step, and clustered with Markov clustering based on cosine similarity. The paper reports that this pipeline segments known classes more accurately than the explicit-map baseline and also discovers novel classes, outperforming state-of-the-art approaches on both NYUv2 and Replica.

Load-bearing premise

The entire clustering pipeline rests on the assumption that the segment feature vectors produced by Feature Modulation have meaningful pairwise cosine similarities, but as written each vector is nonzero only at its own mask pixels, so distinct segments share no nonzero entries and every pairwise cosine similarity is zero.

Editorial extensions

If this is right

  • Open-world scene understanding can be built on a single implicit representation that simultaneously renders appearance, geometry, and semantics, rather than maintaining a separate explicit segmentation map.
  • The same pipeline works in closed-world settings, so a single framework could serve both known-class segmentation and incremental discovery without switching representations.
  • Because the method needs only a pre-trained 2D segmentation network plus posed RGB-D frames, it removes the cost of dense pixel annotation and interactive labeling for a new scene.
  • Training time is reported at roughly eight hours per scene on a single GPU, comparable to a supervised semantic NeRF, suggesting that the discovery capability does not add a large training overhead.
  • The per-class improvements on NYUv2 and the Replica comparison against sparse-label and interactive baselines indicate that implicit feature aggregation can compensate for errors in geometric over-segmentation.

Reading between the lines

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

  • The KL-divergence training recipe may transfer to other implicit scene representations, such as hash-grid or Gaussian-splatting fields; testing NeurNCD on those backbones would reveal whether the gains come from the implicit representation itself or from the specific MLP architecture.
  • The method inherits a reliance on convexity-based geometric segmentation, so heavily non-convex or heavily occluded objects are likely to remain fragmented; augmenting geometric segmentation with learned grouping could raise novel-class mIoU further.
  • The entropy field produced by Embedding-NeRF could double as an uncertainty signal for active learning, letting an embodied agent request human labels only for high-entropy regions and extend the class vocabulary incrementally.
  • If the feature modulation step in practice aggregates embeddings per segment rather than assigning them by mask locations, then the published Algorithm 1 and Equation (12) do not describe the implemented clustering input, and the method should be re-specified as a general per-segment feature aggregation recipe.
Share X Bluesky LinkedIn Reddit HN

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 NeurNCD, a framework for novel class discovery that replaces explicit 3D segmentation maps with an implicit neural representation called Embedding-NeRF. The pipeline extracts semantic embeddings from a pretrained RGB-D segmentation network (ESANet), trains a per-scene NeRF with a photometric loss and a KL-divergence embedding loss, segments depth images into convex sub-instance segments, queries the NeRF output to obtain embeddings and entropy for each segment, concatenates these into segment-level feature vectors, and finally runs Markov clustering based on cosine similarity. The authors report state-of-the-art results on NYUv2 and Replica for both known-class segmentation and novel-class discovery, and include ablations of the major components.

Significance. If the method were correctly specified and reproducible, the paper would introduce a novel use of implicit neural representations for open-world semantic segmentation and novel class discovery, which is an interesting direction for embodied perception. The paper also attempts a principled KL-divergence objective for transferring semantic information into a NeRF and offers quantitative comparisons against both explicit-map baselines and supervised implicit baselines. However, the paper does not release code, and the central algorithmic step as written is degenerate: Algorithm 1 constructs segment vectors with disjoint nonzero supports, making the cosine similarities in Eq. (12) identically zero. The reported results therefore cannot be produced by the described method, and the contribution cannot be verified in its current form.

major comments (4)
  1. [Section 3.3, Algorithm 1 and Eq. (12)] The feature modulation step as written makes the clustering input degenerate. Geometric segmentation partitions the depth image, so masks of distinct segments are disjoint; Algorithm 1 (lines 15–16) copies combined features only into the positions of each segment's own mask and leaves all other entries zero. After the reshape on line 21, for any i != j the dot product H_i · H_j in Eq. (12) contains only terms where one factor is zero, so the cosine similarity is identically zero for every pair of distinct segments. Markov clustering then receives a graph with no positive edges and cannot merge over-segmented pieces, so the reported improvement from clustering cannot be produced by the described method. If the actual implementation aggregates features per segment (e.g., by averaging), that operation is neither documented nor released as code, and the paper would need to specify and validate it before the claims can be assessed.
  2. [Section 3.1, Eq. (9)] The entropy term used in feature modulation is not defined in a reproducible way. The sentence introducing Eq. (9) states that semantic embedding E_i 'obtained by fusion' is sent to 'the two upsampling modules' to obtain U_i^o, but no fusion mechanism, upsampling architecture, or input–output relation is specified anywhere in the paper. Moreover, Eq. (9) presents epsilon_i as a scalar while Algorithm 1 and Eq. (11) treat entropy as a per-pixel vector concatenated with the embedding; the discrepancy makes the exact feature vector H_j^i ambiguous.
  3. [Section 4.2, NYUv2 evaluation] The evaluation protocol is transductive in a way that is not compared fairly with the baselines. The paper trains a separate Embedding-NeRF per scene on the same images that are later evaluated ('the official split of 654 images is used for testing'; 'train a separate Embedding-NeRF model for each scene'), so the model has already fitted the test frames before the segmentation metrics are computed. The comparison with [19] and [27], which are evaluated in an incremental/online setting, therefore conflates scene memorization with generalization; a held-out view or scene split is needed to support the data-efficiency claim.
  4. [Section 4.5, Table 3] The ablation table does not isolate the contribution of entropy (EP): the rows labeled with four checkmarks are ambiguous about which components are active (GS, PSSN, EP, SE, or a different combination), and the text's claim that 'incorporating only entropy features into segments yields inferior outcomes' is not backed by any row with only EP added. The reported ablation for the central components is therefore not verifiable from the table as printed.
minor comments (5)
  1. [Section 3.4] The text attributes Markov clustering to reference [37], but [37] is a survey and the deep MCL method is [39]; the exact algorithm and the hand-tuned parameter set (beyond inflation=12) should be identified for reproducibility.
  2. [Section 4.2] The loss weight lambda in Eq. (8) is never given; without its value and the schedule, the KL-loss contribution cannot be reproduced.
  3. [Figure 2 caption and keywords] 'leverge' should be 'leverage', and the keywords list 'Neural Radiation Field' should be 'Neural Radiance Field'.
  4. [Tables 1 and 2] The per-class numbers and mIoU totals would benefit from standard deviations over multiple runs and from a clearer statement of which classes are known versus novel; Table 2's 'Our -81.3 50.6 89.1 89.7' row also contains an unexplained dash.
  5. [Section 3.1, Eq. (6)] Eq. (6) and the surrounding text do not define t_k, delta_k, or the coarse/fine sampling scheme; the paper should either define these or cite the original NeRF formulation with the needed notation.

Circularity Check

1 steps flagged · score 6.0 of 10

Algorithm 1's masked feature vectors make Eq. 12's cross-segment dot products zero by construction, so the MCL clustering step and the reported improvements cannot follow from the described method.

  1. self definitional [Section 3.3 (Algorithm 1) and Section 3.4 (Eq. 12)]
    "if mask[h,w]==1 then H[i,h,w,:S_emb+1]←combined[h,w] (Algorithm 1, lines 15-16); Eq. 12: Similarity(s_m^i, s_n^i) = H_m^i H_n^i / (||H_m^i|| ||H_n^i||), where m≠n."

    Algorithm 1 builds each segment feature H_i by writing embedding and entropy values only at pixels where that segment's mask equals 1, then reshaping H to (N, -1). Because geometric segmentation (Eq. 10) partitions the depth image into disjoint segments, rows i and j of the reshaped H have non-zero entries on disjoint pixel sets. Every term in the dot product H_m^i · H_n^i for m≠n therefore has at least one zero factor, making the cosine similarity in Eq. 12 identically zero for every distinct segment pair. Feeding this all-zero-off-diagonal similarity matrix to Markov clustering cannot merge over-segmented fragments; each segment remains isolated.

full rationale

The central derivation from per-segment features to clustering is self-definitionally degenerate. Algorithm 1's feature modulation defines H_j^i as a masked per-pixel vector whose non-zero support is exactly segment j's mask. Since the segments from Eq. 10 are disjoint, the dot product in Eq. 12 is zero for all pairs of distinct segments by construction; Markov clustering then cannot associate any two segments. Thus the paper's central claim—that NeurNCD significantly outperforms prior work by fusing Embedding-NeRF features and clustering—does not follow from the stated algorithm. This is not a matter of author intent or self-citation; it is a specific reduction of Eq. 12 to a degenerate constant given Algorithm 1. The paper's self-citations are not load-bearing, and the broad concept of using a neural field to smooth pre-trained embeddings is not inherently circular. However, because the clustering step, a core component of the proposed method, is forced to produce isolated clusters by the paper's own definitions, the reported experimental outcome cannot be reproduced from the described method. Score 6 reflects this construction-level circularity in the central prediction pipeline.

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

The central claim depends on three nested assumptions: the teacher features are informative for unknown classes, the KL loss transfers them faithfully into the NeRF, and the NeRF-rendered features improve clustering. None of these is isolated in the experiments. The only explicit free parameter affecting the final output is the MCL inflation, hand-tuned on a single scene.

free parameters (2)
  • MCL inflation = 12
    Hand-tuned on the kitchen_0004 scene of NYUv2 and then applied to all scenes (Section 3.4). This parameter directly controls cluster granularity and affects the final mIoU.
  • Loss weight lambda = not reported
    The total loss is L = L_p + λ L_e (Eq. 8), but λ is never specified. It is a hand-chosen trade-off between photometric and embedding losses.
assumptions (4)
  • domain assumption Pre-trained ESANet embeddings trained on 9 known classes contain enough structure for clustering both known and novel classes.
    The entire method relies on the teacher network's feature space to separate novel classes, although the teacher never saw those classes during training.
  • ad hoc to paper The KL divergence between softmax-normalized embeddings is a suitable objective for transferring semantic information from 2D to 3D.
    The paper states KL minimization 'reduces distances between genuine prototypes' etc., but provides no derivation or evidence for these properties over other losses such as L2 or cross-entropy.
  • domain assumption Implicit neural representation yields 'low-noise, hole-free' 3D structures that improve clustering over explicit maps.
    This is the motivating assumption of the paper; it is not directly tested because the ablation does not compare against an explicit-map variant using the same teacher features.
  • domain assumption The geometric segmentation method assumes real-world objects have convex surfaces.
    This assumption is inherited from [9] and [19]; it limits performance on non-convex objects as acknowledged in Section 3.2.

how reviews work

0 comments
Cite this review

Pith. "Pith review of NeurNCD: Novel Class Discovery via Implicit Neural Representation." pith.science (2026). https://pith.science/paper/7ZWQYVA5

@misc{pith2026250606412,
  author       = {Pith},
  title        = {Pith review of: NeurNCD: Novel Class Discovery via Implicit Neural Representation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/7ZWQYVA5}},
  note         = {Machine review of arXiv:2506.06412}
}
read the original abstract

Discovering novel classes in open-world settings is crucial for real-world applications. Traditional explicit representations, such as object descriptors or 3D segmentation maps, are constrained by their discrete, hole-prone, and noisy nature, which hinders accurate novel class discovery. To address these challenges, we introduce NeurNCD, the first versatile and data-efficient framework for novel class discovery that employs the meticulously designed Embedding-NeRF model combined with KL divergence as a substitute for traditional explicit 3D segmentation maps to aggregate semantic embedding and entropy in visual embedding space. NeurNCD also integrates several key components, including feature query, feature modulation and clustering, facilitating efficient feature augmentation and information exchange between the pre-trained semantic segmentation network and implicit neural representations. As a result, our framework achieves superior segmentation performance in both open and closed-world settings without relying on densely labelled datasets for supervised training or human interaction to generate sparse label supervision. Extensive experiments demonstrate that our method significantly outperforms state-of-the-art approaches on the NYUv2 and Replica datasets.

Figures

Figures reproduced from arXiv: 2506.06412 by the authors.

Figure 1
Figure 1. NeurNCD leverages implicit neural representation, replacing traditional explicit 3D segmentation maps[ [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. An overview of our method. For input RGB-D image, the pre-trained semantic segmentation network [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Quantitative results for known and unknown classes in the NYUv2 dataset. With the powerful feature propagation [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Embedding-NeRF 3D position (𝑥, 𝑦, 𝑧) and viewing direction (𝜃, 𝜙) are fed into the network after positional en￾coding (PE). Volume density 𝜎 and semantic embedding 𝑒 are functions of 3D position while colours 𝑐 additionally depend on viewing direction. In the context o…
Figure 5
Figure 5. Figure 5: Results on the NYUv2 dataset. The third column [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 6
Figure 6. Figure 6: Results on the Replica dataset. The third and fourth [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

44 extracted references · 34 canonical work pages

  1. [19]

    Yoshikatsu Nakajima, Byeongkeun Kang, Hideo Saito, and Kris Kitani. 2019. Incremental class discovery for semantic segmentation with RGBD sensing. In Proceedings of the IEEE/CVF international conference on computer vision. 972–981

    work page 2019

  2. [27]

    Keisuke Tateno, Federico Tombari, and Nassir Navab. 2015. Real-time and scal- able incremental segmentation on dense slam. In2015 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). IEEE, 4465–4472

    work page 2015

  3. [1]

    Barron, Ben Mildenhall, Dor Verbin, Pratul P

    Jonathan T. Barron, Ben Mildenhall, Dor Verbin, Pratul P. Srinivasan, and Pe- ter Hedman. 2022. Mip-NeRF 360: Unbounded Anti-Aliased Neural Radiance Fields. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). 5470–5479

    work page 2022

  4. [2]

    Hermann Blum, Marcus G Müller, Abel Gawel, Roland Siegwart, and Cesar Cadena. 2023. SCIM: Simultaneous Clustering, Inference, and Mapping for Open- World Semantic Scene Understanding. InRobotics Research. Springer, 119–135

    work page 2023

  5. [3]

    Liang-Chieh Chen, Yukun Zhu, George Papandreou, Florian Schroff, and Hartwig Adam. 2018. Encoder-decoder with atrous separable convolution for semantic image segmentation. InProceedings of the European conference on computer vision (ECCV). 801–818

    2018

  6. [4]

    Xiaokang Chen, Kwan-Yee Lin, Jingbo Wang, Wayne Wu, Chen Qian, Hongsheng Li, and Gang Zeng. 2020. Bi-directional cross-modality feature propagation with separation-and-aggregation gate for RGB-D semantic segmentation. InComputer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Proceedings, Part XI. Springer, 561–577

    work page 2020

  7. [5]

    Zheng Chen, Chen Wang, Yuan-Chen Guo, and Song-Hai Zhang. 2022. Struct- NeRF: Neural Radiance Fields for Indoor Scenes with Structural Hints.arXiv preprint arXiv:2209.05277(2022)

    work page Pith review arXiv 2022

  8. [6]

    Camille Couprie, Clément Farabet, Laurent Najman, and Yann LeCun. 2013. Indoor semantic segmentation using depth information. (2013)

    work page 2013

Show all 44 references
  1. [7]

    Jonas Frey, Hermann Blum, Francesco Milano, Roland Siegwart, and Cesar Ca- dena. 2022. Continual Adaptation of Semantic Segmentation Using Complemen- tary 2D-3D Data Representations.IEEE Robotics and Automation Letters7, 4 (2022), 11665–11672. https://doi.org/10.1109/LRA.2022.3203812

    2022

  2. [8]

    Xiao Fu, Shangzhan Zhang, Tianrun Chen, Yichong Lu, Lanyun Zhu, Xiaowei Zhou, Andreas Geiger, and Yiyi Liao. 2022. Panoptic NeRF: 3D-to-2D Label Transfer for Panoptic Urban Scene Segmentation.arXiv preprint arXiv:2203.15224 (2022)

    2022 arXiv

  3. [9]

    Fadri Furrer, Tonci Novkovic, Marius Fehr, Abel Gawel, Margarita Grinvald, Torsten Sattler, Roland Siegwart, and Juan Nieto. 2018. Incremental object data- base: Building 3d models from multiple partial observations. In2018 IEEE/RSJ International Conference on Intelligent Robo...

    2018

  4. [10]

    Haoyu Guo, Sida Peng, Haotong Lin, Qianqian Wang, Guofeng Zhang, Hujun Bao, and Xiaowei Zhou. 2022. Neural 3D Scene Reconstruction with the Manhattan- world Assumption. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 5511–5520

    2022

  5. [11]

    Saurabh Gupta, Pablo Arbeláez, Ross Girshick, and Jitendra Malik. 2015. Indoor scene understanding with rgb-d images: Bottom-up segmentation, object detec- tion and semantic segmentation.International Journal of Computer Vision112, 2 (2015), 133–149

    2015

  6. [12]

    Greg Hamerly and Charles Elkan. 2003. Learning the k in k-means.Advances in neural information processing systems16 (2003)

    2003

  7. [13]

    Diederik P Kingma and Jimmy Ba. 2014. Adam: A method for stochastic opti- mization. (2014)

    2014

  8. [14]

    Zechao Li, Yanpeng Sun, Liyan Zhang, and Jinhui Tang. 2021. CTNet: Context- based tandem network for semantic segmentation.IEEE Transactions on Pattern Analysis and Machine Intelligence(2021)

    2021

  9. [15]

    Huayao Liu, Jiaming Zhang, Kailun Yang, Xinxin Hu, and Rainer Stiefelhagen

  10. [16]

    Yun Liu, Peng-Tao Jiang, Vahan Petrosyan, Shi-Jie Li, Jiawang Bian, Le Zhang 0001, and Ming-Ming Cheng. 2018. Del: Deep embedding learning for efficient image segmentation.. InIJCAI, Vol. 864. 870

    2018

  11. [17]

    Ben Mildenhall, Pratul P Srinivasan, Matthew Tancik, Jonathan T Barron, Ravi Ramamoorthi, and Ren Ng. 2021. Nerf: Representing scenes as neural radiance fields for view synthesis.Commun. ACM65, 1 (2021), 99–106

    2021

  12. [18]

    Thomas Müller, Alex Evans, Christoph Schied, and Alexander Keller. 2022. Instant neural graphics primitives with a multiresolution hash encoding.arXiv preprint arXiv:2201.05989(2022)

    2022 arXiv

  13. [20]

    Yoshikatsu Nakajima, Keisuke Tateno, Federico Tombari, and Hideo Saito. 2018. Fast and accurate semantic mapping through geometric-based incremental seg- mentation. In2018 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). IEEE, 385–392

    2018

  14. [21]

    Manolis Savva, Abhishek Kadian, Oleksandr Maksymets, Yili Zhao, Erik Wijmans, Bhavana Jain, Julian Straub, Jia Liu, Vladlen Koltun, Jitendra Malik, et al. 2019. Habitat: A platform for embodied ai research. InProceedings of the IEEE/CVF International Conference on Computer Vis...

    2019

  15. [22]

    Daniel Seichter, Mona Köhler, Benjamin Lewandowski, Tim Wengefeld, and Horst-Michael Gross. 2021. Efficient rgb-d semantic segmentation for indoor scene analysis. In2021 IEEE International Conference on Robotics and Automation (ICRA). IEEE, 13525–13531

    2021

  16. [23]

    Nathan Silberman, Derek Hoiem, Pushmeet Kohli, and Rob Fergus. 2012. Indoor segmentation and support inference from rgbd images. InEuropean conference on computer vision. Springer, 746–760

    2012

  17. [24]

    Shuran Song, Samuel P Lichtenberg, and Jianxiong Xiao. 2015. Sun rgb-d: A rgb-d scene understanding benchmark suite. InProceedings of the IEEE conference on computer vision and pattern recognition. 567–576

    2015

  18. [25]

    Julian Straub, Thomas Whelan, Lingni Ma, Yufan Chen, Erik Wijmans, Simon Green, Jakob J Engel, Raul Mur-Artal, Carl Ren, Shobhit Verma, et al. 2019. The Replica dataset: A digital replica of indoor spaces.arXiv preprint arXiv:1906.05797 (2019)

    2019 arXiv

  19. [26]

    Zhiqiang Tao, Hongfu Liu, Huazhu Fu, and Yun Fu. 2019. Multi-view saliency- guided clustering for image cosegmentation.IEEE Transactions on Image Process- ing28, 9 (2019), 4634–4645

    2019

  20. [28]

    Haithem Turki, Deva Ramanan, and Mahadev Satyanarayanan. 2022. Mega- NERF: Scalable Construction of Large-Scale NeRFs for Virtual Fly-Throughs. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). 12922–12931

    2022

  21. [29]

    Han Vanholder. 2016. Efficient inference with tensorrt. InGPU Technology Conference, Vol. 1. 2

    2016

  22. [30]

    Suhani Vora, Noha Radwan, Klaus Greff, Henning Meyer, Kyle Genova, Mehdi SM Sajjadi, Etienne Pot, Andrea Tagliasacchi, and Daniel Duckworth. 2021. Nesf: Neural semantic fields for generalizable semantic segmentation of 3d scenes. arXiv preprint arXiv:2111.13260(2021)

    2021 arXiv

  23. [31]

    Huan Wang, Jian Ren, Zeng Huang, Kyle Olszewski, Menglei Chai, Yun Fu, and Sergey Tulyakov. 2022. R2L: Distilling Neural Radiance Field to Neural Light Field for Efficient Novel View Synthesis.arXiv preprint arXiv:2203.17261(2022)

    2022 arXiv

  24. [32]

    Junming Wang, Zekai Sun, Xiuxian Guan, Tianxiang Shen, Zongyuan Zhang, Tianyang Duan, Dong Huang, Shixiong Zhao, and Heming Cui. 2024. AGRNav: Efficient and Energy-Saving Autonomous Navigation for Air-Ground Robots in Occlusion-Prone Environments.arXiv preprint arXiv:2403.11607(2024)

    2024 arXiv

  25. [33]

    Xiaoyang Wang, Jimin Xiao, Bingfeng Zhang, and Limin Yu. 2022. CARD: Semi- supervised semantic segmentation via class-agnostic relation based denoising. In Proc. IJCAI. 1451–1457

    2022

  26. [34]

    Yikai Wang, Xinghao Chen, Lele Cao, Wenbing Huang, Fuchun Sun, and Yunhe Wang. 2022. Multimodal token fusion for vision transformers. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 12186–12195

    2022

  27. [35]

    Zijin Wu, Xingyi Li, Juewen Peng, Hao Lu, Zhiguo Cao, and Weicai Zhong. 2022. DoF-NeRF: Depth-of-Field Meets Neural Radiance Fields. InProceedings of the 30th ACM International Conference on Multimedia. 1718–1729

    2022

  28. [36]

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

    2022

  29. [37]

    Rui Xu and Donald Wunsch. 2005. Survey of clustering algorithms.IEEE Trans- actions on neural networks16, 3 (2005), 645–678

    2005

  30. [38]

    Hong-Ming Yang, Xu-Yao Zhang, Fei Yin, Qing Yang, and Cheng-Lin Liu. 2020. Convolutional prototype network for open set recognition.IEEE Transactions on Pattern Analysis and Machine Intelligence44, 5 (2020), 2358–2370

    2020

  31. [39]

    Minxiang Ye, Yifei Zhang, Shiqiang Zhu, Anhuan Xie, and Dan Zhang. 2022. Deep Markov Clustering for Panoptic Segmentation. InICASSP 2022-2022 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, 2380–2384

    2022

  32. [40]

    Yuyang Zhao, Zhun Zhong, Nicu Sebe, and Gim Hee Lee. 2022. Novel Class Discovery in Semantic Segmentation. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 4340–4349

    2022

  33. [41]

    Shuaifeng Zhi, Tristan Laidlow, Stefan Leutenegger, and Andrew J Davison. 2021. In-place scene labelling and understanding with implicit scene representation. In Proceedings of the IEEE/CVF International Conference on Computer Vision. 15838– 15847

    2021

  34. [42]

    Shuaifeng Zhi, Edgar Sucar, Andre Mouton, Iain Haughton, Tristan Laidlow, and Andrew J Davison. 2022. iLabel: Revealing Objects in Neural Fields.IEEE Robotics and Automation Letters(2022)

    2022

  35. [43]

    Qian-Yi Zhou, Jaesik Park, and Vladlen Koltun. 2018. Open3D: A modern library for 3D data processing.arXiv preprint arXiv:1801.09847(2018)

    2018 arXiv

  36. [2022]

    CMX: Cross-modal fusion for RGB-X semantic segmentation with trans- formers.arXiv preprint arXiv:2203.04838(2022)

    2022 arXiv

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.