Pith. sign in

REVIEW 3 major objections 5 minor 130 references

Squeeze3D: Your 3D Generation Model is Secretly an Extreme Neural Compressor

T0 review · 3 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

Pith's one-line read The paper claims that a frozen pre-trained 3D generator, bridged to a pre-trained encoder by two small mapping networks trained only on synthetic outputs, acts as an extreme lossy codec, squeezing meshes, point clouds, and radiance fields…

desk verdict Squeeze3D has a genuinely new mechanism and plausible mesh results, but the evaluation oversells the extreme-compression story and dodges the hard off-manifold test. read the letter →

arxiv 2506.07932 v1 pith:7H4OFCOF submitted 2025-06-09 cs.GR cs.CVcs.LG

classification cs.GRcs.CVcs.LG
keywords 3Dcompressiongenerativemodelslatentspacemappingneuralmeshpointcloudradiancefieldslossy
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 paper aims to establish that a pre-trained 3D generative model, kept completely frozen, can serve as the decompression engine of an extreme lossy codec. Two small mapping networks are trained to translate the latent space of any existing 3D encoder into the latent space of the generator, so that a compressed code of a few kilobytes can be expanded back into a recognizable mesh, point cloud, or radiance field. Training needs no real 3D dataset: the pairs used for supervision come from running the generator itself and encoding its outputs. If the claim holds, a 6.43 MB textured mesh becomes a 3 KB code, a 117 KB point cloud becomes a 2 KB code, and a 58 MB radiance field becomes about a 90 KB code, at compression ratios one to two orders of magnitude beyond prior codecs. The paper also argues that this design improves automatically as 3D generators become more expressive.

What carries the argument

The load-bearing mechanism is the pair of trainable mappings $F^E_\theta$ and $F^D_\theta$ between two frozen latent spaces, one belonging to the encoder $E$ and one to the generator $\mathcal G$. A Gram-loss term, $\lambda_{\text{gram}}\|F^E_\theta(z_E)F^E_\theta(z_E)^\top - I\|_F^2$, drives the compressed codes toward orthonormality, which the authors show is needed because unregularized codes concentrate information along a few dominant singular directions and waste most of their dimensions. The compressed latent dimension then acts as the rate knob of the codec: the ablation on point clouds shows quality rises with code size up to a point, then saturates.

What would settle it

Encode a curated set of objects from categories the generator has never seen (for example, mechanical CAD parts or indoor room scans), store only the compressed codes, and reconstruct; if the reconstructions are no closer to the inputs than random samples from the generator by standard metrics such as LPIPS or Chamfer distance, the claim that Squeeze3D acts as a general 3D compressor fails for out-of-prior inputs.

Watch

Extended reading notes

Core claim

Squeeze3D's central claim is that disparate latent spaces can be bridged by learned maps, making an existing encoder-generator pair into a codec without retraining either model. For an object $G$, the encoder yields $z_E=E(G)$; the forward mapping $F^E_\theta$ compresses it to $z_{\text{comp}}$; the reverse mapping $F^D_\theta$ lifts that code to a generator latent $z_G$; and the frozen generator $\mathcal{G}(z_G,c)$ redraws the object. The authors report compression ratios up to 2187x for textured meshes, about 58x for point clouds, and about 619x for radiance fields, with reconstruction quality close to that of methods that compress far less aggressively. Because the mapping networks are trained exclusively on synthetic generator outputs, the pipeline is format-agnostic and can be re-targeted to a new encoder or generator by retraining only the two small networks.

Load-bearing premise

The load-bearing premise is that any object fed to the codec lies close enough to the generator's output distribution that the encoder's latent can be inverted into a generator latent whose output resembles that object; if the object is outside the generator's prior, the stored code contains no genuine object identity and decompression becomes a hallucination.

Editorial extensions

If this is right

  • A textured mesh can be stored and transmitted as about 3 KB, a point cloud as about 2 KB, and a radiance field as about 90 KB, without per-object training.
  • Compression and decompression are single forward passes through small networks plus the frozen generator, so the method avoids the long per-scene optimizations of many neural compression methods; the paper reports roughly 270 ms to compress and 1.5 s to decompress a mesh.
  • The same recipe applies to any encoder-generator pair, so supporting a new 3D format or a stronger generator only requires retraining the two mapping networks.
  • Reconstruction quality is capped by the generator's representational power, which means Squeeze3D inherits future generator improvements for free.
  • Compressed codes form a continuous, interpolable space, so linearly interpolating two codes yields an intermediate mesh.

Reading between the lines

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

  • The same latent-bridging recipe could be lifted to 2D or video: a frozen image or video diffusion model would act as the decoder, mapping networks would be trained only on synthetic outputs, and the achievable ratios would be determined by how well the generator covers the test distribution.
  • The rate-distortion frontier of this approach is set by the generator's prior, not by the codec, so a meaningful benchmark would measure reconstruction fidelity as a function of how far test objects fall from that prior; the paper's appendix already shows a 4.22 dB PSNR drop on out-of-distribution radiance fields.
  • The Gram-loss finding suggests a general design rule for neural latent codecs: without an orthonormality regularizer, the effective rank of the compressed code is far below its nominal dimension, so any learned bottleneck should include such a term.
  • A hybrid system could detect when the input is out of the generator's support and fall back to a classical codec; the paper flags the need for such a fallback, and a concrete test would be a decision rule based on reconstruction likelihood or latent density.
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

3 major / 5 minor

Summary. The paper proposes Squeeze3D, a compression framework that uses a pre-trained 3D encoder to map a mesh, point cloud, or radiance field to a compact latent, then two small mapping networks (forward and reverse) to translate that latent into the latent space of a pre-trained 3D generator, which reconstructs the object. The mapping networks are trained exclusively on synthetic generator outputs using a reconstruction loss on generator latents plus a Gram orthogonality loss. The paper reports extreme compression ratios (up to 2187x for textured meshes, 58.5x for point clouds, and 619x for radiance fields) and claims visual quality comparable to many existing methods across the three formats.

Significance. If the method works as claimed, the contribution is significant: it demonstrates that a frozen generative prior can serve as an extreme lossy codec, eliminating per-object optimization and supporting flexible encoder/generator swaps. The paper is clearly written, the training procedure is reproducible from the description, and the ablation on compressed size is useful. However, the evaluation currently supports the extreme-ratio claim better than the quality-comparability claim, and the load-bearing distribution-coverage assumption is never tested with genuinely off-manifold data.

major comments (3)
  1. [§4.3, Table 3] The claim that Squeeze3D maintains visual quality 'comparable to many existing methods' is not supported for point clouds. Squeeze3D achieves PointSSIM 0.4484 versus 0.9722 for Draco‡ and 0.9535 for Draco§, and PCQM 1.8437 versus 3.2875 for Draco‡. A PointSSIM drop of more than half is not 'comparable' in the usual sense, even at a higher compression ratio. The paper should either revise the wording, add a rate-distortion analysis that makes the trade-off explicit, or include a quality-matched comparison at similar bitrates.
  2. [§3.2, §5, §B.2, §C.1] The load-bearing distribution-coverage assumption is never tested. The training set consists only of pairs (E(G(z,c)), z) sampled from the frozen generator's own outputs, so the mapping networks only learn to invert the encoder on the generator's output manifold. The out-of-distribution evaluation in §B.2 is composed of outputs of other 3D generative models (OpenLRM, InstantMesh, Hunyuan3D-2, TRELLIS, SPAR3D, SORA-3D), and the radiance-field OOD set in §C.2 is the 'ai' subset of NeRF-MAE, the same model family used as encoder and generator. The main test sets (Objaverse, ShapeNet, NeRF-MAE) lie within the generator's training prior. No real scanned, LiDAR, or CAD object is compressed. For an off-manifold object, zcomp may contain no identity information and decompression becomes hallucination. The limitation section concedes this qualitatively, but the paper never quantifies the degradation as a function of distance from the generator manifold, which is needed to support the claim that Squeeze3D is a general 3D codec.
  3. [§4.1, Tables 4 and 8] The radiance-field experiment uses NeRF-MAE as both the encoder and the generator, and the OOD test is the 'ai' subset of the same dataset. This conflates the generative prior with the encoder and does not exercise the claimed cross-model bridging. The compressed size for this leg is 24,000 dimensions (Table 5), far larger than the mesh and point-cloud codes, so the headline 619x ratio is achieved on a very large original representation with a relatively large code. The paper should clarify what the radiance-field experiment demonstrates beyond a single autoencoder with a trained bottleneck, and, if possible, include a second generator for this format.
minor comments (5)
  1. [§4.3] The text states the point-cloud compression ratio is '117 (117 / 1.00)', while Table 3 reports CR 58.5 (117 / 2.00) and the abstract says 55x; these numbers should be reconciled.
  2. [Figures 4 and 5] The captions contain the placeholder text 'Lorem ipsum', which should be replaced with actual descriptions.
  3. [Equation (4)] The Gram loss is written for a single sample; since the Frobenius norm of a vector outer product is simply the vector norm squared, the intended batch-wise orthogonality should be stated explicitly.
  4. [§C.2] There are typos ('seprately', 'datsets', 'and ae') that should be corrected.
  5. [Table 2] The reported compression ratio 2187.0748 does not match the displayed sizes 6.43 MB / 0.003 MB ≈ 2143; please verify the arithmetic.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: Squeeze3D's training target is the generator's own latent, but the paper's claims are empirical and the dependence on generator coverage is openly stated as a limitation, not hidden in the derivation.

full rationale

The mapping networks in Squeeze3D are trained, per Eq. (4), to map encoder latents of synthetic generator outputs back to the generator latents that produced them, with the Gram loss encouraging orthonormality of the compressed code. This design honestly recognizes the generator as both the prior and the quality ceiling, as explicitly conceded in Section 5: 'the decompressed outputs from Squeeze3D fundamentally cannot exceed the quality of what the generator can produce.' The main compression results, however, are measured on held-out datasets (Objaverse, ShapeNet, and NeRF-MAE) using frozen pre-trained encoders and generators; they are reported empirical measurements rather than predictions derived from the training loss. The 'out-of-distribution' mesh collection in Section B.2 draws from other 3D generative model demos, which limits external validity for real off-manifold data but does not make the method's own equations circular. No load-bearing self-citations or author-imported uniqueness theorems appear, and the distribution-coverage assumption is stated as a limitation rather than disguised as a result. Consequently, the central pipeline does not reduce to its inputs by construction; the main risk is correctness/generalization, not circularity.

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

The paper introduces no new physical, mathematical, or ontological entities. The compressed latent zcomp is the actual stored bitstream, and the mapping networks are trained artifacts whose behavior is benchmarked externally on Objaverse, ShapeNet, and NeRF-MAE test objects. The absence of invented entities does not offset the free parameters and domain assumptions above, which carry the central claim.

free parameters (4)
  • Compressed latent dimension dC = 770 (InstantMesh), 1024 (OpenLRM, Shap-E, LION), 1024 to 8192 (LION ablations), 24000 (radiance fields)
    Hand-chosen per encoder-generator pair; it directly sets the stored artifact size and therefore the compression ratio. No rate-distortion optimization over dC is performed (Table 5, §C.4).
  • Loss weights λgram and λgen = Not reported
    Defined in Eq. (4) but never instantiated in the paper, which blocks reproduction of the training objective.
  • LLaMA3 prompt distribution for mesh training data = 2500 prompts repeated 4 times = 10000 objects
    The synthetic mesh corpus, and thus the set of objects the compressor can represent, is determined by an unreleased prompt list (§4.1).
  • Generator-target representation per pair = Triplanes (InstantMesh, OpenLRM); implicit MLP (Shap-E); global and local latents (LION)
    Selected empirically, e.g. 'we experimentally observed better performance with generating triplane representations, thus our results report this setting' (§A.2); the choice fixes which part of the generator's latent space is being targeted.
assumptions (4)
  • domain assumption The pre-trained encoder E maps each 3D object to a latent vector that preserves the object's identity.
    Invoked in Eqs. (1)-(2); if the encoder discards texture, topology, or fine geometry, decompression cannot recover them. MeshAnything, used as the mesh encoder, is trained for mesh generation, not as a lossless descriptor, and the paper does not measure how much object information E retains.
  • domain assumption Objects seen at test time lie near the output distribution of the frozen generator G.
    Training pairs are sampled from G's own outputs (§3.2), so the mapping learns to invert E only on G's output manifold. For objects outside that manifold, acknowledged in §5 as outliers, the stored code contains no object identity and decompression is hallucination. This assumption, not the method's equations, carries the rate-distortion claims.
  • ad hoc to paper NeRF-MAE's decoder can serve as the generative model for the radiance-field experiment.
    §4.1 states 'For radiance fields, we train mapping networks for NeRF-MAE as the encoder and the generator', with training fields produced from the NeRF-MAE dataset. NeRF-MAE is a masked autoencoder, not a generative model; this leg reduces to autoencoding through the MAE's own bottleneck.
  • domain assumption The Gram semi-orthogonality constraint does not discard information needed for reconstruction.
    Eq. (4) adds the term λgram times the Frobenius norm of F(zE)F(zE)^T minus the identity, to decorrelate compressed dimensions. The empirical singular-value motivation in §3.2 does not establish that the constraint preserves rate-distortion optimality, and the weights λgram and λgen are never reported.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Squeeze3D: Your 3D Generation Model is Secretly an Extreme Neural Compressor." pith.science (2026). https://pith.science/paper/7H4OFCOF

@misc{pith2026250607932,
  author       = {Pith},
  title        = {Pith review of: Squeeze3D: Your 3D Generation Model is Secretly an Extreme Neural Compressor},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/7H4OFCOF}},
  note         = {Machine review of arXiv:2506.07932}
}
read the original abstract

We propose Squeeze3D, a novel framework that leverages implicit prior knowledge learnt by existing pre-trained 3D generative models to compress 3D data at extremely high compression ratios. Our approach bridges the latent spaces between a pre-trained encoder and a pre-trained generation model through trainable mapping networks. Any 3D model represented as a mesh, point cloud, or a radiance field is first encoded by the pre-trained encoder and then transformed (i.e. compressed) into a highly compact latent code. This latent code can effectively be used as an extremely compressed representation of the mesh or point cloud. A mapping network transforms the compressed latent code into the latent space of a powerful generative model, which is then conditioned to recreate the original 3D model (i.e. decompression). Squeeze3D is trained entirely on generated synthetic data and does not require any 3D datasets. The Squeeze3D architecture can be flexibly used with existing pre-trained 3D encoders and existing generative models. It can flexibly support different formats, including meshes, point clouds, and radiance fields. Our experiments demonstrate that Squeeze3D achieves compression ratios of up to 2187x for textured meshes, 55x for point clouds, and 619x for radiance fields while maintaining visual quality comparable to many existing methods. Squeeze3D only incurs a small compression and decompression latency since it does not involve training object-specific networks to compress an object.

Figures

Figures reproduced from arXiv: 2506.07932 by the authors.

Figure 1
Figure 1. We showcase extreme compression of 3D models while preserving perceptual quality. [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Overview of our Method. Squeeze3D bridges arbitrary latent spaces between 3D encoders and generators through trainable mapping networks. During compression, a 3D geometry is encoded and then transformed into a compact representation via the forward mapping network. During decompression, the reverse mapping network converts this representation into the generator’s latent space, which is then used to reconstruct the o… view at source ↗
Figure 3
Figure 3. Training Squeeze3D. We show an overview of (a) our process of creating synthetic data to train the mapping networks and (b) our process of training the mapping networks. Given a pre-trained 3D generator G and encoder E, we first sample a diverse collection of condi￾tioning inputs C = {ci} N i=1 appropriate for the generator model (e.g. text prompts for text-to-3D generators, images for image-to-3D generators, or ran… view at source ↗
Figures from the paper (21 more)
Figure 4
Figure 4. Figure 4: Qualitative mesh compression results. We compare Squeeze3D to state-of-the-art methods. Our approach maintains visually important geometric details. Additional results in§ C. To address this, we propose the Gram loss term that is computed on the outputs of the first ma…
Figure 5
Figure 5. Figure 5: Qualitative point cloud compression results. We show qualitative results comparing Squeeze3D to state-of-the-art methods. Our ap￾proach achieves significantly higher compression ratios while maintaining perceptually important geometric details. Training Dataset Creatio…
Figure 6
Figure 6. Figure 6: Qualitative radiance field compression results. We show qualitative results comparing Squeeze3D to state-of-the-art methods. Our approach achieves a significantly higher compression ratio while maintaining visually important geometric details. While many other compress…
Figure 7
Figure 7. Figure 7: Network Architectures. For each of the encoder-decoder pairs, we train the mapping networks, which are feed-forward neural networks. We summarize network archi￾tectures in [PITH_FULL_IMAGE:figures/full_fig_p018_7.png]
Figure 8
Figure 8. Figure 8: Interpolation. The compressed representations we obtain can also be interpolated. In these examples we obtain the compressed representation for the leftmost and rightmost meshes and linearly interpolate between them. Ground Truth Reconstruction [PITH_FULL_IMAGE:figure…
Figure 9
Figure 9. Figure 9: Multi-view visualization of compressed and reconstructed meshes. The consistent ap￾pearance across different viewing angles demonstrates that Squeeze3D learns correct transformations between latent spaces and produces coherent 3D reconstructions. This confirms that our…
Figure 10
Figure 10. Figure 10: Squeeze3D preserves geometry details. We show some meshes compressed with Squeeze3D as wireframes. Notice that Squeeze3D preserves many finegrained geometric details [PITH_FULL_IMAGE:figures/full_fig_p023_10.png]
Figure 11
Figure 11. Figure 11: Compression results using different 3D genera￾tors. Squeeze3D is agnostic to the choice of a 3D generation model. Thus, we show compression results with the 3D gener￾ators: OpenLRM, and Shap-E. We choose 3D meshes that lie in the representation capacity of the chosen …
Figure 13
Figure 13. Figure 13: Results Library. 25 [PITH_FULL_IMAGE:figures/full_fig_p025_13.png]
Figure 14
Figure 14. Figure 14: Results Library. 26 [PITH_FULL_IMAGE:figures/full_fig_p026_14.png]
Figure 15
Figure 15. Figure 15: Results Library. 27 [PITH_FULL_IMAGE:figures/full_fig_p027_15.png]
Figure 16
Figure 16. Figure 16: Results Library. 28 [PITH_FULL_IMAGE:figures/full_fig_p028_16.png]
Figure 17
Figure 17. Figure 17: Results Library. 29 [PITH_FULL_IMAGE:figures/full_fig_p029_17.png]
Figure 18
Figure 18. Figure 18: Results Library. 30 [PITH_FULL_IMAGE:figures/full_fig_p030_18.png]
Figure 19
Figure 19. Figure 19: Results Library. 31 [PITH_FULL_IMAGE:figures/full_fig_p031_19.png]
Figure 20
Figure 20. Figure 20: Results Library. 32 [PITH_FULL_IMAGE:figures/full_fig_p032_20.png]
Figure 21
Figure 21. Figure 21: Results Library. 33 [PITH_FULL_IMAGE:figures/full_fig_p033_21.png]
Figure 22
Figure 22. Figure 22: Results Library. 34 [PITH_FULL_IMAGE:figures/full_fig_p034_22.png]
Figure 23
Figure 23. Figure 23: Results Library. 35 [PITH_FULL_IMAGE:figures/full_fig_p035_23.png]
Figure 24
Figure 24. Figure 24: Results Library. 36 [PITH_FULL_IMAGE:figures/full_fig_p036_24.png]
Figure 25
Figure 25. Figure 25: Results Library. 37 [PITH_FULL_IMAGE:figures/full_fig_p037_25.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

130 extracted references · 65 canonical work pages

  1. [1]

    Towards a point cloud structural similarity metric

    Evangelos Alexiou and Touradj Ebrahimi. Towards a point cloud structural similarity metric. In2020 IEEE International Conference on Multimedia & Expo Workshops (ICMEW), pages 1–6, 2020

  2. [2]

    Diff-PCC: Diffusion-based neural compression for 3d point clouds, 2024

    Anonymous. Diff-PCC: Diffusion-based neural compression for 3d point clouds, 2024

  3. [3]

    Sal: Sign agnostic learning of shapes from raw data

    Matan Atzmon and Yaron Lipman. Sal: Sign agnostic learning of shapes from raw data. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 2565–2574, 2020

  4. [4]

    Jimmy Lei Ba, Jamie Ryan Kiros, and Geoffrey E. Hinton. Layer normalization, 2016

  5. [5]

    Andrew Brock, Theodore Lim, J. M. Ritchie, and Nick Weston. Generative and discriminative voxel modeling with convolutional neural networks, 2016

  6. [6]

    Recursively generated b-spline surfaces on arbitrary topological meshes

    Edwin Catmull and James Clark. Recursively generated b-spline surfaces on arbitrary topological meshes. InSeminal graphics: pioneering efforts that shaped the field, pages 183–188. 1998

  7. [7]

    Chang, Thomas Funkhouser, Leonidas Guibas, Pat Hanrahan, Qixing Huang, Zimo Li, Silvio Savarese, Manolis Savva, Shuran Song, Hao Su, Jianxiong Xiao, Li Yi, and Fisher Yu

    Angel X. Chang, Thomas Funkhouser, Leonidas Guibas, Pat Hanrahan, Qixing Huang, Zimo Li, Silvio Savarese, Manolis Savva, Shuran Song, Hao Su, Jianxiong Xiao, Li Yi, and Fisher Yu. Shapenet: An information-rich 3d model repository, 2015

  8. [8]

    3d shape tokenization via latent flow matching, 2025

    Jen-Hao Rick Chang, Yuyang Wang, Miguel Angel Bautista Martin, Jiatao Gu, Xiaoming Zhao, Josh Susskind, and Oncel Tuzel. 3d shape tokenization via latent flow matching, 2025

Show all 130 references
  1. [9]

    Tensorf: Tensorial radiance fields

    Anpei Chen, Zexiang Xu, Andreas Geiger, Jingyi Yu, and Hao Su. Tensorf: Tensorial radiance fields. In European conference on computer vision, pages 333–350. Springer, 2022

  2. [10]

    Deep compression autoencoder for efficient high-resolution diffusion models, 2025

    Junyu Chen, Han Cai, Junsong Chen, Enze Xie, Shang Yang, Haotian Tang, Muyang Li, Yao Lu, and Song Han. Deep compression autoencoder for efficient high-resolution diffusion models, 2025

  3. [11]

    How far can we compress instant-ngp-based nerf? InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 20321–20330, June 2024

    Yihang Chen, Qianyi Wu, Mehrtash Harandi, and Jianfei Cai. How far can we compress instant-ngp-based nerf? InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 20321–20330, June 2024

  4. [12]

    Meshanything: Artist-created mesh generation with autoregressive transformers, 2024

    Yiwen Chen, Tong He, Di Huang, Weicai Ye, Sijin Chen, Jiaxiang Tang, Xin Chen, Zhongang Cai, Lei Yang, Gang Yu, Guosheng Lin, and Chi Zhang. Meshanything: Artist-created mesh generation with autoregressive transformers, 2024

  5. [13]

    3dtopia-xl: Scaling high-quality 3d asset generation via primitive diffusion.arXiv preprint arXiv:2409.12957, 2024

    Zhaoxi Chen, Jiaxiang Tang, Yuhao Dong, Ziang Cao, Fangzhou Hong, Yushi Lan, Tengfei Wang, Haozhe Xie, Tong Wu, Shunsuke Saito, et al. 3dtopia-xl: Scaling high-quality 3d asset generation via primitive diffusion.arXiv preprint arXiv:2409.12957, 2024

  6. [14]

    Learning implicit fields for generative shape modeling

    Zhiqin Chen and Hao Zhang. Learning implicit fields for generative shape modeling. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 5939–5948, 2019

  7. [15]

    Appearance-preserving simplification

    Jonathan Cohen, Marc Olano, and Dinesh Manocha. Appearance-preserving simplification. InProceed- ings of the 25th annual conference on Computer graphics and interactive techniques, pages 115–122, 1998

  8. [16]

    Octformer: Efficient octree-based transformer for point cloud compression with local enhancement.Proceedings of the AAAI Conference on Artificial Intelligence, 37(1):470–478, Jun

    Mingyue Cui, Junhua Long, Mingjian Feng, Boyang Li, and Huang Kai. Octformer: Efficient octree-based transformer for point cloud compression with local enhancement.Proceedings of the AAAI Conference on Artificial Intelligence, 37(1):470–478, Jun. 2023

  9. [17]

    Neusdfusion: A spatial-aware generative model for 3d shape completion, reconstruction, and generation, 2024

    Ruikai Cui, Weizhe Liu, Weixuan Sun, Senbo Wang, Taizhang Shang, Yang Li, Xibin Song, Han Yan, Zhennan Wu, Shenzhou Chen, Hongdong Li, and Pan Ji. Neusdfusion: A spatial-aware generative model for 3d shape completion, reconstruction, and generation, 2024

  10. [18]

    On the effectiveness of weight-encoded neural implicit 3d shapes, 2021

    Thomas Davies, Derek Nowrouzezahrai, and Alec Jacobson. On the effectiveness of weight-encoded neural implicit 3d shapes, 2021

  11. [19]

    Geometry compression

    Michael Deering. Geometry compression. InProceedings of the 22nd Annual Conference on Com- puter Graphics and Interactive Techniques, SIGGRAPH ’95, page 13–20, New York, NY , USA, 1995. Association for Computing Machinery

  12. [20]

    Objaverse-xl: A universe of 10m+ 3d objects, 2023

    Matt Deitke, Ruoshi Liu, Matthew Wallingford, Huong Ngo, Oscar Michel, Aditya Kusupati, Alan Fan, Christian Laforte, Vikram V oleti, Samir Yitzhak Gadre, Eli VanderBilt, Aniruddha Kembhavi, Carl V ondrick, Georgia Gkioxari, Kiana Ehsani, Ludwig Schmidt, and Ali Farhadi. Objave...

  13. [21]

    An image is worth 16x16 words: Transformers for image recognition at scale

    Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Sylvain Gelly, Jakob Uszkoreit, and Neil Houlsby. An image is worth 16x16 words: Transformers for image recognition at...

  14. [22]

    Neural geometry fields for meshes

    Venkataram Edavamadathil Sivaram, Tzu-Mao Li, and Ravi Ramamoorthi. Neural geometry fields for meshes. InACM SIGGRAPH 2024 Conference Papers, SIGGRAPH ’24, New York, NY , USA, 2024. Association for Computing Machinery

  15. [23]

    Lightgaussian: Unbounded 3d gaussian compression with 15x reduction and 200+ fps, 2024

    Zhiwen Fan, Kevin Wang, Kairun Wen, Zehao Zhu, Dejia Xu, and Zhangyang Wang. Lightgaussian: Unbounded 3d gaussian compression with 15x reduction and 200+ fps, 2024

  16. [24]

    Google/draco: a library for compressing and decompressing 3d geometric meshes and point clouds, 2018

    Frank Galligan, Michael Hemmer, Ondrej Stava, Fan Zhang, and Jamieson Brettle. Google/draco: a library for compressing and decompressing 3d geometric meshes and point clouds, 2018

  17. [25]

    Galvis, Xingxing Zuo, Simon Schaefer, and Stefan Leutengger

    Juan D. Galvis, Xingxing Zuo, Simon Schaefer, and Stefan Leutengger. Sc-diff: 3d shape completion with latent diffusion models, 2024

  18. [26]

    Strivec: Sparse tri-vector radiance fields, 2023

    Quankai Gao, Qiangeng Xu, Hao Su, Ulrich Neumann, and Zexiang Xu. Strivec: Sparse tri-vector radiance fields, 2023

  19. [27]

    Rui Gao and Rajeev K. Jaiman. H-siren: Improving implicit neural representations with hyperbolic periodic functions, 2024

  20. [28]

    Heckbert

    Michael Garland and Paul S. Heckbert. Surface simplification using quadric error metrics. InProceedings of the 24th Annual Conference on Computer Graphics and Interactive Techniques, SIGGRAPH ’97, page 209–216, USA, 1997. ACM Press/Addison-Wesley Publishing Co

  21. [29]

    On quantizing implicit neural representations

    Cameron Gordon, Shin-Fang Chng, Lachlan MacDonald, and Simon Lucey. On quantizing implicit neural representations. InProceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, pages 341–350, 2023

  22. [30]

    Mesh convolutional autoencoder for semi-regular meshes of different sizes

    Sara Hahner and Jochen Garcke. Mesh convolutional autoencoder for semi-regular meshes of different sizes. InProceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV), pages 885–894, January 2022

  23. [31]

    Identity mappings in deep residual networks

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Identity mappings in deep residual networks. In Bastian Leibe, Jiri Matas, Nicu Sebe, and Max Welling, editors,Computer Vision – ECCV 2016, pages 630–645, Cham, 2016. Springer International Publishing

  24. [32]

    Openlrm: Open-source large reconstruction models

    Zexin He and Tengfei Wang. Openlrm: Open-source large reconstruction models. https://github. com/3DTopia/OpenLRM, 2023

  25. [33]

    Gaussian error linear units (gelus), 2023

    Dan Hendrycks and Kevin Gimpel. Gaussian error linear units (gelus), 2023

  26. [34]

    Deep geometric texture synthesis.arXiv preprint arXiv:2007.00074, 2020

    Amir Hertz, Rana Hanocka, Raja Giryes, and Daniel Cohen-Or. Deep geometric texture synthesis.arXiv preprint arXiv:2007.00074, 2020

  27. [35]

    Masked autoencoder for self-supervised pre-training on lidar point clouds

    Georg Hess, Johan Jaxing, Elias Svensson, David Hagerman, Christoffer Petersson, and Lennart Svensson. Masked autoencoder for self-supervised pre-training on lidar point clouds. In2023 IEEE/CVF Winter Conference on Applications of Computer Vision Workshops (WACVW), page 350–35...

  28. [36]

    Lrm: Large reconstruction model for single image to 3d, 2024

    Yicong Hong, Kai Zhang, Jiuxiang Gu, Sai Bi, Yang Zhou, Difan Liu, Feng Liu, Kalyan Sunkavalli, Trung Bui, and Hao Tan. Lrm: Large reconstruction model for single image to 3d, 2024

  29. [37]

    Piecewise smooth surface reconstruction

    Hugues Hoppe, Tony DeRose, Tom Duchamp, Mark Halstead, Hubert Jin, John McDonald, Jean Schweitzer, and Werner Stuetzle. Piecewise smooth surface reconstruction. InProceedings of the 21st annual conference on Computer graphics and interactive techniques, pages 295–302, 1994

  30. [38]

    Rendering-oriented 3d point cloud attribute compression using sparse tensor-based transformer, 2025

    Xiao Huo, Junhui Hou, Shuai Wan, and Fuzheng Yang. Rendering-oriented 3d point cloud attribute compression using sparse tensor-based transformer, 2025

  31. [39]

    Nerf-mae: Masked autoencoders for self-supervised 3d representation learning for neural radiance fields

    Muhammad Zubair Irshad, Sergey Zakharov, Vitor Guizilini, Adrien Gaidon, Zsolt Kira, and Rares Ambrus. Nerf-mae: Masked autoencoders for self-supervised 3d representation learning for neural radiance fields. InEuropean Conference on Computer Vision, pages 434–453. Springer, 2024

  32. [40]

    Neural 3d scene compression via model compression.arXiv preprint arXiv:2105.03120, 2021

    Berivan Isik. Neural 3d scene compression via model compression.arXiv preprint arXiv:2105.03120, 2021. 12

  33. [41]

    Efficient learning with sine-activated low-rank matrices, 2025

    Yiping Ji, Hemanth Saratchandran, Cameron Gordon, Zeyu Zhang, and Simon Lucey. Efficient learning with sine-activated low-rank matrices, 2025

  34. [42]

    Generative latent coding for ultra-low bitrate image compression

    Zhaoyang Jia, Jiahao Li, Bin Li, Houqiang Li, and Yan Lu. Generative latent coding for ultra-low bitrate image compression. In2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 26088–26098, 2024

  35. [43]

    Local implicit grid representations for 3d scenes

    Chiyu Jiang, Avneesh Sud, Ameesh Makadia, Jingwei Huang, Matthias Nießner, Thomas Funkhouser, et al. Local implicit grid representations for 3d scenes. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 6001–6010, 2020

  36. [44]

    Shap-e: Generating conditional 3d implicit functions, 2023

    Heewoo Jun and Alex Nichol. Shap-e: Generating conditional 3d implicit functions, 2023

  37. [45]

    Prunerf: Segment- centric dataset pruning via 3d spatial consistency.arXiv preprint arXiv:2406.00798, 2024

    Yeonsung Jung, Heecheol Yun, Joonhyung Park, Jin-Hwa Kim, and Eunho Yang. Prunerf: Segment- centric dataset pruning via 3d spatial consistency.arXiv preprint arXiv:2406.00798, 2024

  38. [46]

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

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

  39. [47]

    Auto-encoding variational bayes, 2013

    Diederik P Kingma, Max Welling, et al. Auto-encoding variational bayes, 2013

  40. [48]

    Nerf-vae: A geometry aware 3d scene generative model

    Adam R Kosiorek, Heiko Strathmann, Daniel Zoran, Pol Moreno, Rosalia Schneider, Sona Mokrá, and Danilo Jimenez Rezende. Nerf-vae: A geometry aware 3d scene generative model. InInternational Conference on Machine Learning, pages 5742–5752. PMLR, 2021

  41. [49]

    Corto: Mesh compression library

    Visual Computing Lab. Corto: Mesh compression library. https://github.com/cnr-isti-vclab/ corto, 2025. Accessed: March 03, 2025

  42. [50]

    Ln3diff: Scalable latent neural fields diffusion for speedy 3d generation

    Yushi Lan, Fangzhou Hong, Shuai Yang, Shangchen Zhou, Xuyi Meng, Bo Dai, Xingang Pan, and Chen Change Loy. Ln3diff: Scalable latent neural fields diffusion for speedy 3d generation. InEuropean Conference on Computer Vision, pages 112–130. Springer, 2024

  43. [51]

    Compact 3d gaussian representation for radiance field

    Joo Chan Lee, Daniel Rho, Xiangyu Sun, Jong Hwan Ko, and Eunbyung Park. Compact 3d gaussian representation for radiance field. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 21719–21728, June 2024

  44. [52]

    Spectral mesh simplification.Computer Graphics Forum, 39(2):315–324, 2020

    Thibault Lescoat, Hsueh-Ti Derek Liu, Jean-Marc Thiery, Alec Jacobson, Tamy Boubekeur, and Maks Ovsjanikov. Spectral mesh simplification.Computer Graphics Forum, 39(2):315–324, 2020

  45. [53]

    Compressing volumetric radiance fields to 1 mb

    Lingzhi Li, Zhen Shen, Zhongshu Wang, Li Shen, and Liefeng Bo. Compressing volumetric radiance fields to 1 mb. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 4222–4231, 2023

  46. [54]

    Compressing volumetric radiance fields to 1 mb

    Lingzhi Li, Zhen Shen, Zhongshu Wang, Li Shen, and Liefeng Bo. Compressing volumetric radiance fields to 1 mb. In2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 4222–4231, 2023

  47. [55]

    Compact real-time radiance fields with neural codebook

    Lingzhi Li, Zhongshu Wang, Zhen Shen, Li Shen, and Ping Tan. Compact real-time radiance fields with neural codebook. In2023 IEEE International Conference on Multimedia and Expo (ICME), pages 2189–2194. IEEE, 2023

  48. [56]

    NeRFCodec: Neural Feature Compression Meets Neural Radiance Fields for Memory-Efficient Scene Representation

    Sicheng Li, Hao Li, Yiyi Liao, and Lu Yu. NeRFCodec: Neural Feature Compression Meets Neural Radiance Fields for Memory-Efficient Scene Representation . In2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 21274–21283, Los Alamitos, CA, USA, June...

  49. [57]

    Nerfcodec: Neural feature compression meets neural radiance fields for memory-efficient scene representation, 2024

    Sicheng Li, Hao Li, Yiyi Liao, and Lu Yu. Nerfcodec: Neural feature compression meets neural radiance fields for memory-efficient scene representation, 2024

  50. [58]

    An improved decimation of triangle meshes based on curvature

    Wei Li, Yufei Chen, Zhicheng Wang, Weidong Zhao, and Lin Chen. An improved decimation of triangle meshes based on curvature. In Duoqian Miao, Witold Pedrycz, Dominik ´Slezak, Georg Peters, Qinghua Hu, and Ruizhi Wang, editors,Rough Sets and Knowledge Technology, pages 260–271,...

  51. [59]

    Alvarez, Sanja Fidler, Chen Feng, and Anima Anandkumar

    Yiming Li, Zhiding Yu, Christopher Choy, Chaowei Xiao, Jose M. Alvarez, Sanja Fidler, Chen Feng, and Anima Anandkumar. V oxformer: Sparse voxel transformer for camera-based 3d semantic scene completion, 2023. 13

  52. [60]

    Transpcc: Towards deep point cloud compression via transformers

    Zujie Liang and Fan Liang. Transpcc: Towards deep point cloud compression via transformers. In Proceedings of the 2022 International Conference on Multimedia Retrieval, ICMR ’22, page 1–5, New York, NY , USA, 2022. Association for Computing Machinery

  53. [61]

    A comprehensive study and comparison of core technologies for mpeg 3-d point cloud compression.IEEE Transactions on Broadcasting, 66(3):701–717, 2019

    Hao Liu, Hui Yuan, Qi Liu, Junhui Hou, and Ju Liu. A comprehensive study and comparison of core technologies for mpeg 3-d point cloud compression.IEEE Transactions on Broadcasting, 66(3):701–717, 2019

  54. [62]

    Neural subdivision.arXiv preprint arXiv:2005.01819, 2020

    Hsueh-Ti Derek Liu, Vladimir G Kim, Siddhartha Chaudhuri, Noam Aigerman, and Alec Jacobson. Neural subdivision.arXiv preprint arXiv:2005.01819, 2020

  55. [63]

    Variational autoencoder for 3d voxel compression

    Juncheng Liu, Steven Mills, and Brendan McCane. Variational autoencoder for 3d voxel compression. In 2020 35th International Conference on Image and Vision Computing New Zealand (IVCNZ), pages 1–6, 2020

  56. [64]

    Smooth subdivision surfaces based on triangles

    Charles Loop. Smooth subdivision surfaces based on triangles. 1987

  57. [65]

    Zhe Luo, Wenjing Jia, and Stuart W. Perry. Transformer-based geometric point cloud compression with local neighbor aggregation. InDICTA, pages 223–228, 2023

  58. [66]

    Shapesplat: A large-scale dataset of gaussian splats and their self-supervised pretraining.arXiv preprint arXiv:2408.10906, 2024

    Qi Ma, Yue Li, Bin Ren, Nicu Sebe, Ender Konukoglu, Theo Gevers, Luc Van Gool, and Danda Pani Paudel. Shapesplat: A large-scale dataset of gaussian splats and their self-supervised pretraining.arXiv preprint arXiv:2408.10906, 2024

  59. [67]

    Da Silva Cruz

    Miguel Marques and Luís A. Da Silva Cruz. Explorations on 3d point clouds coding using transformers and patches. In2022 10th European Workshop on Visual Information Processing (EUVIP), pages 1–6, 2022

  60. [68]

    Julien N. P. Martel, David B. Lindell, Connor Z. Lin, Eric R. Chan, Marco Monteiro, and Gordon Wetzstein. Acorn: Adaptive coordinate networks for neural scene representation, 2021

  61. [69]

    Oc- cupancy networks: Learning 3d reconstruction in function space

    Lars Mescheder, Michael Oechsle, Michael Niemeyer, Sebastian Nowozin, and Andreas Geiger. Oc- cupancy networks: Learning 3d reconstruction in function space. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 4460–4470, 2019

  62. [70]

    Pcqm: A full-reference quality metric for colored 3d point clouds

    Gabriel Meynet, Yana Nehmé, Julie Digne, and Guillaume Lavoué. Pcqm: A full-reference quality metric for colored 3d point clouds. In2020 Twelfth International Conference on Quality of Multimedia Experience (QoMEX), pages 1–6, 2020

  63. [71]

    Implicit surface representations as layers in neural networks

    Mateusz Michalkiewicz, Jhony K Pontes, Dominic Jack, Mahsa Baktashmotlagh, and Anders Eriks- son. Implicit surface representations as layers in neural networks. InProceedings of the IEEE/CVF International Conference on Computer Vision, pages 4743–4752, 2019

  64. [72]

    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.Commun. ACM, 65(1):99–106, December 2021

  65. [73]

    Variational autoencoders for 3d data processing.Artificial Intelligence Review, 57(2):42, 2024

    Szilárd Molnár and Levente Tamás. Variational autoencoders for 3d data processing.Artificial Intelligence Review, 57(2):42, 2024

  66. [74]

    Instant neural graphics primitives with a multiresolution hash encoding.ACM Transactions on Graphics, 41(4):1–15, July 2022

    Thomas Müller, Alex Evans, Christoph Schied, and Alexander Keller. Instant neural graphics primitives with a multiresolution hash encoding.ACM Transactions on Graphics, 41(4):1–15, July 2022

  67. [75]

    TT-NF: Tensor train neural fields, 2023

    Anton Obukhov, Mikhail Usvyatsov, Christos Sakaridis, Konrad Schindler, and Luc Van Gool. TT-NF: Tensor train neural fields, 2023

  68. [76]

    Deepsdf: Learning continuous signed distance functions for shape representation

    Jeong Joon Park, Peter Florence, Julian Straub, Richard Newcombe, and Steven Lovegrove. Deepsdf: Learning continuous signed distance functions for shape representation. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 165–174, 2019

  69. [77]

    Convolutional occupancy networks

    Songyou Peng, Michael Niemeyer, Lars Mescheder, Marc Pollefeys, and Andreas Geiger. Convolutional occupancy networks. InComputer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Proceedings, Part III 16, pages 523–540. Springer, 2020

  70. [78]

    Neural nerf compression.arXiv preprint arXiv:2406.08943, 2024

    Tuan Pham and Stephan Mandt. Neural nerf compression.arXiv preprint arXiv:2406.08943, 2024

  71. [79]

    Neural mesh simplification

    Rolandos Alexandros Potamias, Stylianos Ploumpis, and Stefanos Zafeiriou. Neural mesh simplification. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 18583– 18592, 2022. 14

  72. [80]

    Pointnet++: Deep hierarchical feature learning on point sets in a metric space

    Charles Ruizhongtai Qi, Li Yi, Hao Su, and Leonidas J Guibas. Pointnet++: Deep hierarchical feature learning on point sets in a metric space. In I. Guyon, U. V on Luxburg, S. Bengio, H. Wallach, R. Fergus, S. Vishwanathan, and R. Garnett, editors,Advances in Neural Information...

  73. [81]

    Nerfmeshing: Distilling neural radiance fields into geometrically-accurate 3d meshes, 2023

    Marie-Julie Rakotosaona, Fabian Manhardt, Diego Martin Arroyo, Michael Niemeyer, Abhijit Kundu, and Federico Tombari. Nerfmeshing: Distilling neural radiance fields into geometrically-accurate 3d meshes, 2023

  74. [82]

    Arce, Nestor Porras-Diaz, Oleg Ieremeiev, Andrii Rubel, Vladimir Lukin, Mateusz Kopytek, Piotr Lech, Jarosław Fastowicz, and Krzysztof Okarma

    Andres Ramirez-Jaime, Gonzalo R. Arce, Nestor Porras-Diaz, Oleg Ieremeiev, Andrii Rubel, Vladimir Lukin, Mateusz Kopytek, Piotr Lech, Jarosław Fastowicz, and Krzysztof Okarma. Generative diffusion models for compressed sensing of satellite lidar data: Evaluating image quality ...

  75. [83]

    Xcube: Large-scale 3d generative modeling using sparse voxel hierarchies

    Xuanchi Ren, Jiahui Huang, Xiaohui Zeng, Ken Museth, Sanja Fidler, and Francis Williams. Xcube: Large-scale 3d generative modeling using sparse voxel hierarchies. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024

  76. [84]

    L3dg: Latent 3d gaussian diffusion

    Barbara Roessle, Norman Müller, Lorenzo Porzi, Samuel Rota Bulò, Peter Kontschieder, Angela Dai, and Matthias Nießner. L3dg: Latent 3d gaussian diffusion. InSIGGRAPH Asia 2024 Conference Papers, December 2024

  77. [85]

    U-net: Convolutional networks for biomedical image segmentation

    Olaf Ronneberger, Philipp Fischer, and Thomas Brox. U-net: Convolutional networks for biomedical image segmentation. In Nassir Navab, Joachim Hornegger, William M. Wells, and Alejandro F. Frangi, editors,Medical Image Computing and Computer-Assisted Intervention – MICCAI 2015,...

  78. [86]

    Rossignac

    J. Rossignac. Edgebreaker: connectivity compression for triangle meshes.IEEE Transactions on Visualization and Computer Graphics, 5(1):47–61, 1999

  79. [87]

    Schroeder, Jonathan A

    William J. Schroeder, Jonathan A. Zarge, and William E. Lorensen. Decimation of triangle meshes. SIGGRAPH Comput. Graph., 26(2):65–70, July 1992

  80. [88]

    Emerging mpeg standards for point cloud compression.IEEE Journal on Emerging and Selected Topics in Circuits and Systems, 9(1):133–148, 2018

    Sebastian Schwarz, Marius Preda, Vittorio Baroncini, Madhukar Budagavi, Pablo Cesar, Philip A Chou, Robert A Cohen, Maja Krivoku ´ca, Sébastien Lasserre, Zhu Li, et al. Emerging mpeg standards for point cloud compression.IEEE Journal on Emerging and Selected Topics in Circuits...

  81. [89]

    3d point cloud attribute compression using diffusion-based texture-aware intra prediction.IEEE Transactions on Circuits and Systems for Video Technology, 34(10):9633–9646, 2024

    Yiting Shao, Xiaodong Yang, Wei Gao, Shan Liu, and Ge Li. 3d point cloud attribute compression using diffusion-based texture-aware intra prediction.IEEE Transactions on Circuits and Systems for Video Technology, 34(10):9633–9646, 2024

  82. [90]

    Polyscope, 2019

    Nicholas Sharp et al. Polyscope, 2019. www.polyscope.run

  83. [91]

    Accelerating nerfs: Optimizing neural radiance fields with special- ized hardware architectures

    William Shen and Willie McClinton. Accelerating nerfs: Optimizing neural radiance fields with special- ized hardware architectures

  84. [92]

    Distilled low rank neural radiance field with quantization for light field compression.arXiv preprint arXiv:2208.00164, 2022

    Jinglei Shi and Christine Guillemot. Distilled low rank neural radiance field with quantization for light field compression.arXiv preprint arXiv:2208.00164, 2022

  85. [93]

    Binary radiance fields.Advances in neural information processing systems, 36, 2024

    Seungjoo Shin and Jaesik Park. Binary radiance fields.Advances in neural information processing systems, 36, 2024

  86. [94]

    Meshgpt: Generating triangle meshes with decoder-only transformers

    Yawar Siddiqui, Antonio Alliegro, Alexey Artemov, Tatiana Tommasi, Daniele Sirigatti, Vladislav Rosov, Angela Dai, and Matthias Nießner. Meshgpt: Generating triangle meshes with decoder-only transformers. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern ...

  87. [95]

    Implicit neural representations with periodic activation functions

    Vincent Sitzmann, Julien Martel, Alexander Bergman, David Lindell, and Gordon Wetzstein. Implicit neural representations with periodic activation functions. In H. Larochelle, M. Ranzato, R. Hadsell, M.F. Balcan, and H. Lin, editors,Advances in Neural Information Processing Sys...

  88. [96]

    Scene representation networks: Continuous 3d-structure-aware neural scene representations.Advances in Neural Information Processing Systems, 32, 2019

    Vincent Sitzmann, Michael Zollhöfer, and Gordon Wetzstein. Scene representation networks: Continuous 3d-structure-aware neural scene representations.Advances in Neural Information Processing Systems, 32, 2019. 15

  89. [97]

    Dropout: A simple way to prevent neural networks from overfitting.Journal of Machine Learning Research, 15(56):1929–1958, 2014

    Nitish Srivastava, Geoffrey Hinton, Alex Krizhevsky, Ilya Sutskever, and Ruslan Salakhutdinov. Dropout: A simple way to prevent neural networks from overfitting.Journal of Machine Learning Research, 15(56):1929–1958, 2014

  90. [98]

    Explicit surface remeshing

    Vitaly Surazhsky and Craig Gotsman. Explicit surface remeshing. InProceedings of the 2003 Eurograph- ics/ACM SIGGRAPH symposium on Geometry processing, pages 20–30, 2003

  91. [99]

    Piecewise regular meshes: Construction and compression.Graphical Models, 64(3-4):183–198, 2002

    Andrzej Szymczak, Jarek Rossignac, and Davis King. Piecewise regular meshes: Construction and compression.Graphical Models, 64(3-4):183–198, 2002

  92. [100]

    Variable bitrate neural fields

    Towaki Takikawa, Alex Evans, Jonathan Tremblay, Thomas Müller, Morgan McGuire, Alec Jacobson, and Sanja Fidler. Variable bitrate neural fields. InACM SIGGRAPH 2022 Conference Proceedings, SIGGRAPH ’22, New York, NY , USA, 2022. Association for Computing Machinery

  93. [101]

    Neural geometric level of detail: Real-time rendering with implicit 3D shapes

    Towaki Takikawa, Joey Litalien, Kangxue Yin, Karsten Kreis, Charles Loop, Derek Nowrouzezahrai, Alec Jacobson, Morgan McGuire, and Sanja Fidler. Neural geometric level of detail: Real-time rendering with implicit 3D shapes. 2021

  94. [102]

    Fourier features let networks learn high frequency functions in low dimensional domains

    Matthew Tancik, Pratul Srinivasan, Ben Mildenhall, Sara Fridovich-Keil, Nithin Raghavan, Utkarsh Singhal, Ravi Ramamoorthi, Jonathan Barron, and Ren Ng. Fourier features let networks learn high frequency functions in low dimensional domains. In H. Larochelle, M. Ranzato, R. Ha...

  95. [103]

    Compressible-composable nerf via rank-residual decomposition

    Jiaxiang Tang, Xiaokang Chen, Jingbo Wang, and Gang Zeng. Compressible-composable nerf via rank-residual decomposition. In S. Koyejo, S. Mohamed, A. Agarwal, D. Belgrave, K. Cho, and A. Oh, editors,Advances in Neural Information Processing Systems, volume 35, pages 14798–14809...

  96. [104]

    Del- icate textured mesh recovery from nerf via adaptive surface refinement.arXiv preprint arXiv:2303.02091, 2022

    Jiaxiang Tang, Hang Zhou, Xiaokang Chen, Tianshu Hu, Errui Ding, Jingdong Wang, and Gang Zeng. Del- icate textured mesh recovery from nerf via adaptive surface refinement.arXiv preprint arXiv:2303.02091, 2022

  97. [105]

    Geometric compression through topological surgery.ACM Trans

    Gabriel Taubin and Jarek Rossignac. Geometric compression through topological surgery.ACM Trans. Graph., 17(2):84–115, April 1998

  98. [106]

    Triangle mesh compression.Proceedings - Graphics Interface, pages 26–34, 1998

    Costa Touma and Craig Gotsman. Triangle mesh compression.Proceedings - Graphics Interface, pages 26–34, 1998. Graphics Interface ’98 ; Conference date: 18-06-1998 Through 20-06-1998

  99. [107]

    Neural discrete representation learning

    Aaron van den Oord, Oriol Vinyals, and koray kavukcuoglu. Neural discrete representation learning. In I. Guyon, U. V on Luxburg, S. Bengio, H. Wallach, R. Fergus, S. Vishwanathan, and R. Garnett, editors, Advances in Neural Information Processing Systems, volume 30. Curran Ass...

  100. [108]

    Sparse tensor-based multiscale representation for point cloud geometry compression.IEEE Transactions on Pattern Analysis and Machine Intelligence, 45(7):9055–9071, 2023

    Jianqiang Wang, Dandan Ding, Zhu Li, Xiaoxing Feng, Chuntong Cao, and Zhan Ma. Sparse tensor-based multiscale representation for point cloud geometry compression.IEEE Transactions on Pattern Analysis and Machine Intelligence, 45(7):9055–9071, 2023

  101. [109]

    Neural radiance fields with hash-low-rank decomposition.Applied Sciences, 14(23), 2024

    Jiaxin Wang, Weichen Dai, Kangcheng Ma, and Wanzeng Kong. Neural radiance fields with hash-low-rank decomposition.Applied Sciences, 14(23), 2024

  102. [110]

    latentsplat: Autoencoding variational gaussians for fast generalizable 3d reconstruction

    Christopher Wewer, Kevin Raj, Eddy Ilg, Bernt Schiele, and Jan Eric Lenssen. latentsplat: Autoencoding variational gaussians for fast generalizable 3d reconstruction. InEuropean Conference on Computer Vision, pages 456–473. Springer, 2024

  103. [111]

    Direct3d: Scalable image-to-3d generation via 3d latent diffusion transformer.arXiv preprint arXiv:2405.14832, 2024

    Shuang Wu, Youtian Lin, Feihu Zhang, Yifei Zeng, Jingxi Xu, Philip Torr, Xun Cao, and Yao Yao. Direct3d: Scalable image-to-3d generation via 3d latent diffusion transformer.arXiv preprint arXiv:2405.14832, 2024

  104. [112]

    Structured 3d latents for scalable and versatile 3d generation.arXiv preprint arXiv:2412.01506, 2024

    Jianfeng Xiang, Zelong Lv, Sicheng Xu, Yu Deng, Ruicheng Wang, Bowen Zhang, Dong Chen, Xin Tong, and Jiaolong Yang. Structured 3d latents for scalable and versatile 3d generation.arXiv preprint arXiv:2412.01506, 2024

  105. [113]

    Hollownerf: Pruning hashgrid-based nerfs with trainable collision mitigation

    Xiufeng Xie, Riccardo Gherardi, Zhihong Pan, and Stephen Huang. Hollownerf: Pruning hashgrid-based nerfs with trainable collision mitigation. InProceedings of the IEEE/CVF International Conference on Computer Vision, pages 3480–3490, 2023

  106. [114]

    Neural fields in visual computing and beyond

    Yiheng Xie, Towaki Takikawa, Shunsuke Saito, Or Litany, Shiqin Yan, Numair Khan, Federico Tombari, James Tompkin, Vincent sitzmann, and Srinath Sridhar. Neural fields in visual computing and beyond. Computer Graphics Forum, 41(2):641–676, 2022. 16

  107. [115]

    Instantmesh: Efficient 3d mesh generation from a single image with sparse-view large reconstruction models.arXiv preprint arXiv:2404.07191, 2024

    Jiale Xu, Weihao Cheng, Yiming Gao, Xintao Wang, Shenghua Gao, and Ying Shan. Instantmesh: Efficient 3d mesh generation from a single image with sparse-view large reconstruction models.arXiv preprint arXiv:2404.07191, 2024

  108. [116]

    Coarse-to-fine transformer for lossless 3d medical image compression

    Xiaoxuan Yang, Guo Lu, Donghui Feng, Zhengxue Cheng, Guosheng Yu, and Li Song. Coarse-to-fine transformer for lossless 3d medical image compression. In2024 IEEE International Conference on Visual Communications and Image Processing (VCIP), pages 1–5, 2024

  109. [117]

    Vq-nerf: Vector quantization enhances implicit neural representations.arXiv preprint arXiv:2310.14487, 2023

    Yiying Yang, Wen Liu, Fukun Yin, Xin Chen, Gang Yu, Jiayuan Fan, and Tao Chen. Vq-nerf: Vector quantization enhances implicit neural representations.arXiv preprint arXiv:2310.14487, 2023

  110. [118]

    Adversarial autoencoders for compact representations of 3d point clouds.Computer Vision and Image Understanding, 193:102921, 2020

    Maciej Zamorski, Maciej Zi˛ eba, Piotr Klukowski, Rafał Nowak, Karol Kurach, Wojciech Stokowiec, and Tomasz Trzci´nski. Adversarial autoencoders for compact representations of 3d point clouds.Computer Vision and Image Understanding, 193:102921, 2020

  111. [119]

    Lion: Latent point diffusion models for 3d shape generation

    Xiaohui Zeng, Arash Vahdat, Francis Williams, Zan Gojcic, Or Litany, Sanja Fidler, and Karsten Kreis. Lion: Latent point diffusion models for 3d shape generation. InAdvances in Neural Information Processing Systems (NeurIPS), 2022

  112. [120]

    3dshape2vecset: A 3d shape repre- sentation for neural fields and generative diffusion models.ACM Transactions On Graphics (TOG), 42(4):1–16, 2023

    Biao Zhang, Jiapeng Tang, Matthias Niessner, and Peter Wonka. 3dshape2vecset: A 3d shape repre- sentation for neural fields and generative diffusion models.ACM Transactions On Graphics (TOG), 42(4):1–16, 2023

  113. [121]

    Compress3d: A compressed latent space for 3d generation from a single image

    Bowen Zhang, Tianyu Yang, Yu Li, Lei Zhang, and Xi Zhao. Compress3d: A compressed latent space for 3d generation from a single image. In Aleš Leonardis, Elisa Ricci, Stefan Roth, Olga Russakovsky, Torsten Sattler, and Gül Varol, editors,Computer Vision – ECCV 2024, pages 276–2...

  114. [122]

    3d representation in 512-byte:variational tokenizer is the key for autoregressive 3d generation, 2024

    Jinzhi Zhang, Feng Xiong, and Mu Xu. 3d representation in 512-byte:variational tokenizer is the key for autoregressive 3d generation, 2024

  115. [123]

    G3pt: Unleash the power of autoregressive modeling in 3d generation via cross-scale querying transformer.arXiv preprint arXiv:2409.06322, 2024

    Jinzhi Zhang, Feng Xiong, and Mu Xu. G3pt: Unleash the power of autoregressive modeling in 3d generation via cross-scale querying transformer.arXiv preprint arXiv:2409.06322, 2024

  116. [124]

    Transformer and upsampling-based point cloud compression

    Junteng Zhang, Gexin Liu, Dandan Ding, and Zhan Ma. Transformer and upsampling-based point cloud compression. InProceedings of the 1st International Workshop on Advances in Point Cloud Compression, Processing and Analysis, APCCPA ’22, page 33–39, New York, NY , USA, 2022. Asso...

  117. [125]

    Efros, Eli Shechtman, and Oliver Wang

    Richard Zhang, Phillip Isola, Alexei A. Efros, Eli Shechtman, and Oliver Wang. The unreasonable effectiveness of deep features as a perceptual metric. In2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 586–595, 2018

  118. [126]

    Rate-aware compression for nerf-based volumetric video

    Zhiyu Zhang, Guo Lu, Huanxiong Liang, Zhengxue Cheng, Anni Tang, and Li Song. Rate-aware compression for nerf-based volumetric video. InProceedings of the 32nd ACM International Conference on Multimedia, pages 3974–3983, 2024

  119. [127]

    Michelangelo: Conditional 3d shape generation based on shape-image-text aligned latent representation.Advances in neural information processing systems, 36:73969–73982, 2023

    Zibo Zhao, Wen Liu, Xin Chen, Xianfang Zeng, Rui Wang, Pei Cheng, Bin Fu, Tao Chen, Gang Yu, and Shenghua Gao. Michelangelo: Conditional 3d shape generation based on shape-image-text aligned latent representation.Advances in neural information processing systems, 36:73969–73982, 2023

  120. [128]

    Vq-nerf: Neural reflectance decomposition and editing with vector quantization.IEEE Transactions on Visualization and Computer Graphics, 30(9):6247–6260, 2023

    Hongliang Zhong, Jingbo Zhang, and Jing Liao. Vq-nerf: Neural reflectance decomposition and editing with vector quantization.IEEE Transactions on Visualization and Computer Graphics, 30(9):6247–6260, 2023

  121. [129]

    Fully convolutional mesh autoencoder using efficient spatially varying kernels

    Yi Zhou, Chenglei Wu, Zimo Li, Chen Cao, Yuting Ye, Jason Saragih, Hao Li, and Yaser Sheikh. Fully convolutional mesh autoencoder using efficient spatially varying kernels. In H. Larochelle, M. Ranzato, R. Hadsell, M.F. Balcan, and H. Lin, editors,Advances in Neural Informatio...

  122. [130]

    within-block

    Denis Zorin, Peter Schröder, and Wim Sweldens. Interpolating subdivision for meshes with arbitrary topology. InProceedings of the 23rd annual conference on Computer graphics and interactive techniques, pages 189–192, 1996. 17 Appendix Contents A Additional Implementation Detai...

Pith tools

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