arxiv: 2605.14594 · v1 · submitted 2026-05-14 · 💻 cs.CV · cs.GR

Recognition: 1 theorem link

· Lean Theorem

TOPOS: High-Fidelity and Efficient Industry-Grade 3D Head Generation

Bojun Xiong , Zoubin Bi , Xinghui Peng , Yunmu Wang , Junchen Deng , Jun Liang , Jing Li , Bowen Cai

show 1 more author

Huan Fu

Authors on Pith no claims yet

Pith reviewed 2026-05-15 01:28 UTC · model grok-4.3

classification 💻 cs.CV cs.GR

keywords 3D head generationsingle-image 3Dfixed topologydigital humansvariational autoencoderrectified flowUV textureindustrial 3D pipeline

0 comments

The pith

TOPOS generates single-image 3D heads locked to one fixed studio topology so every output shares identical vertices for rigging and animation.

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

The paper presents TOPOS as a complete pipeline that takes one portrait and outputs a 3D head mesh plus texture that already follows the exact vertex layout studios require for skinning and animation. It first trains a variational autoencoder whose Perceiver Resampler maps point clouds from arbitrary input meshes onto this single reference topology while keeping geometry and semantic labels intact. A rectified-flow transformer then samples new heads from the resulting structured latent space, and a fine-tuned image model produces UV textures that stay aligned to the mesh. The result is claimed to beat both classical face-reconstruction techniques and general 3D generators on head-specific fidelity and usability. A reader cares because current generative models produce inconsistent topologies that force costly remeshing steps before any asset can enter a real production pipeline.

Core claim

TOPOS recovers both geometry and appearance for heads under one fixed reference topology by training TOPOS-VAE to map diverse point clouds into that topology via its Perceiver Resampler, then training TOPOS-DiT, a rectified flow transformer, to generate new heads from a single image inside the learned latent space, and finally attaching TOPOS-Texture to output relightable UV maps that remain spatially registered to the mesh.

What carries the argument

TOPOS-VAE variational autoencoder that uses the Perceiver Resampler to map point clouds sampled from arbitrary head meshes onto one fixed reference topology while preserving vertex correspondence and geometric detail.

If this is right

All generated heads share exact vertex ordering, so skinning weights and blend shapes transfer directly between any pair without additional correspondence solving.
The structured latent space lets the rectified flow transformer produce new heads at high resolution while remaining compatible with existing studio asset libraries.
UV textures generated by TOPOS-Texture stay pixel-aligned to the fixed mesh, supporting immediate use in relighting and shader pipelines.
The method outperforms both classical single-image reconstruction and topology-agnostic 3D generative models on metrics that matter for digital-human production.

Where Pith is reading between the lines

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

The same resampler-plus-flow design could be applied to full-body or hand generation if a canonical topology is first defined for those domains.
Fixed-topology outputs would reduce the compute cost of batch asset creation in virtual production environments that require thousands of consistent characters.
Because the topology is enforced at generation time, downstream neural rendering or real-time avatar systems could cache shared rig data across many generated heads.

Load-bearing premise

The Perceiver Resampler can map point clouds taken from head meshes of many different topologies onto the single target topology while keeping the geometric accuracy and semantic vertex labels needed for later rigging.

What would settle it

Generate a batch of heads, load them into a standard animation package with the reference rig, and observe whether skinning weights transfer without manual fixes or whether animation artifacts appear at the same vertex locations across all outputs.

Figures

Figures reproduced from arXiv: 2605.14594 by Bojun Xiong, Bowen Cai, Huan Fu, Jing Li, Junchen Deng, Jun Liang, Xinghui Peng, Yunmu Wang, Zoubin Bi.

**Figure 1.** Figure 1: Our proposed TOPOS framework is capable of generating high-fidelity 3D head mesh and texture map given a single image. From left to right are the [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗

**Figure 2.** Figure 2: The top part of this figure shows the structure of our TOPOS-VAE, which utilizes Perceiver Resampler to encode the input pointclouds and GNN decoder [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗

**Figure 3.** Figure 3: The pipeline of our proposed TOPOS-Texture. [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗

**Figure 5.** Figure 5: The first two lines show the results of our texture map augmentation. [PITH_FULL_IMAGE:figures/full_fig_p006_5.png] view at source ↗

**Figure 6.** Figure 6: Qualitative results reconstructed by different methods on our test dataset [PITH_FULL_IMAGE:figures/full_fig_p007_6.png] view at source ↗

**Figure 8.** Figure 8: Qualitative results of our ablation studies on TOPOS-VAE using the [PITH_FULL_IMAGE:figures/full_fig_p007_8.png] view at source ↗

**Figure 10.** Figure 10: Qualitative results generated by different methods on our test dataset [PITH_FULL_IMAGE:figures/full_fig_p008_10.png] view at source ↗

**Figure 11.** Figure 11: Qualitative results generated by different methods on our test [PITH_FULL_IMAGE:figures/full_fig_p008_11.png] view at source ↗

**Figure 12.** Figure 12: Qualitative results generated by different methods on in-the-wild images. For each method, we show both the geometry and texture renderings [PITH_FULL_IMAGE:figures/full_fig_p009_12.png] view at source ↗

**Figure 13.** Figure 13: Qualitative results of our ablation studies on TOPOS-Texture. The [PITH_FULL_IMAGE:figures/full_fig_p009_13.png] view at source ↗

**Figure 14.** Figure 14: The animation results of our generated 3D head mesh. From left to right are the input image, edited image, generated geometry, mesh topology, [PITH_FULL_IMAGE:figures/full_fig_p010_14.png] view at source ↗

**Figure 15.** Figure 15: More generative results of our method on different in-the-wild images. From left to right are the input images, edited images, generated geometry, [PITH_FULL_IMAGE:figures/full_fig_p015_15.png] view at source ↗

read the original abstract

High-fidelity 3D head generation plays a crucial role in the film, animation and video game industries. In industrial pipelines, studios typically enforce a fixed reference topology across all head assets, as such a clean and uniform topology is a prerequisite for production-level rigging, skinning and animation. In this paper, we present TOPOS, a framework tailored for single image conditioned 3D head generation that jointly recovers geometry and appearance under such an industry-standard topology. In contrast to general 3D generative models which produce triangle meshes with inconsistent topology and numerous vertices, hindering semantic correspondence and asset-level reuse, TOPOS generates head meshes with a fixed, studio-style topology, enabling consistent vertex-level correspondence across all generated heads. To model heads under this unified topology, we proposed a novel variational autoencoder structure, termed TOPOS-VAE. Inspired by multi-model large language models (MLLMs), our TOPOS-VAE leverages the Perceiver Resampler to convert input pointclouds sampled from head meshes of diverse topologies into the target reference topology. Building upon TOPOS-VAE's structured latent space, we train a rectified flow transformer, TOPOS-DiT, to efficiently generate high-fidelity head meshes from a single image. We further present TOPOS-Texture, an end-to-end module that produces relightable UV texture maps from the same portrait image via fine-tuning a multimodal image generative model. The generated textures are spatially aligned with the underlying mesh geometry and faithfully preserve high-frequency appearance details. Extensive experiments demonstrate that TOPOS achieves state-of-the-art performance on 3D head generation, surpassing both classical face reconstruction methods and general 3D object generative models, highlighting its effectiveness for digital human creation.

Editorial analysis

A structured set of objections, weighed in public.

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

Desk Editor's Note private letter to a colleague

TOPOS gives a targeted pipeline for single-image fixed-topology head meshes but its SOTA claim sits on experiments we cannot yet inspect.

read the letter

The paper's core move is to enforce a single studio-style mesh topology from the start so the output can go straight into rigging and animation. It does this with a VAE that first resamples arbitrary head point clouds into the reference topology via a Perceiver Resampler, then trains a rectified-flow DiT on the resulting latent space to produce the mesh from one image, plus a separate fine-tuned texture module that aligns UV maps to the geometry. That combination is the concrete new piece; prior single-image head work either allowed varying topology or did not target production constraints this directly. The practical framing around asset reuse and semantic correspondence is also useful and matches real pipeline needs. The abstract states that the system beats both classical reconstruction methods and general 3D generators, but supplies no tables, ablations, or error breakdowns, so the performance margin remains unverified from the text alone. The resampler step is the load-bearing assumption: if it fails to preserve landmark positions or low-frequency shape across diverse inputs, the downstream consistency claims weaken. The paper would have been stronger with even a short quantitative check on vertex error or landmark accuracy for that module. Readers working on constrained 3D asset creation in games or film will find the problem statement and architecture relevant; the rest of the community can skip it. The work is coherent enough on its own terms to merit referee time once the experiments are visible.

Referee Report

2 major / 2 minor

Summary. The paper introduces TOPOS, a single-image conditioned framework for generating 3D head meshes with a fixed industry-standard topology suitable for rigging and animation. It proposes TOPOS-VAE, which uses a Perceiver Resampler to map point clouds from arbitrary head topologies into a reference topology, followed by TOPOS-DiT (a rectified flow transformer) for latent-space generation and TOPOS-Texture for aligned relightable UV maps. The central claim is that this yields state-of-the-art performance on 3D head generation, outperforming both classical face reconstruction methods and general 3D generative models.

Significance. If the performance claims and topology-preservation properties hold, the work would be significant for industrial digital-human pipelines in film, animation, and games, where fixed-topology assets are a prerequisite for downstream skinning and animation without additional retopology steps. The combination of VAE-based topology unification with a DiT generator and texture module addresses a practical gap between academic 3D generation and production requirements.

major comments (2)

[Abstract / TOPOS-VAE] Abstract / TOPOS-VAE description: the claim that the Perceiver Resampler reliably converts point clouds sampled from diverse head topologies into the fixed reference topology while preserving both low-frequency geometry and high-frequency semantic landmarks (eye corners, lip vertices) required for rigging is load-bearing for all downstream claims of consistent vertex correspondence and SOTA performance, yet the manuscript provides no quantitative validation such as per-vertex RMSE, landmark correspondence rates, or ablation comparing resampled versus ground-truth fixed-topology meshes.
[Experiments] Experimental section (presumably §4 or §5): the abstract asserts state-of-the-art results surpassing classical reconstruction and general 3D models, but without visible quantitative tables, ablation studies on the resampler, or error analysis on topology fidelity, the support for the central performance claim cannot be evaluated.

minor comments (2)

[Method] Clarify the exact input/output dimensions and training objective of the Perceiver Resampler within TOPOS-VAE; the current high-level description leaves the mechanism for semantic correspondence ambiguous.
[Related Work] Add explicit comparison to prior topology-aware generation works (e.g., those using fixed-topology templates or mesh autoencoders) in the related-work section to better position the novelty of the Perceiver-based unification.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive review and for acknowledging the practical relevance of fixed-topology 3D head generation for industrial pipelines. We agree that stronger quantitative support is required for the Perceiver Resampler claims and for the SOTA assertions. Below we respond to each major comment and commit to the necessary revisions.

read point-by-point responses

Referee: [Abstract / TOPOS-VAE] Abstract / TOPOS-VAE description: the claim that the Perceiver Resampler reliably converts point clouds sampled from diverse head topologies into the fixed reference topology while preserving both low-frequency geometry and high-frequency semantic landmarks (eye corners, lip vertices) required for rigging is load-bearing for all downstream claims of consistent vertex correspondence and SOTA performance, yet the manuscript provides no quantitative validation such as per-vertex RMSE, landmark correspondence rates, or ablation comparing resampled versus ground-truth fixed-topology meshes.

Authors: We acknowledge that the manuscript does not currently include explicit quantitative metrics (per-vertex RMSE, landmark error rates, or direct ablation against ground-truth fixed-topology meshes) for the Perceiver Resampler. While the VAE training objective and qualitative results support topology unification, these metrics are indeed necessary to substantiate the load-bearing claims. We will add a dedicated quantitative evaluation subsection that reports per-vertex RMSE, landmark correspondence accuracy on semantic points, and an ablation comparing resampled meshes to ground-truth reference-topology meshes. These additions will appear in the revised Experiments section. revision: yes
Referee: [Experiments] Experimental section (presumably §4 or §5): the abstract asserts state-of-the-art results surpassing classical reconstruction and general 3D models, but without visible quantitative tables, ablation studies on the resampler, or error analysis on topology fidelity, the support for the central performance claim cannot be evaluated.

Authors: The current experimental section contains quantitative comparisons, yet we agree that the tables, resampler-specific ablations, and topology-fidelity error analysis are not presented with sufficient clarity or completeness to fully support the SOTA claims. We will revise the Experiments section to feature expanded quantitative tables (geometry and texture metrics against both classical and generative baselines), explicit ablations isolating the resampler, and a focused error analysis on topology fidelity. These changes will make the performance evidence directly evaluable. revision: yes

Circularity Check

0 steps flagged

No circularity in TOPOS derivation chain

full rationale

The paper's chain proceeds from a design choice (fixed industry topology) to a Perceiver-Resampler VAE that maps diverse point clouds into that topology, followed by a rectified-flow DiT conditioned on image latents and a fine-tuned texture generator. No equations, fitted parameters, or self-citations are shown that make any claimed output (geometry, correspondence, or SOTA metrics) equivalent to the inputs by construction. The resampler is presented as an adaptation of prior MLLM work rather than a self-referential fit; downstream performance is asserted via experiments, not by renaming training objectives. The derivation therefore remains independent of its own fitted values.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 3 invented entities

Review based on abstract only; no numerical free parameters, training details, or explicit axioms are stated. The framework introduces three named modules whose internal parameters are not enumerated.

axioms (1)

domain assumption A fixed reference topology is a prerequisite for production-level rigging, skinning and animation in industrial pipelines.
Explicitly stated as the motivation for the work.

invented entities (3)

TOPOS-VAE no independent evidence
purpose: Variational autoencoder that converts diverse-topology point clouds to fixed reference topology using Perceiver Resampler
New structure proposed to enforce unified topology
TOPOS-DiT no independent evidence
purpose: Rectified flow transformer that generates head meshes from single image using the structured latent space
New generative model built on TOPOS-VAE latents
TOPOS-Texture no independent evidence
purpose: End-to-end module producing relightable UV texture maps aligned with the mesh
New texture generation component

pith-pipeline@v0.9.0 · 5639 in / 1443 out tokens · 47273 ms · 2026-05-15T01:28:49.983886+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

145 extracted references · 145 canonical work pages · 12 internal anchors

[1]

Advances in neural information processing systems , volume=

Denoising diffusion probabilistic models , author=. Advances in neural information processing systems , volume=

work page
[2]

Denoising Diffusion Implicit Models

Denoising Diffusion Implicit Models , author=. arXiv:2010.02502 , year=

work page internal anchor Pith review Pith/arXiv arXiv 2010
[3]

International conference on machine learning , pages=

Improved denoising diffusion probabilistic models , author=. International conference on machine learning , pages=. 2021 , organization=

work page 2021
[4]

Flow Matching for Generative Modeling

Flow matching for generative modeling , author=. arXiv preprint arXiv:2210.02747 , year=

work page internal anchor Pith review Pith/arXiv arXiv
[5]

Flow Matching Guide and Code

Flow matching guide and code , author=. arXiv preprint arXiv:2412.06264 , year=

work page internal anchor Pith review Pith/arXiv arXiv
[6]

Advances in neural information processing systems , volume=

Learning a probabilistic latent space of object shapes via 3d generative-adversarial modeling , author=. Advances in neural information processing systems , volume=

work page
[7]

Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

3d shapenets: A deep representation for volumetric shapes , author=. Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

work page
[8]

International conference on machine learning , pages=

Learning representations and generative models for 3d point clouds , author=. International conference on machine learning , pages=. 2018 , organization=

work page 2018
[9]

Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

Diffusion probabilistic models for 3d point cloud generation , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

work page
[10]

Advances in neural information processing systems , volume=

Lion: Latent point diffusion models for 3d shape generation , author=. Advances in neural information processing systems , volume=

work page
[11]

Computer Graphics Forum , volume=

OctFusion: Octree-based Diffusion Models for 3D Shape Generation , author=. Computer Graphics Forum , volume=. 2025 , organization=

work page 2025
[12]

Proceedings of the Special Interest Group on Computer Graphics and Interactive Techniques Conference Conference Papers , pages=

Octgpt: Octree-based multiscale autoregressive models for 3d shape generation , author=. Proceedings of the Special Interest Group on Computer Graphics and Interactive Techniques Conference Conference Papers , pages=

work page
[13]

ACM Transactions on Graphics (ToG) , volume=

Locally attentional sdf diffusion for controllable 3d shape generation , author=. ACM Transactions on Graphics (ToG) , volume=. 2023 , publisher=

work page 2023
[14]

Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

Structured 3d latents for scalable and versatile 3d generation , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

work page
[15]

Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

Sparseflex: High-resolution and arbitrary-topology 3d shape modeling , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

work page
[16]

Seminal graphics: pioneering efforts that shaped the field , pages=

Marching cubes: A high resolution 3D surface construction algorithm , author=. Seminal graphics: pioneering efforts that shaped the field , pages=

work page
[17]

Advances in Neural Information Processing Systems , volume=

Meshxl: Neural coordinate field for generative 3d foundation models , author=. Advances in Neural Information Processing Systems , volume=

work page
[18]

arXiv preprint arXiv:2406.10163 , year=

Meshanything: Artist-created mesh generation with autoregressive transformers , author=. arXiv preprint arXiv:2406.10163 , year=

work page arXiv
[19]

Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

Meshgpt: Generating triangle meshes with decoder-only transformers , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

work page
[20]

Advances in neural information processing systems , volume=

Attention is all you need , author=. Advances in neural information processing systems , volume=

work page
[21]

1999 , isbn =

Blanz, Volker and Vetter, Thomas , title =. 1999 , isbn =. doi:10.1145/311535.311556 , booktitle =

work page doi:10.1145/311535.311556 1999
[22]

2009 sixth IEEE international conference on advanced video and signal based surveillance , pages=

A 3D face model for pose and illumination invariant face recognition , author=. 2009 sixth IEEE international conference on advanced video and signal based surveillance , pages=. 2009 , organization=

work page 2009
[23]

Li, Tianye and Bolkart, Timo and Black, Michael. J. and Li, Hao and Romero, Javier , journal =. Learning a model of facial shape and expression from. 2017 , pages =

work page 2017
[24]

2021 , eprint=

High-Resolution Image Synthesis with Latent Diffusion Models , author=. 2021 , eprint=

work page 2021
[25]

Advances in neural information processing systems , volume=

Flamingo: a visual language model for few-shot learning , author=. Advances in neural information processing systems , volume=

work page
[26]

Advances in neural information processing systems , volume=

Direct preference optimization: Your language model is secretly a reward model , author=. Advances in neural information processing systems , volume=

work page
[27]

1998 , publisher=

Reinforcement learning: An introduction , author=. 1998 , publisher=

work page 1998
[28]

2020 , booktitle=

NeRF: Representing Scenes as Neural Radiance Fields for View Synthesis , author=. 2020 , booktitle=

work page 2020
[29]

3D Gaussian Splatting for Real-Time Radiance Field Rendering , journal =

Kerbl, Bernhard and Kopanas, Georgios and Leimk. 3D Gaussian Splatting for Real-Time Radiance Field Rendering , journal =. 2023 , url =

work page 2023
[30]

International Journal of Computer Vision , volume=

Large scale 3D morphable models , author=. International Journal of Computer Vision , volume=. 2018 , publisher=

work page 2018
[31]

Proceedings of the IEEE international conference on computer vision , pages=

Internet based morphable model , author=. Proceedings of the IEEE international conference on computer vision , pages=

work page
[32]

in-the-wild

3d face morphable models" in-the-wild" , author=. Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

work page
[33]

ACM Transactions on Graphics (ToG) , volume=

Learning an animatable detailed 3D face model from in-the-wild images , author=. ACM Transactions on Graphics (ToG) , volume=. 2021 , publisher=

work page 2021
[34]

European conference on computer vision , pages=

Realy: Rethinking the evaluation of 3d face reconstruction , author=. European conference on computer vision , pages=. 2022 , organization=

work page 2022
[35]

Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

Combining 3d morphable models: A large scale face-and-head model , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

work page
[36]

Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

Ghum & ghuml: Generative 3d human shape and articulated pose models , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

work page
[37]

Proceedings of the ieee/cvf conference on computer vision and pattern recognition , pages=

Facescape: a large-scale high quality 3d face dataset and detailed riggable 3d face prediction , author=. Proceedings of the ieee/cvf conference on computer vision and pattern recognition , pages=

work page
[38]

European conference on computer vision , pages=

Towards metrical reconstruction of human faces , author=. European conference on computer vision , pages=. 2022 , organization=

work page 2022
[39]

arXiv preprint arXiv:2404.04104 , year=

SMIRK: 3D Facial Expressions through Analysis-by-Neural-Synthesis , author=. arXiv preprint arXiv:2404.04104 , year=

work page arXiv
[40]

Proceedings of the AAAI conference on artificial intelligence , volume=

Flowface: Semantic flow-guided shape-aware face swapping , author=. Proceedings of the AAAI conference on artificial intelligence , volume=

work page
[41]

arXiv preprint arXiv:2505.00615 , year=

Pixel3dmm: Versatile screen-space priors for single-image 3d face reconstruction , author=. arXiv preprint arXiv:2505.00615 , year=

work page arXiv
[42]

Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

Instant multi-view head capture through learnable registration , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

work page
[43]

Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

Topologically consistent multi-view face inference using volumetric sampling , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

work page
[44]

Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

Learning detailed face reconstruction from a single image , author=. Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

work page
[45]

Proceedings of the IEEE international conference on computer vision , pages=

Unrestricted facial geometry reconstruction using image-to-image translation , author=. Proceedings of the IEEE international conference on computer vision , pages=

work page
[46]

Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

Df2net: A dense-fine-finer network for detailed 3d face reconstruction , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

work page
[47]

Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

Efficient geometry-aware 3d generative adversarial networks , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

work page
[48]

Neu S : Learning neural implicit surfaces by volume rendering for multi-view reconstruction

Neus: Learning neural implicit surfaces by volume rendering for multi-view reconstruction , author=. arXiv preprint arXiv:2106.10689 , year=

work page arXiv
[49]

The Thirty-eighth Annual Conference on Neural Information Processing Systems , year=

Generalizable and Animatable Gaussian Head Avatar , author=. The Thirty-eighth Annual Conference on Neural Information Processing Systems , year=

work page
[50]

Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

LHM: Large Animatable Human Reconstruction Model for Single Image to 3D in Seconds , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

work page
[51]

IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , year=

GaussianAvatar: Towards Realistic Human Avatar Modeling from a Single Video via Animatable 3D Gaussians , author=. IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , year=

work page
[52]

arXiv preprint arXiv:2508.19754 , year=

FastAvatar: Towards Unified Fast High-Fidelity 3D Avatar Reconstruction with Large Gaussian Reconstruction Transformers , author=. arXiv preprint arXiv:2508.19754 , year=

work page arXiv
[53]

Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

Occupancy networks: Learning 3d reconstruction in function space , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

work page
[54]

Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

Deepsdf: Learning continuous signed distance functions for shape representation , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

work page
[55]

Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

Deformed implicit field: Modeling 3d shapes with learned dense correspondence , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

work page
[56]

Point-E: A System for Generating 3D Point Clouds from Complex Prompts

Point-e: A system for generating 3d point clouds from complex prompts , author=. arXiv preprint arXiv:2212.08751 , year=

work page internal anchor Pith review Pith/arXiv arXiv
[57]

arXiv preprint arXiv:2505.14521 , year=

Sparc3d: Sparse representation and construction for high-resolution 3d shapes modeling , author=. arXiv preprint arXiv:2505.14521 , year=

work page arXiv
[58]

Bowen Jing, Bonnie Berger, and Tommi Jaakkola

Perceiver io: A general architecture for structured inputs & outputs , author=. arXiv preprint arXiv:2107.14795 , year=

work page arXiv
[59]

ACM Transactions On Graphics (TOG) , volume=

3dshape2vecset: A 3d shape representation for neural fields and generative diffusion models , author=. ACM Transactions On Graphics (TOG) , volume=. 2023 , publisher=

work page 2023
[60]

Advances in neural information processing systems , volume=

Michelangelo: Conditional 3d shape generation based on shape-image-text aligned latent representation , author=. Advances in neural information processing systems , volume=

work page
[61]

Hunyuan3D 2.0: Scaling Diffusion Models for High Resolution Textured 3D Assets Generation

Hunyuan3d 2.0: Scaling diffusion models for high resolution textured 3d assets generation , author=. arXiv preprint arXiv:2501.12202 , year=

work page internal anchor Pith review Pith/arXiv arXiv
[62]

Tech report , year=

Native and Compact Structured Latents for 3D Generation , author=. Tech report , year=

work page
[63]

arXiv preprint arXiv:2405.16890 , year=

Pivotmesh: Generic 3d mesh generation via pivot vertices guidance , author=. arXiv preprint arXiv:2405.16890 , year=

work page arXiv
[64]

arXiv preprint arXiv:2409.18114 , year=

Edgerunner: Auto-regressive auto-encoder for artistic mesh generation , author=. arXiv preprint arXiv:2409.18114 , year=

work page arXiv
[65]

Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

Meshanything v2: Artist-created mesh generation with adjacent mesh tokenization , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

work page
[66]

arXiv preprint arXiv:2507.02477 , year=

Mesh Silksong: Auto-Regressive Mesh Generation as Weaving Silk , author=. arXiv preprint arXiv:2507.02477 , year=

work page arXiv
[67]

Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

Scaling mesh generation via compressive tokenization , author=. Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

work page
[68]

Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

Nautilus: Locality-aware autoencoder for scalable mesh generation , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

work page
[69]

arXiv preprint arXiv:2505.13573 , year=

FreeMesh: Boosting Mesh Generation with Coordinates Merging , author=. arXiv preprint arXiv:2505.13573 , year=

work page arXiv
[70]

arXiv preprint arXiv:2412.09548 , year=

Meshtron: High-fidelity, artist-like 3d mesh generation at scale , author=. arXiv preprint arXiv:2412.09548 , year=

work page arXiv
[71]

arXiv preprint arXiv:2503.16653 , year=

iflame: Interleaving full and linear attention for efficient mesh generation , author=. arXiv preprint arXiv:2503.16653 , year=

work page arXiv
[72]

Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

Meshllm: Empowering large language models to progressively understand and generate 3d mesh , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

work page
[73]

arXiv preprint arXiv:2411.09595 , year=

Llama-mesh: Unifying 3d mesh generation with language models , author=. arXiv preprint arXiv:2411.09595 , year=

work page arXiv
[74]

Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

Diffusion model alignment using direct preference optimization , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

work page
[75]

Fast Graph Representation Learning with PyTorch Geometric

Fast graph representation learning with PyTorch Geometric , author=. arXiv preprint arXiv:1903.02428 , year=

work page internal anchor Pith review Pith/arXiv arXiv 1903
[76]

International conference on machine learning , pages=

Neural message passing for quantum chemistry , author=. International conference on machine learning , pages=. 2017 , organization=

work page 2017
[77]

Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

Dynamic edge-conditioned filters in convolutional neural networks on graphs , author=. Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

work page
[78]

Advances in neural information processing systems , volume=

Inductive representation learning on large graphs , author=. Advances in neural information processing systems , volume=

work page
[79]

ACM Transactions on Graphics (tog) , volume=

Dynamic graph cnn for learning on point clouds , author=. ACM Transactions on Graphics (tog) , volume=. 2019 , publisher=

work page 2019
[80]

ACM Transactions on Graphics (TOG) , volume=

Learning the geodesic embedding with graph neural networks , author=. ACM Transactions on Graphics (TOG) , volume=. 2023 , publisher=

work page 2023

Showing first 80 references.