Pith. sign in

REVIEW 3 major objections 5 minor 101 references

PARTE: Part-Guided Texturing for 3D Human Reconstruction from a Single Image

T0 review · 3 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read PARTE shows that conditioning texture generation on explicit 3D human part labels fixes the part-misalignment failure of single-image 3D human reconstruction.

desk verdict PARTE delivers a solid empirical step for part-guided texturing in single-image 3D human reconstruction, but the headline Part IoU gain is partly circular because the same Sapiens segmenter supplies the part labels and the evaluation metric. read the letter →

arxiv 2507.17332 v4 pith:LL3H5IBB submitted 2025-07-23 cs.CV

Hyeongjin Nam , Donghwan Kim , Gyeongsik Moon , Kyoung Mu Lee This is my paper · ORCID
classification cs.CV
keywords 3Dhumanreconstructionsingle-imagepart-guidedtexturingpartsegmentationdiffusionmodelscoredistillationtexturealignment
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 claims that the main quality bottleneck in reconstructing a textured 3D human from one photograph is part misalignment: colors bleed across regions, so a jacket's pattern smears into the pants and the back view is painted without regard to where one garment ends and another begins. To fix this, PARTE treats 3D human part labels as an explicit intermediate and a conditioning signal: a PartSegmenter labels five part classes on the reconstructed surface by rendering normal maps from thirty viewpoints, segmenting each view, and voting the labels into 3D, and a PartTexturer then optimizes surface colors under a reconstruction loss plus score-distillation sampling, with a custom diffusion network that takes the part segments, the front-view image, and per-part text prompts as joint conditions. The reported result is state-of-the-art texture fidelity and part alignment on the THuman2.1 and HuMMan benchmarks, with Part IoU rising from 0.588 (previous best) to 0.641 on THuman2.1, and the same texturer improving the textures of four existing reconstruction pipelines when grafted onto their geometry. The paper's broader assertion is that the structural coherence of human parts is the right cue for inferring textures in regions invisible in the input photo, so reconstruction should be decomposed into part segmentation first and part-conditioned generation second.

What carries the argument

Two coupled modules carry the argument. PartSegmenter contains SegmentNet, a network built from the frozen Transformer layers of a pre-trained human segmentation model, with two branches that exchange features through self-attention: one reads normal maps rendered from the textureless surface at each of thirty viewpoints, capturing geometric cues such as cloth boundaries in regions invisible to the camera, and the other reads the front-view image's part segments, capturing semantic cues such as cloth style; the per-view 2D part maps are unprojected onto the mesh and aggregated by majority voting into per-vertex 3D labels. PartTexturer contains PartDiffusion, a pre-trained latent diffusion network modified with fusion layers that inject three conditions through self-attention: part segments at the rendering viewpoint, per-part image patches formed by multiplying the front-view image with its part segments, and per-part text prompts produced by a captioning model; these conditions enter the score-distillation loss that optimizes the MLP color field, so each region is painted from its own appearance cue rather than from one global prompt.

What would settle it

Run PARTE on a test set of people wearing outfits outside the five-class taxonomy (dresses, skirts, layered or multi-piece clothing, accessories) using ground-truth 3D part annotations, and compare texture misalignment against standard garments: the part-guidance claim predicts a sharp rise in texture bleeding wherever segmentation fails. Alternatively, deliberately corrupt the part segmentations fed to PartTexturer (swap upper/lower labels or erase a part) and check that texture quality drops in proportion; if texture quality is insensitive to segmentation quality, the part-guidance mechanism is not the source of the reported gains.

Watch

Extended reading notes

Core claim

On the paper's own terms, the discovery is that texture misalignment across human parts, a failure common to both regression-based and diffusion-based single-image 3D human reconstruction, can be removed by explicitly conditioning texture generation on dense 3D part segmentations rather than on global image context alone. The framework first builds a textureless human mesh with an off-the-shelf geometry method, labels its surface with five part categories by aggregating 2D segmentations from thirty viewpoints through majority voting in 3D, and then optimizes an MLP color field under a front-view reconstruction loss plus a score-distillation-sampling loss, using a custom diffusion network that ingests the rendered part segments, the front-view image, and per-part text prompts. The experiments report consistent gains: on THuman2.1, PSNR of 22.175 with Part IoU 0.641 versus 21.089 and 0.588 for the best prior method, and similar margins on HuMMan; ablations show that removing part guidance drops Part IoU from 0.641 to 0.461, and that grafting the same texturer onto the geometry of 2K2K, SiTH, HumanRef, and SIFU raises each method's texture quality without altering its geometry.

Load-bearing premise

The method is only as good as its part labels: if the 3D part segmentation is wrong, the part-guided texturing is confidently wrong, and the paper's own limitation section concedes that unseen garment types such as dresses are mis-segmented and therefore mis-textured.

Editorial extensions

If this is right

  • Part-guided texturing acts as a plug-and-play upgrade: replacing the texturing stage of 2K2K, SiTH, HumanRef, SIFU, and TeCH improves their PSNR, LPIPS, and Part IoU without changing their geometry, so existing reconstruction pipelines can adopt PARTE as their texture stage.
  • On ground-truth geometry, PartDiffusion outperforms the diffusion texturing schemes of DreamFusion-style score distillation, HumanRef's reference U-Net, and TeCH's DreamBooth, isolating the texturing gain from geometry errors.
  • The 3D part segmentation intermediate makes the output editable: a single garment can be re-textured by inpainting its 2D projection and re-running PartTexturer, and cloth surfaces can be cut out as 3D assets along part boundaries.
  • Because part labels are voted from thirty viewpoints, the guidance is 3D-consistent rather than front-view-only, which is what allows coherent back- and side-view texturing from a single photo.

Reading between the lines

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

  • The success of this two-stage split suggests a general recipe for compositional 3D reconstruction: identify the object's semantic parts first, then let a generative prior paint each part under explicit region constraints; the same scheme could transfer to animals, vehicles, or multi-object scenes where sub-regions have independent appearances.
  • Per-part text prompts generated automatically from the input image give the diffusion model region-level semantic control at near-zero annotation cost, a cheap way to add part-level controllability to other text-to-3D or image-to-3D pipelines.
  • A stress test the paper does not run: deliberately corrupt the part segmentations fed to PartTexturer and measure the texture-quality drop; the causal claim that part guidance drives the gains predicts a steep, monotone degradation, which would also quantify how much headroom remains for improving SegmentNet.
  • The dress failure implies the five-class taxonomy is the current ceiling; extending to finer or open-vocabulary part labels (separating hair from face, adding accessories, adding garment classes such as dresses) is the natural next step and probably necessary for unrestricted clothing.
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. PARTE is a single-image 3D human reconstruction method that makes explicit use of 3D human part information to guide texture synthesis. The pipeline first reconstructs a textureless surface with TeCH, then PartSegmenter predicts dense part labels on that surface by aggregating 2D segmentations from normal-map renderings at 30 viewpoints, where SegmentNet combines normal-map features with front-view Sapiens segments. PartTexturer then optimizes an MLP color field with a reconstruction loss on the front view and an SDS loss whose diffusion predictor, PartDiffusion, is conditioned on rendered part maps, masked front-view image patches, and per-part text prompts from BLIP. The paper reports improved PSNR, LPIPS, and Part IoU over PIFu, 2K2K, SiTH, HumanRef, SIFU, and TeCH on THuman2.1 and HuMMan, plus ablations and plug-in experiments showing that PARTE improves the texturing of other geometry reconstruction methods.

Significance. The modular design is a clean and plausible contribution: separating dense 3D part estimation from part-conditioned texture synthesis is well motivated, and the plug-in results in Table S1 show that the texturing module can benefit existing geometry pipelines. The PSNR and LPIPS gains, though modest in absolute terms, are consistent across two datasets and across ablations, and they are not affected by the segmentation-metric loop. The main weakness is the evaluation of part alignment: Part IoU is computed with Sapiens, the same model that produces the pseudo-GT labels, the front-view conditioning segments, and the novel-view training segments. As the paper's own limitation section demonstrates for dresses, a Sapiens-consistent texture is not necessarily a correctly part-aligned texture. The part-alignment claim therefore needs an independent check before the headline result can be accepted.

major comments (3)
  1. [Sec. 5.2, Tables 2 and 3; Secs. 3.3/4.3/S5.2] The part-alignment evidence is partly circular. Part IoU is computed by applying Sapiens to rendered images of both the reconstruction and the GT (Sec. 5.2), while Sapiens is also used to generate the pseudo-GT part labels for THuman2.1 and HuMMan (Sec. 5.1), to extract the front-view segments that feed SegmentNet (Sec. 3.3), and to generate the novel-view part segments used to train PartDiffusion (Sec. S5.2). PARTE is therefore optimized, through both modules, to be consistent with Sapiens, whereas the baselines are not; the Part IoU gains may partly measure self-consistency with the labeler rather than true part alignment. Sec. S7 makes the risk concrete: for dresses, Sapiens-based labeling fails, and the Sapiens-based evaluator cannot penalize the resulting misalignment. Please provide an independent part-alignment evaluation, e.g., with a different segmentation model, with 4D-DRESS's own semantic labels, or with human raters.
  2. [Table 1 and Sec. 5.1] The 3D part segmentation metrics (Part CD and Label Acc) are computed against GT labels that are themselves Sapiens outputs, and SegmentNet is initialized from Sapiens and conditioned on Sapiens front-view segments. The comparison with Human3D, which is fine-tuned on the same pseudo-GT but does not use Sapiens at inference, is therefore not a neutral test of segmentation quality. I request results on a dataset with human-verified part labels or with a held-out labeler.
  3. [Sec. S6, Table S4] The 4D-DRESS evaluation is said to use a dataset with accurate 3D part labels, but the table does not state whether Part IoU is computed against those ground-truth labels or via Sapiens. If the GT labels are used, this would be a valuable non-circular check and should be stated prominently; if Sapiens is still used, the circularity concern applies to this table as well.
minor comments (5)
  1. [Sec. S7, Abstract, Conclusion] The admitted limitation that unseen cloth types such as dresses cause segmentation failures and incorrect texturing should be reflected in the abstract and conclusion, which currently claim state-of-the-art 3D human reconstruction without this scope condition.
  2. [Eq. (1) and Sec. S5.2] The relative weighting of L_recon and L_SDS is not reported; the supplementary gives the guidance scale and timestep range but not the loss weights. Please add these values for reproducibility.
  3. [Sec. 5.2] Please specify whether background and the 'others' class are excluded from Part IoU, and whether the metric is averaged per view or over pooled pixels; this affects comparability with baseline numbers.
  4. [Table 3] Because PARTE reuses TeCH's geometry, PARTE's P2S/CD values are identical to TeCH's; the caption should state this explicitly to avoid the impression that PARTE independently produces those geometry numbers.
  5. [Sec. 3.3] The sentence stating that all Transformer weights are initialized from Sapiens and kept frozen is ambiguous because the inserted self-attention layers are trainable; please clarify which parameters are optimized.

Circularity Check

1 steps flagged · score 4.0 of 10

Part IoU and the 3D part-segmentation metrics are computed with Sapiens, the same model that supplies PARTE's pseudo-GT labels and conditioning signals, so the headline part-alignment gain is partly self-consistency; PSNR and LPIPS remain independent.

  1. self definitional [Sec. 5.1 (Datasets); Sec. 5.2 (Evaluation Metrics); also Sec. 3.3, Sec. 4.3, Sec. S5.2]
    "To obtain ground truth (GT) part segmentations for the scans, we apply Sapiens [38] to 360 rendered images from uniformly distributed viewpoints and aggregate the results in 3D space. ... Part IoU evaluates the part alignment of the reconstructed human textures in 2D space. It is computed as the average IoU of part segments of the rendered images, where each segment is obtained by applying Sapiens [38] to the rendered images."

    Part alignment is not measured against an independent label source. The GT part labels used for supervision and evaluation are Sapiens outputs (Sec. 5.1), the front-view and novel-view Sapiens segments are used as conditioning and training inputs (Sec. 3.3, Sec. 4.3, Sec. S5.2), and Part IoU is computed by running Sapiens on rendered images (Sec. 5.2). Thus PARTE is explicitly trained and conditioned to produce textures whose Sapiens-based segmentations coincide with Sapiens-based pseudo-GT, and then scored on that same coincidence. Baselines are not conditioned on Sapiens segments, so the Part IoU advantage partly measures in-distribution self-consistency rather than true part alignment.

full rationale

The core method is not circular: PartSegmenter and PartDiffusion are trained on held-out THuman2.1 scans with reconstruction and SDS objectives, and the headline texture-fidelity claim is independently supported by PSNR and LPIPS on rendered images from held-out scans. There is no load-bearing self-citation chain; Sapiens is an external model. The circularity is confined to the part-alignment and part-segmentation evaluation: the pseudo-GT labels, the conditioning signals, and the Part IoU metric all come from the same Sapiens segmenter, so PARTE is rewarded for reproducing Sapiens labels that it was trained and conditioned to reproduce. This can inflate Part IoU and the related 3D segmentation metrics over baselines independent of true part alignment. The score is 4 rather than higher because PSNR and LPIPS provide independent evidence for the central texture-quality claim, and the paper's own S7 limitation exposes the failure mode rather than hiding it.

Assumptions & free parameters 7 free parameters · 4 assumptions · 2 invented entities

The listed hand-chosen hyperparameters affect the reconstruction quality but are standard design choices. The domain assumptions are the load-bearing premises: Sapiens label reliability, SDS prior validity, normal-map informativeness, and TeCH geometry quality. No new physical entities are introduced; PartSegmenter and PartDiffusion are new modules validated on held-out data.

free parameters (7)
  • number_of_part_categories = 5
    Hand-chosen taxonomy (face & hair, upper, lower, footwear, others) inherited from Sapiens; affects all downstream part guidance and evaluation.
  • viewpoints_for_3D_voting = 30
    Number of rendered normal-map viewpoints for 2D-to-3D label aggregation (Sec 3.2).
  • SegmentNet_transformer_layers_L = 10
    Number of Transformer layers before inserted self-attention in SegmentNet (Sec 3.3).
  • SDS_guidance_scale = 100
    Classifier-free guidance scale in Eq. (2), set in Sec S5.2; high value is common for SDS but is a hand-chosen hyperparameter.
  • MLP_hidden_dimension = 32
    Texture MLP width in PartTexturer (Sec S5.2).
  • hash_encoding_max_resolution = 2048
    Positional encoding resolution for texture MLP (Sec S5.2).
  • noise_timestep_range = [0.02, 0.98]
    SDS random timestep range (Sec S5.2).
assumptions (4)
  • domain assumption Sapiens provides reliable 2D human part segmentation for arbitrary images.
    Used to create pseudo-GT labels for training and evaluation (Sec 5.1) and to extract conditioning part segments (Sec 3.3, 4.3). Segmentation errors propagate to part guidance.
  • domain assumption A pre-trained diffusion model's score function is a valid prior for human appearance in SDS loss.
    Eq. (2) uses SDS with StableDiffusion to infer invisible textures; standard in SDS literature but not proven for this task.
  • domain assumption Normal maps of a textureless human surface contain sufficient structural cues to infer part labels in invisible regions.
    Core premise of PartSegmenter (Sec 3.2, Fig. 3); supported by ablations but still a modeling assumption.
  • domain assumption TeCH geometry reconstruction is accurate enough for part segmentation and texturing.
    PARTE inherits TeCH geometry (P2S 2.984 cm, CD 3.008 cm); geometric errors degrade normal maps and texturing.
invented entities (2)
  • PartSegmenter independent evidence
    purpose: Predicts 3D part labels on a textureless human surface from normal maps and front-view image.
    New module; validated on held-out THuman2.1 and HuMMan via Label Acc and Part CD.
  • PartDiffusion independent evidence
    purpose: Part-guided diffusion network that infers texture colors for novel views conditioned on part segments, input image, and text prompts.
    New architecture; validated in 2D image generation and 3D texturing ablations.

how reviews work

0 comments
Cite this review

Pith. "Pith review of PARTE: Part-Guided Texturing for 3D Human Reconstruction from a Single Image." pith.science (2026). https://pith.science/paper/LL3H5IBB

@misc{pith2026250717332,
  author       = {Pith},
  title        = {Pith review of: PARTE: Part-Guided Texturing for 3D Human Reconstruction from a Single Image},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/LL3H5IBB}},
  note         = {Machine review of arXiv:2507.17332}
}
read the original abstract

The misaligned human texture across different human parts is one of the main limitations of existing 3D human reconstruction methods. Each human part, such as a jacket or pants, should maintain a distinct texture without blending into others. The structural coherence of human parts serves as a crucial cue to infer human textures in the invisible regions of a single image. However, most existing 3D human reconstruction methods do not explicitly exploit such part segmentation priors, leading to misaligned textures in their reconstructions. In this regard, we present PARTE, which utilizes 3D human part information as a key guide to reconstruct 3D human textures. Our framework comprises two core components. First, to infer 3D human part information from a single image, we propose a 3D part segmentation module (PartSegmenter) that initially reconstructs a textureless human surface and predicts human part labels based on the textureless surface. Second, to incorporate part information into texture reconstruction, we introduce a part-guided texturing module (PartTexturer), which acquires prior knowledge from a pre-trained image generation network on texture alignment of human parts. Extensive experiments demonstrate that our framework achieves state-of-the-art quality in 3D human reconstruction. The project page is available at https://hygenie1228.github.io/PARTE/.

Figures

Figures reproduced from arXiv: 2507.17332 by the authors.

Figure 1
Figure 1. Overview of PARTE. Our framework reconstructs a realistic 3D textured human from a single image by effectively improving part alignment with 3D human part information. Despite remarkable advances in 3D reconstruction of hu￾man geometry [23, 55, 72, 85, 86], reconstructing high￾fidelity textures has been relatively unexplored. Most ex￾isting methods [6, 11, 25, 27, 35, 71, 89, 96, 100] train their networks on 3D huma… view at source ↗
Figure 2
Figure 2. Overall pipeline of PARTE. Our framework initially reconstructs a textureless human mesh from a single image and textures it based on two core modules. PartSegmenter predicts 3D human parts of the textureless human mesh by incorporating information from the input image and normal maps of the textureless human surface. Based on the 3D human part segmentations, PartTexturer reconstructs the human textures using PartDi… view at source ↗
Figure 3
Figure 3. Detailed process of PartSegmenter. (a) From a tex￾tureless human surface, PartSegmenter performs normal rendering and acquires part segments via SegmentNet. The part segments are used to vote for 3D human part labels on the human surface. (b) SegmentNet incorporates features from the normal map and the front-view image for accurate part segmentation. 3.2. 3D part segmentation [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: (a) illustrates an overview of PartDiffusion. PartD￾iffusion takes three conditioning inputs, part segments S k , input image I, and text prompts {Pi} n i=1 and predicts the noise corresponding to the noisy image x k t , where k is a rendering viewpoint. To integrate t…
Figure 5
Figure 5. Figure 5: Impact of (a) PartSegmenter and (b) PartDiffusion in 3D human texture reconstruction. duces ambiguity in capturing semantic cues, such as cloth style. On the other hand, our PartSegmenter utilizes front￾view image information along with geometric information (i.e., nor…
Figure 6
Figure 6. Figure 6: Qualitative comparison with existing 3D human reconstruction methods: SiTH [27], HumanRef [94], and TeCH [34]. THuman2.1 HuMMan 3D geometry reconstruction 3D texture reconstruction 3D geometry reconstruction 3D texture reconstruction Methods P2S↓ CD↓ PSNR↑ LPIPS↓ Part …

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

101 extracted references · 77 canonical work pages

  1. [1]

    Gaussian shell maps for efficient 3D human generation

    Rameen Abdal, Wang Yifan, Zifan Shi, Yinghao Xu, Ryan Po, Zhengfei Kuang, Qifeng Chen, Dit-Yan Yeung, and Gordon Wetzstein. Gaussian shell maps for efficient 3D human generation. In CVPR, 2024. 3

    2024

  2. [2]

    Single-image 3D human digitization with shape-guided diffusion

    Badour AlBahar, Shunsuke Saito, Hung-Yu Tseng, Changil Kim, Johannes Kopf, and Jia-Bin Huang. Single-image 3D human digitization with shape-guided diffusion. In SIG- GRAPH Asia, 2023. 1

    2023

  3. [3]

    Learning to re- construct people in clothing from a single rgb camera

    Thiemo Alldieck, Marcus Magnor, Bharat Lal Bhatnagar, Christian Theobalt, and Gerard Pons-Moll. Learning to re- construct people in clothing from a single rgb camera. In CVPR, 2019. 2

    2019

  4. [4]

    imGHUM: Implicit generative models of 3D human shape and articulated pose

    Thiemo Alldieck, Hongyi Xu, and Cristian Sminchisescu. imGHUM: Implicit generative models of 3D human shape and articulated pose. In ICCV, 2021. 3

    2021

  5. [5]

    Photorealistic monocular 3D reconstruction of hu- mans wearing clothing

    Thiemo Alldieck, Mihai Zanfir, and Cristian Sminchis- escu. Photorealistic monocular 3D reconstruction of hu- mans wearing clothing. In CVPR, 2022. 2

    2022

  6. [6]

    Photorealistic monocular 3D reconstruction of hu- mans wearing clothing

    Thiemo Alldieck, Mihai Zanfir, and Cristian Sminchis- escu. Photorealistic monocular 3D reconstruction of hu- mans wearing clothing. In CVPR, 2022. 1

    2022

  7. [7]

    Learning on graph with lapla- cian regularization

    Rie Ando and Tong Zhang. Learning on graph with lapla- cian regularization. In NeurIPS, 2006. 3

    2006

  8. [8]

    MultiDiffusion: Fusing diffusion paths for controlled image generation

    Omer Bar-Tal, Lior Yariv, Yaron Lipman, and Tali Dekel. MultiDiffusion: Fusing diffusion paths for controlled image generation. Proceedings of Machine Learning Research ,

Show all 101 references
  1. [9]

    HuMMan: Multi-modal 4D human dataset for versatile sensing and modeling

    Zhongang Cai, Daxuan Ren, Ailing Zeng, Zhengyu Lin, Tao Yu, Wenjia Wang, Xiangyu Fan, Yang Gao, Yi- fan Yu, Liang Pan, Fangzhou Hong, Mingyuan Zhang, Chen Change Loy, Lei Yang, and Ziwei Liu. HuMMan: Multi-modal 4D human dataset for versatile sensing and modeling. In ECCV, 2022. 1, 6

    2022

  2. [10]

    DreamAvatar: Text-and-shape guided 3D human avatar generation via diffusion models

    Yukang Cao, Yan-Pei Cao, Kai Han, Ying Shan, and Kwan- Yee K Wong. DreamAvatar: Text-and-shape guided 3D human avatar generation via diffusion models. In CVPR,

  3. [11]

    Gener- alizable human gaussians from single-view image

    Jinnan Chen, Chen Li, Jianfeng Zhang, Lingting Zhu, Buzhen Huang, Hanlin Chen, and Gim Hee Lee. Gener- alizable human gaussians from single-view image. arXiv preprint arXiv:2406.06050, 2024. 1

    2024 arXiv

  4. [12]

    Fan- tasia3D: Disentangling geometry and appearance for high- quality text-to-3D content creation

    Rui Chen, Yongwei Chen, Ningxin Jiao, and Kui Jia. Fan- tasia3D: Disentangling geometry and appearance for high- quality text-to-3D content creation. In ICCV, 2023. 2

    2023

  5. [13]

    PrimDiffusion: V olumetric primitives diffusion for 3D human generation

    Zhaoxi Chen, Fangzhou Hong, Haiyi Mei, Guangcong Wang, Lei Yang, and Ziwei Liu. PrimDiffusion: V olumetric primitives diffusion for 3D human generation. In NeurIPS,

  6. [14]

    Structured 3D features for reconstructing con- trollable avatars

    Enric Corona, Mihai Zanfir, Thiemo Alldieck, Ed- uard Gabriel Bazavan, Andrei Zanfir, and Cristian Smin- chisescu. Structured 3D features for reconstructing con- trollable avatars. In CVPR, 2023. 2

    2023

  7. [15]

    An efficient method of triangu- lating equi-valued surfaces by using tetrahedral cells

    Akio Doi and Akio Koide. An efficient method of triangu- lating equi-valued surfaces by using tetrahedral cells. IE- ICE TRANSACTIONS on Information and Systems , 1991. 3

    1991

  8. [16]

    Soft-gated warping-GAN for pose- guided person image synthesis

    Haoye Dong, Xiaodan Liang, Ke Gong, Hanjiang Lai, Jia Zhu, and Jian Yin. Soft-gated warping-GAN for pose- guided person image synthesis. NIPS, 2018. 3

    2018

  9. [17]

    AG3D: Learning to gen- erate 3D avatars from 2D image collections

    Zijian Dong, Xu Chen, Jinlong Yang, Michael J Black, Ot- mar Hilliges, and Andreas Geiger. AG3D: Learning to gen- erate 3D avatars from 2D image collections. InICCV, 2023. 3

    2023

  10. [18]

    Collaborative regression of expressive bodies using moderation

    Yao Feng, Vasileios Choutas, Timo Bolkart, Dimitrios Tzionas, and Michael J Black. Collaborative regression of expressive bodies using moderation. In 3DV, 2021. 3

    2021

  11. [19]

    StyleGAN-Human: A data-centric odyssey of human gen- eration

    Jianglin Fu, Shikai Li, Yuming Jiang, Kwan-Yee Lin, Chen Qian, Chen Change Loy, Wayne Wu, and Ziwei Liu. StyleGAN-Human: A data-centric odyssey of human gen- eration. In ECCV, 2022. 2, 4, 6, 7, 8, 9

    2022

  12. [20]

    Text-guided 3D human generation from 2D collections

    Tsu-Jui Fu, Wenhan Xiong, Yixin Nie, Jingyu Liu, Barlas Oguz, and William Yang Wang. Text-guided 3D human generation from 2D collections. In EMNLP, 2023. 3

    2023

  13. [21]

    ConTex-Human: Free-view rendering of human from a single image with texture-consistent synthesis

    Xiangjun Gao, Xiaoyu Li, Chaopeng Zhang, Qi Zhang, Yanpei Cao, Ying Shan, and Long Quan. ConTex-Human: Free-view rendering of human from a single image with texture-consistent synthesis. In CVPR, 2024. 2

    2024

  14. [22]

    StylePeople: A generative model of fullbody human avatars

    Artur Grigorev, Karim Iskakov, Anastasia Ianina, Renat Bashirov, Ilya Zakharkin, Alexander Vakhitov, and Victor Lempitsky. StylePeople: A generative model of fullbody human avatars. In CVPR, 2021. 3

    2021

  15. [23]

    High-fidelity 3D hu- man digitization from single 2k resolution images

    Sang-Hun Han, Min-Gyu Park, Ju Hong Yoon, Ju-Mi Kang, Young-Jae Park, and Hae-Gon Jeon. High-fidelity 3D hu- man digitization from single 2k resolution images. In CVPR, 2023. 1, 2, 8

    2023

  16. [24]

    ClothFlow: A flow-based model for clothed person generation

    Xintong Han, Xiaojun Hu, Weilin Huang, and Matthew R Scott. ClothFlow: A flow-based model for clothed person generation. In ICCV, 2019. 3

    2019

  17. [25]

    ARCH++: Animation-ready clothed human reconstruction revisited

    Tong He, Yuanlu Xu, Shunsuke Saito, Stefano Soatto, and Tony Tung. ARCH++: Animation-ready clothed human reconstruction revisited. In ICCV, 2021. 1, 2

    2021

  18. [26]

    Learning locally editable virtual humans

    Hsuan-I Ho, Lixin Xue, Jie Song, and Otmar Hilliges. Learning locally editable virtual humans. In CVPR, 2023. 1

    2023

  19. [27]

    SiTH: Single-view textured human reconstruction with image-conditioned dif- fusion

    I Ho, Jie Song, Otmar Hilliges, et al. SiTH: Single-view textured human reconstruction with image-conditioned dif- fusion. In CVPR, 2024. 1, 2, 8, 6

    2024

  20. [28]

    Classifier-free diffusion guidance

    Jonathan Ho and Tim Salimans. Classifier-free diffusion guidance. In NeurIPS workshop, 2021. 4

    2021

  21. [29]

    AvatarCLIP: zero-shot text- driven generation and animation of 3D avatars.TOG, 2022

    Fangzhou Hong, Mingyuan Zhang, Liang Pan, Zhongang Cai, Lei Yang, and Ziwei Liu. AvatarCLIP: zero-shot text- driven generation and animation of 3D avatars.TOG, 2022. 3

    2022

  22. [30]

    EV A3D: Compositional 3D human generation from 2D image collections

    Fangzhou Hong, Zhaoxi Chen, LAN Yushi, Liang Pan, and Ziwei Liu. EV A3D: Compositional 3D human generation from 2D image collections. In ICLR, 2023

    2023

  23. [31]

    StructLDM: Structured latent diffusion for 3D human generation

    Tao Hu, Fangzhou Hong, and Ziwei Liu. StructLDM: Structured latent diffusion for 3D human generation. In ECCV, 2024

    2024

  24. [32]

    HumanNorm: Learn- ing normal diffusion model for high-quality and realistic 3D human generation

    Xin Huang, Ruizhi Shao, Qi Zhang, Hongwen Zhang, Ying Feng, Yebin Liu, and Qing Wang. HumanNorm: Learn- ing normal diffusion model for high-quality and realistic 3D human generation. In CVPR, 2024

    2024

  25. [33]

    DreamWaltz: Make a scene with complex 3D animatable avatars

    Yukun Huang, Jianan Wang, Ailing Zeng, He Cao, Xi- anbiao Qi, Yukai Shi, Zheng-Jun Zha, and Lei Zhang. DreamWaltz: Make a scene with complex 3D animatable avatars. In NeurIPS, 2024. 3

    2024

  26. [34]

    TeCH: Text- guided reconstruction of lifelike clothed humans

    Yangyi Huang, Hongwei Yi, Yuliang Xiu, Tingting Liao, Jiaxiang Tang, Deng Cai, and Justus Thies. TeCH: Text- guided reconstruction of lifelike clothed humans. In 3DV,

  27. [35]

    ARCH: Animatable reconstruction of clothed humans

    Zeng Huang, Yuanlu Xu, Christoph Lassner, Hao Li, and Tony Tung. ARCH: Animatable reconstruction of clothed humans. In CVPR, 2020. 1, 2

    2020

  28. [36]

    Zero-shot text-guided object gen- eration with dream fields

    Ajay Jain, Ben Mildenhall, Jonathan T Barron, Pieter Abbeel, and Ben Poole. Zero-shot text-guided object gen- eration with dream fields. In CVPR, 2022. 2

    2022

  29. [37]

    Avatar- Craft: Transforming text into neural human avatars with parameterized shape and pose control

    Ruixiang Jiang, Can Wang, Jingbo Zhang, Menglei Chai, Mingming He, Dongdong Chen, and Jing Liao. Avatar- Craft: Transforming text into neural human avatars with parameterized shape and pose control. In ICCV, 2023. 3

    2023

  30. [38]

    Sapiens: Foundation for human vi- sion models

    Rawal Khirodkar, Timur Bagautdinov, Julieta Martinez, Su Zhaoen, Austin James, Peter Selednik, Stuart Anderson, and Shunsuke Saito. Sapiens: Foundation for human vi- sion models. In ECCV, 2024. 4, 5, 6, 3

    2024

  31. [39]

    Chupa: Carving 3D clothed humans from skinned shape priors us- ing 2D diffusion probabilistic models

    Byungjun Kim, Patrick Kwon, Kwangho Lee, Myunggi Lee, Sookwan Han, Daesik Kim, and Hanbyul Joo. Chupa: Carving 3D clothed humans from skinned shape priors us- ing 2D diffusion probabilistic models. In ICCV, 2023. 3

    2023

  32. [40]

    Kingma and Jimmy Ba

    Diederik P. Kingma and Jimmy Ba. Adam: A method for stochastic optimization. In ICLR, 2015. 3, 4

    2015

  33. [41]

    DreamHuman: Animatable 3D avatars from text

    Nikos Kolotouros, Thiemo Alldieck, Andrei Zanfir, Ed- uard Bazavan, Mihai Fieraru, and Cristian Sminchisescu. DreamHuman: Animatable 3D avatars from text. In NeurIPS, 2023. 3

    2023

  34. [42]

    Instant 3D human avatar generation using image diffusion models

    Nikos Kolotouros, Thiemo Alldieck, Enric Corona, Ed- uard Gabriel Bazavan, and Cristian Sminchisescu. Instant 3D human avatar generation using image diffusion models. In ECCV, 2024. 1, 3

    2024

  35. [43]

    360-degree textures of people in clothing from a single im- age

    Verica Lazova, Eldar Insafutdinov, and Gerard Pons-Moll. 360-degree textures of people in clothing from a single im- age. In 3DV, 2019. 2

    2019

  36. [44]

    BLIP: Bootstrapping language-image pre-training for uni- fied vision-language understanding and generation

    Junnan Li, Dongxu Li, Caiming Xiong, and Steven Hoi. BLIP: Bootstrapping language-image pre-training for uni- fied vision-language understanding and generation. In ICML, 2022. 5, 4

    2022

  37. [45]

    Object-driven text-to-image synthesis via adversarial training

    Wenbo Li, Pengchuan Zhang, Lei Zhang, Qiuyuan Huang, Xiaodong He, Siwei Lyu, and Jianfeng Gao. Object-driven text-to-image synthesis via adversarial training. In CVPR,

  38. [46]

    Collaging class-specific GANs for semantic image synthesis

    Yuheng Li, Yijun Li, Jingwan Lu, Eli Shechtman, Yong Jae Lee, and Krishna Kumar Singh. Collaging class-specific GANs for semantic image synthesis. In ICCV, 2021. 3

    2021

  39. [47]

    High-fidelity clothed avatar re- construction from a single image

    Tingting Liao, Xiaomei Zhang, Yuliang Xiu, Hongwei Yi, Xudong Liu, Guo-Jun Qi, Yong Zhang, Xuan Wang, Xi- angyu Zhu, and Zhen Lei. High-fidelity clothed avatar re- construction from a single image. In CVPR, 2023. 2

    2023

  40. [48]

    Tingting Liao, Hongwei Yi, Yuliang Xiu, Jiaxiang Tang, Yangyi Huang, Justus Thies, and Michael J. Black. TADA! text to animatable digital avatars. In 3DV, 2024. 3

    2024

  41. [49]

    Microsoft COCO: Common objects in context

    Tsung-Yi Lin, Michael Maire, Serge Belongie, James Hays, Pietro Perona, Deva Ramanan, Piotr Doll ´ar, and C Lawrence Zitnick. Microsoft COCO: Common objects in context. In ECCV, 2014. 2, 4

    2014

  42. [50]

    DeepFashion: Powering robust clothes recognition and retrieval with rich annotations

    Ziwei Liu, Ping Luo, Shi Qiu, Xiaogang Wang, and Xiaoou Tang. DeepFashion: Powering robust clothes recognition and retrieval with rich annotations. In CVPR, 2016. 2, 4

    2016

  43. [51]

    A convnet for the 2020s

    Zhuang Liu, Hanzi Mao, Chao-Yuan Wu, Christoph Feicht- enhofer, Trevor Darrell, and Saining Xie. A convnet for the 2020s. In CVPR, 2022. 4

    2022

  44. [52]

    Decoupled weight decay regularization

    Ilya Loshchilov and Frank Hutter. Decoupled weight decay regularization. In ICLR, 2019. 4

    2019

  45. [53]

    Learning semantic per- son image generation by region-adaptive normalization

    Zhengyao Lv, Xiaoming Li, Xin Li, Fu Li, Tianwei Lin, Dongliang He, and Wangmeng Zuo. Learning semantic per- son image generation by region-adaptive normalization. In CVPR, 2021. 3

    2021

  46. [54]

    Learn- ing to dress 3D people in generative clothing

    Qianli Ma, Jinlong Yang, Anurag Ranjan, Sergi Pujades, Gerard Pons-Moll, Siyu Tang, and Michael J Black. Learn- ing to dress 3D people in generative clothing. In CVPR,

  47. [55]

    En3D: An enhanced generative model for sculpting 3D humans from 2D synthetic data

    Yifang Men, Biwen Lei, Yuan Yao, Miaomiao Cui, Zhouhui Lian, and Xuansong Xie. En3D: An enhanced generative model for sculpting 3D humans from 2D synthetic data. In CVPR, 2024. 1, 3

    2024

  48. [56]

    3D clothed human reconstruction in the wild

    Gyeongsik Moon, Hyeongjin Nam, Takaaki Shiratori, and Kyoung Mu Lee. 3D clothed human reconstruction in the wild. In ECCV, 2022. 2

    2022

  49. [57]

    DeClotH: Decomposable 3D cloth and hu- man body reconstruction from a single image

    Hyeongjin Nam, Donghwan Kim, Jeongtaek Oh, and Ky- oung Mu Lee. DeClotH: Decomposable 3D cloth and hu- man body reconstruction from a single image. In CVPR, 2025

    2025

  50. [58]

    HumanSplat: Generalizable single-image human gaus- sian splatting with structure priors

    Panwang Pan, Zhuo Su, Chenguo Lin, Zhen Fan, Yongjie Zhang, Zeming Li, Tingting Shen, Yadong Mu, and Yebin Liu. HumanSplat: Generalizable single-image human gaus- sian splatting with structure priors. In NeurIPS, 2025. 2

    2025

  51. [59]

    Semantic image synthesis with spatially-adaptive normalization

    Taesung Park, Ming-Yu Liu, Ting-Chun Wang, and Jun- Yan Zhu. Semantic image synthesis with spatially-adaptive normalization. In CVPR, 2019. 3

    2019

  52. [60]

    Automatic differentiation in PyTorch

    Adam Paszke, Sam Gross, Soumith Chintala, Gregory Chanan, Edward Yang, Zachary DeVito, Zeming Lin, Al- ban Desmaison, Luca Antiga, and Adam Lerer. Automatic differentiation in PyTorch. 2017. 2

    2017

  53. [61]

    Georgios Pavlakos, Vasileios Choutas, Nima Ghorbani, Timo Bolkart, Ahmed A. A. Osman, Dimitrios Tzionas, and Michael J. Black. Expressive body capture: 3D hands, face, and body from a single image. In CVPR, 2019. 3

    2019

  54. [62]

    CharacterGen: Efficient 3D char- acter generation from single images with multi-view pose canonicalization

    Hao-Yang Peng, Jia-Peng Zhang, Meng-Hao Guo, Yan-Pei Cao, and Shi-Min Hu. CharacterGen: Efficient 3D char- acter generation from single images with multi-view pose canonicalization. TOG, 2024. 2

    2024

  55. [63]

    Barron, and Ben Milden- hall

    Ben Poole, Ajay Jain, Jonathan T. Barron, and Ben Milden- hall. DreamFusion: Text-to-3D using 2D diffusion. In ICLR, 2023. 2, 5

    2023

  56. [64]

    PointNet: Deep learning on point sets for 3D classification and segmentation

    Charles R Qi, Hao Su, Kaichun Mo, and Leonidas J Guibas. PointNet: Deep learning on point sets for 3D classification and segmentation. In CVPR, 2017. 4

    2017

  57. [65]

    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 NeurIPS, 2017. 4

    2017

  58. [66]

    Learn- ing transferable visual models from natural language super- vision

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. Learn- ing transferable visual models from natural language super- vision. In ICML, 2021. 3, 4

    2021

  59. [67]

    https://renderpeople.com/ 3d-people

    Renderpeople, 2018. https://renderpeople.com/ 3d-people. 1

    2018

  60. [68]

    High-resolution image synthesis with latent diffusion models

    Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Bj ¨orn Ommer. High-resolution image synthesis with latent diffusion models. In CVPR, 2022. 5, 7, 1, 2, 4

    2022

  61. [69]

    High-resolution image synthesis with latent diffusion models

    Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Bj ¨orn Ommer. High-resolution image synthesis with latent diffusion models. In CVPR, 2022. 1, 2, 3

    2022

  62. [70]

    DreamBooth: Fine tuning text-to-image diffusion models for subject-driven generation

    Nataniel Ruiz, Yuanzhen Li, Varun Jampani, Yael Pritch, Michael Rubinstein, and Kfir Aberman. DreamBooth: Fine tuning text-to-image diffusion models for subject-driven generation. In CVPR, 2023. 2

    2023

  63. [71]

    PIFu: Pixel-aligned implicit function for high-resolution clothed human digiti- zation

    Shunsuke Saito, Zeng Huang, Ryota Natsume, Shigeo Mor- ishima, Angjoo Kanazawa, and Hao Li. PIFu: Pixel-aligned implicit function for high-resolution clothed human digiti- zation. In ICCV, 2019. 1, 2, 8

    2019

  64. [72]

    PIFuHD: Multi-level pixel-aligned implicit function for high-resolution 3D human digitization

    Shunsuke Saito, Tomas Simon, Jason Saragih, and Hanbyul Joo. PIFuHD: Multi-level pixel-aligned implicit function for high-resolution 3D human digitization. In CVPR, 2020. 1, 2

    2020

  65. [73]

    DiffHuman: probabilistic photorealistic 3D reconstruction of humans

    Akash Sengupta, Thiemo Alldieck, Nikos Kolotouros, En- ric Corona, Andrei Zanfir, and Cristian Sminchisescu. DiffHuman: probabilistic photorealistic 3D reconstruction of humans. In CVPR, 2024. 2

    2024

  66. [74]

    Deep marching tetrahedra: a hybrid represen- tation for high-resolution 3D shape synthesis

    Tianchang Shen, Jun Gao, Kangxue Yin, Ming-Yu Liu, and Sanja Fidler. Deep marching tetrahedra: a hybrid represen- tation for high-resolution 3D shape synthesis. In NeurIPS,

  67. [75]

    Unsuper- vised person image generation with semantic parsing trans- formation

    Sijie Song, Wei Zhang, Jiaying Liu, and Tao Mei. Unsuper- vised person image generation with semantic parsing trans- formation. In CVPR, 2019. 3

    2019

  68. [76]

    Image synthesis from reconfig- urable layout and style

    Wei Sun and Tianfu Wu. Image synthesis from reconfig- urable layout and style. In ICCV, 2019. 3

    2019

  69. [77]

    3D segmentation of humans in point clouds with synthetic data

    Ayc ¸a Takmaz, Jonas Schult, Irem Kaftan, Mertcan Akc ¸ay, Bastian Leibe, Robert Sumner, Francis Engelmann, and Siyu Tang. 3D segmentation of humans in point clouds with synthetic data. In ICCV, 2023. 4, 6, 7

    2023

  70. [78]

    Complete 3D human reconstruction from a single incomplete image

    Junying Wang, Jae Shin Yoon, Tuanfeng Y Wang, Kr- ishna Kumar Singh, and Ulrich Neumann. Complete 3D human reconstruction from a single incomplete image. In CVPR, 2023. 2

    2023

  71. [79]

    4D-DRESS: A 4D dataset of real-world human clothing with semantic annotations

    Wenbo Wang, Hsuan-I Ho, Chen Guo, Boxiang Rong, Artur Grigorev, Jie Song, Juan Jose Zarate, and Otmar Hilliges. 4D-DRESS: A 4D dataset of real-world human clothing with semantic annotations. In CVPR, 2024. 4

    2024

  72. [80]

    Gene- MAN: Generalizable single-image 3D human reconstruc- tion from multi-source human data

    Wentao Wang, Hang Ye, Fangzhou Hong, Xue Yang, Jianfu Zhang, Yizhou Wang, Ziwei Liu, and Liang Pan. Gene- MAN: Generalizable single-image 3D human reconstruc- tion from multi-source human data. In ICLR, 2025. 2

    2025

  73. [81]

    InstanceDiffusion: Instance- level control for image generation

    Xudong Wang, Trevor Darrell, Sai Saketh Rambhatla, Ro- hit Girdhar, and Ishan Misra. InstanceDiffusion: Instance- level control for image generation. In CVPR, 2024. 3, 7, 4

    2024

  74. [82]

    ProlificDreamer: High- fidelity and diverse text-to-3D generation with variational score distillation

    Zhengyi Wang, Cheng Lu, Yikai Wang, Fan Bao, Chongx- uan Li, Hang Su, and Jun Zhu. ProlificDreamer: High- fidelity and diverse text-to-3D generation with variational score distillation. In NeurIPS, 2023. 2

    2023

  75. [83]

    Photo Wake-Up: 3D character animation from a single photo

    Chung-Yi Weng, Brian Curless, and Ira Kemelmacher- Shlizerman. Photo Wake-Up: 3D character animation from a single photo. In CVPR, 2019. 2

    2019

  76. [84]

    Get3DHuman: Lifting StyleGAN-human into a 3D gener- ative model using pixel-aligned reconstruction priors

    Zhangyang Xiong, Di Kang, Derong Jin, Weikai Chen, Linchao Bao, Shuguang Cui, and Xiaoguang Han. Get3DHuman: Lifting StyleGAN-human into a 3D gener- ative model using pixel-aligned reconstruction priors. In ICCV, 2023. 3

    2023

  77. [85]

    ICON: Implicit clothed humans obtained from normals

    Yuliang Xiu, Jinlong Yang, Dimitrios Tzionas, and Michael J Black. ICON: Implicit clothed humans obtained from normals. In CVPR, 2022. 1

    2022

  78. [86]

    ECON: Explicit clothed humans opti- mized via normal integration

    Yuliang Xiu, Jinlong Yang, Xu Cao, Dimitrios Tzionas, and Michael J Black. ECON: Explicit clothed humans opti- mized via normal integration. In CVPR, 2023. 1

    2023

  79. [87]

    3D human texture es- timation from a single image with transformers

    Xiangyu Xu and Chen Change Loy. 3D human texture es- timation from a single image with transformers. In ICCV,

  80. [88]

    XAGen: 3D expressive human avatars generation

    Zhongcong Xu, Jianfeng Zhang, Jun Hao Liew, Jiashi Feng, and Mike Zheng Shou. XAGen: 3D expressive human avatars generation. In NeurIPS, 2023. 3

    2023

  81. [89]

    Human-3Diffusion: Realistic avatar creation via explicit 3D consistent diffusion models

    Yuxuan Xue, Xianghui Xie, Riccardo Marin, and Gerard Pons-Moll. Human-3Diffusion: Realistic avatar creation via explicit 3D consistent diffusion models. arXiv preprint arXiv:2406.08475, 2024. 1

    2024 arXiv

  82. [90]

    ReCo: Region-controlled text-to- image generation

    Zhengyuan Yang, Jianfeng Wang, Zhe Gan, Linjie Li, Kevin Lin, Chenfei Wu, Nan Duan, Zicheng Liu, Ce Liu, Michael Zeng, et al. ReCo: Region-controlled text-to- image generation. In CVPR, 2023. 3

    2023

  83. [91]

    Function4D: Real-time human vol- umetric capture from very sparse consumer RGBD sensors

    Tao Yu, Zerong Zheng, Kaiwen Guo, Pengpeng Liu, Qiong- hai Dai, and Yebin Liu. Function4D: Real-time human vol- umetric capture from very sparse consumer RGBD sensors. In CVPR, 2021. 1, 6, 2, 4, 5

    2021

  84. [92]

    Semantic human mesh recon- struction with textures

    Xiaoyu Zhan, Jianxin Yang, Yuanqi Li, Jie Guo, Yanwen Guo, and Wenping Wang. Semantic human mesh recon- struction with textures. In CVPR, 2024. 1, 2

    2024

  85. [93]

    3D- aware semantic-guided generative model for human synthe- sis

    Jichao Zhang, Enver Sangineto, Hao Tang, Aliaksandr Siarohin, Zhun Zhong, Nicu Sebe, and Wei Wang. 3D- aware semantic-guided generative model for human synthe- sis. In ECCV, 2022. 3

    2022

  86. [94]

    HumanRef: Single image to 3D hu- man generation via reference-guided diffusion

    Jingbo Zhang, Xiaoyu Li, Qi Zhang, Yanpei Cao, Ying Shan, and Jing Liao. HumanRef: Single image to 3D hu- man generation via reference-guided diffusion. In CVPR,

  87. [95]

    The unreasonable effectiveness of deep features as a perceptual metric

    Richard Zhang, Phillip Isola, Alexei A Efros, Eli Shecht- man, and Oliver Wang. The unreasonable effectiveness of deep features as a perceptual metric. In CVPR, 2018. 6

    2018

  88. [96]

    Global-correlated 3d-decoupling transformer for clothed avatar reconstruction

    Zechuan Zhang, Li Sun, Zongxin Yang, Ling Chen, and Yi Yang. Global-correlated 3d-decoupling transformer for clothed avatar reconstruction. In NeurIPS, 2023. 1

    2023

  89. [97]

    SIFU: Side- view conditioned implicit function for real-world usable clothed human reconstruction

    Zechuan Zhang, Zongxin Yang, and Yi Yang. SIFU: Side- view conditioned implicit function for real-world usable clothed human reconstruction. In CVPR, 2024. 1, 2, 8

    2024

  90. [98]

    Point transformer

    Hengshuang Zhao, Li Jiang, Jiaya Jia, Philip HS Torr, and Vladlen Koltun. Point transformer. In ICCV, 2021. 4

    2021

  91. [99]

    LayoutDiffusion: Controllable diffusion model for layout-to-image generation

    Guangcong Zheng, Xianpan Zhou, Xuewei Li, Zhongang Qi, Ying Shan, and Xi Li. LayoutDiffusion: Controllable diffusion model for layout-to-image generation. In CVPR,

  92. [100]

    PaMIR: Parametric model-conditioned implicit representa- tion for image-based human reconstruction, 2021

    Zerong Zheng, Tao Yu, Yebin Liu, and Qionghai Dai. PaMIR: Parametric model-conditioned implicit representa- tion for image-based human reconstruction, 2021. 1, 2

    2021

  93. [101]

    SEAN: Image synthesis with semantic region-adaptive nor- malization

    Peihao Zhu, Rameen Abdal, Yipeng Qin, and Peter Wonka. SEAN: Image synthesis with semantic region-adaptive nor- malization. In CVPR, 2020. 3

    2020

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.