Pith. sign in

REVIEW 5 major objections 5 minor 2 cited by

VTBench: Comprehensive Benchmark Suite Towards Real-World Virtual Try-on Models

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

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

Pith's one-line read Benchmark grades try-on by texture, size, background, and hands

desk verdict Useful multi-dimensional VTON benchmark, but the human-alignment claim is unsupported by the reported evidence and the Similarity win-ratio computation is undefined. read the letter →

arxiv 2505.19571 v1 pith:2AH6BLSA submitted 2025-05-26 cs.CV

Hu Xiaobin , Liang Yujie , Luo Donghao , Peng Xu , Zhang Jiangning , Zhu Junwei , Wang Chengjie , Fu Yanwei This is my paper
classification cs.CV
keywords virtualtry-onbenchmarkunpairedevaluationhumanalignmenttexturefidelitycross-categorysizefitnessbackgroundconsistencyhandocclusion
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

Virtual try-on models are usually graded with a single distribution-distance score such as FID, or with paired reconstruction metrics that need a ground-truth image of the same person wearing the same garment. These scores miss what humans actually notice: whether the garment's texture survived, whether an item from a different category still fits logically, whether the background stayed intact, and whether occluded hands remain consistent. The paper introduces VTBench, a benchmark suite that splits try-on quality into six dimensions, builds a dedicated test set and an unpaired metric for each, evaluates fifteen recent GAN- and diffusion-based models, and claims that its per-dimension scores align strongly with human preference judgments.

What carries the argument

The load-bearing object is the hierarchical Evaluation Dimension Suite: six disentangled dimensions, each paired with a curated test set and an unpaired evaluator. The four novel evaluators are (1) font texture similarity, which runs an OCR model on the garment and generated images and combines edit-distance, term-frequency, and character-level precision/recall/F1; (2) a vision-language-model reasonability score for cross-category size fitness, feeding model, garment, and result images and counting logical-fitting judgments; (3) background consistency computed in a dilated residual ring around the human mask using pixel and semantic-feature distances; and (4) hand consistency measured as mean joint-position error between source and generated poses. These evaluators let each dimension be scored on a single image without paired ground truth, and their outputs are converted to win ratios and correlated with human preferences to justify the benchmark.

What would settle it

Run the released VTBench test sets on the same fifteen models, collect a much larger human preference study with more models, more samples, and multiple annotator pools, and compute the Spearman correlation per dimension; if the correlations fall well below the paper's qualitative 'strong' claim, the central alignment assertion fails. Alternatively, show a specific dimension where the automated metric ranks models opposite to human preference on a new test set.

Watch

Extended reading notes

Core claim

The paper argues that a single similarity score cannot represent virtual try-on quality, so it decomposes quality into general aesthetics and distribution similarity, garment texture fidelity and size fitness, and background and hand consistency. For each dimension VTBench contributes a test set (font-texture garments, cross-category garment-model pairs, complex-background scenes, hand-occluded poses) and an unpaired metric that works without ground truth: OCR-based text similarity for texture, a vision-language model judging whether a cross-category outfit is logically worn, mask-dilated pixel and semantic-feature distance for background, and joint-position error for hands. On fifteen recent models the paper reports that its per-dimension win ratios track human preference annotations, and it uses those measurements to conclude that DiT-based diffusion models preserve garment details best, that FID and KID scores fail to reflect texture and size fidelity, and that explicit hand priors help occluded-hand reconstruction.

Load-bearing premise

The load-bearing premise is that the human preference data, collected from only six models and twenty samples per dimension, are representative enough to prove that VTBench's automated scores rank models the way people would.

Editorial extensions

If this is right

  • Researchers can replace or supplement FID and KID with per-dimension scores that show whether a model fails on texture, size, background, or hands, rather than an aggregate number.
  • The paper's main observations become training guidance: DiT-based architectures improve garment preservation, FID and KID are not good texture or size indicators, and explicit hand priors help occluded-hand cases.
  • The released test sets, generated results, and human annotations give future virtual try-on models a standard yardstick for comparison.
  • The per-dimension win-ratio validation protocol can be reused to test whether any new evaluation metric aligns with human preference.

Reading between the lines

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

  • The same unpaired evaluator recipe (OCR, vision-language judgment, mask-dilated residual, keypoint error) could carry over to other image-editing benchmarks such as outpainting or object replacement, where ground truth is unavailable.
  • The cross-category size metric uses a single expensive vision-language model; VTBench could be extended by auditing whether smaller or open vision-language models give the same rankings before deployment.
  • The texture metric's dependence on OCR means it only sees text-like textures; adding non-text pattern test sets (stripes, florals) would test whether the same human-alignment result holds for textures without readable characters.
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

5 major / 5 minor

Summary. The paper introduces VTBench, a benchmark suite for image-based virtual try-on that decomposes try-on quality into six dimensions (Similarity, Aesthetics, Texture Fidelity, Size Fitness, Background Consistency, and Hand Consistency) and proposes novel unpaired evaluation metrics for four of them, together with four curated test sets. The authors evaluate 15 recent GAN-, UNet-diffusion-, and DiT-diffusion-based try-on models, provide human preference annotations, and claim that the automated metrics align strongly with human perception across all dimensions. The paper also draws conclusions about architecture families, in particular that DiT-based models are superior at garment preservation and that hand priors improve hand-occluded reconstruction.

Significance. If the human-alignment claim were properly established, VTBench would be a useful community resource: it targets real-world challenges (texture transfer, cross-category size adaptation, complex backgrounds, hand occlusion) that existing FID/KID-style evaluations do not address, it provides curated test sets and protocols, and it includes an extensive set of current baselines. The proposed task-specific unpaired evaluators are a promising direction, and the paper is commendable for planning to release test sets, generated results, and human annotations. However, as submitted, the central validation evidence is too thin and partially underspecified: the correlation statistics behind the 'strong alignment' claim are not reported, the Similarity-dimension win-ratio computation is undefined, and key metric details (ETSS normalization, VLM prompting) are not reproducible. The benchmark may still be valuable, but it is not yet a validated evaluation suite.

major comments (5)
  1. [Section 4.2, Fig. 3] The central claim that 'VTBench evaluations across all dimensions exhibit a strong alignment with human perceptual judgments' is not supported by the reported evidence. The paper states that linear regression and Spearman's rank correlation are computed, but it never reports the correlation coefficients, p-values, or confidence intervals. With only six models in the human study (Section 3.3), Spearman's rho can take only a limited set of discrete values and is unlikely to reach conventional significance levels, and 20 samples per dimension is a small basis for a 'large-scale' validation. Please report exact statistics for every dimension and, ideally, enlarge the model set and sample count before claiming strong alignment.
  2. [Section 4.2, Similarity dimension] The per-model 'win ratio' from VTBench automated evaluations is not defined for the Similarity dimension. FID and KID are distribution-level metrics computed over sets of images; they do not produce a per-image score from which pairwise preferences between individual outputs can be derived. The manuscript does not specify the additional scoring rule used to convert FID/KID into a win ratio, so the Similarity panel in Fig. 3 is not reproducible. Please define this mapping explicitly, or exclude Similarity from the per-dimension win-ratio analysis and validate only the dimensions with well-defined per-sample evaluators.
  3. [Section 3.1.2, Eq. (3)] The ETSS formula has an ambiguity in its normalization. If w1, w2, and w3 are all set to 0.2 and the third term is a single average of (Ep + Er + Ef), the effective weight of that composite term depends on whether w3 multiplies the sum or each component; the text as written is not explicit about which interpretation is intended. As a result, the intended score range of ETSS is unclear, which complicates interpretation of the texture-fidelity numbers in Table 1. Please clarify whether w3 applies to the sum or to each of Ep, Er, and Ef, and state the resulting range of ETSS.
  4. [Section 3.1.2, Eq. (4)] The Size Fitness metric depends on an unspecified VLM prompt and a closed API (Qwen-VL-Max). The paper does not report the prompt wording, decoding settings, or API version, so the numbers in the 'VLMs' column of Table 1 cannot be reproduced by the community. In addition, the VLM judgments are not validated separately; their reliability is inferred only from the small human study. Please release the exact prompt and evaluation code, and consider providing results with an open-source VLM or a detailed protocol for the closed API.
  5. [Table 1] The GAN-based baseline rows list '-' in the 'VLMs' (Size Fitness) column, so the claim of evaluating all 15 models across all six dimensions is not actually met. The paper should either run the size-fitness evaluation for the GAN models or explicitly state the per-dimension coverage of baselines. In addition, Table 1 reports no error bars or confidence intervals for any metric, which is particularly important for FID/KID because they are known to be sensitive to sample size and stochastic generation.
minor comments (5)
  1. [Eq. (2)] There is a typo in the denominator: “TF(sp||” should be “TF(sp)”, and the definitions of TP, FP, and FN via set intersection on strings are not mathematically precise; please rewrite these definitions in terms of character or token sets.
  2. [Section 4.2] The text refers to “VBench’s per-dimension assessments”; this should be “VTBench”.
  3. [Section 3.3] The phrase “five-teen binary combinations” contains a typo (“fifteen”), and the description of the annotation effort as “massive” is inconsistent with the stated design of six models and 20 samples per dimension; please reword.
  4. [Eqs. (6) and (7)] The exact computation of “|Isyn − Im|DINO,MD” is not defined; please specify how the DINO feature distance is computed and which DINO model/version is used, and similarly clarify the masked L1 computation for Epixel.
  5. [Eq. (8)] The JPE hand-consistency error is reported in raw pixel units (e.g., 4109 in Table 1) without normalization by image size and without a protocol for cases where hand joints are not detected; please specify the normalization and the failure-handling strategy.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation: VTBench's metrics are independently defined and the human-alignment validation is an external check, not a fitted input.

full rationale

The paper's central contributions are benchmark test sets and unpaired evaluation metrics. The metrics are defined from fixed, pre-existing components: FID/KID for similarity, a CLIP-based aesthetic predictor, an OCR-based font-texture similarity (ETSS), a VLM-based cross-category size evaluator, pixel/DINO background consistency, and joint-position-error for hand consistency. None of these metrics is fitted to human preference labels; the human preference study in Section 3.3 is an independent validation step that compares automated win ratios with human win ratios. Even though the Spearman correlations are not reported and the annotation study uses only 6 models and 20 samples per dimension, that is a statistical-evidence weakness, not a circularity. The ETSS weights (w1=w2=w3=0.2) are hand-chosen constants, not learned from human judgments. The authors' own models (FitDit, CrossVTON, VTON-HandFit) appear among 15 baselines, and Section 5 cites the authors' prior architecture analysis, but the benchmark scores are produced by fixed model-agnostic metrics and the cited architecture claims are independent factual statements. No equation or conclusion reduces by construction to its own input, and no fitted parameter is renamed as a prediction. Therefore the paper exhibits no significant circularity.

Assumptions & free parameters 4 free parameters · 3 assumptions · 0 invented entities

The central claims rest on several hand-chosen parameters and on the reliability of proprietary/external models, plus an untested disentanglement assumption. The weights in the texture metric are particularly fragile because the normalization is inconsistent.

free parameters (4)
  • ETSS weights w1, w2, w3 = 0.2 each
    Manually set weights in Eq. (3) without justification; sum to 0.6, leaving the maximum FTS score below 1.
  • Dilation amount for background mask MD = unspecified
    Eq. (5) dilates the union of masks but the kernel size and iterations are not reported, so the background region varies by implementation.
  • VLM prompt for size fitness = unspecified
    The prompt passed to Qwen-VL-Max determines what counts as 'logically reasonable'; it is not given in the paper, making the metric non-reproducible.
  • Number of hand joints Ns in JPE = unspecified
    JPE in Eq. (8) depends on the skeleton definition and joint count provided by HaMeR; not specified.
assumptions (3)
  • domain assumption External pretrained models (PPOCRv3, Qwen-VL-Max, DINO, HaMeR) produce reliable outputs for the target tasks.
    The metrics assume each off-the-shelf model's errors are negligible; OCR failure or VLM hallucination directly corrupts texture and size scores.
  • ad hoc to paper The six evaluation dimensions are disentangled, so each metric isolates one quality without interference from others.
    The paper claims 'eliminating potential interference' in Section 3.1, but does not test this independence; e.g., background edits affect aesthetics.
  • domain assumption Human preference annotations are accurate ground truth.
    Human alignment is the gold standard throughout; annotator disagreement is not reported.

how reviews work

0 comments
Cite this review

Pith. "Pith review of VTBench: Comprehensive Benchmark Suite Towards Real-World Virtual Try-on Models." pith.science (2026). https://pith.science/paper/2AH6BLSA

@misc{pith2026250519571,
  author       = {Pith},
  title        = {Pith review of: VTBench: Comprehensive Benchmark Suite Towards Real-World Virtual Try-on Models},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/2AH6BLSA}},
  note         = {Machine review of arXiv:2505.19571}
}
read the original abstract

While virtual try-on has achieved significant progress, evaluating these models towards real-world scenarios remains a challenge. A comprehensive benchmark is essential for three key reasons:(1) Current metrics inadequately reflect human perception, particularly in unpaired try-on settings;(2)Most existing test sets are limited to indoor scenarios, lacking complexity for real-world evaluation; and (3) An ideal system should guide future advancements in virtual try-on generation. To address these needs, we introduce VTBench, a hierarchical benchmark suite that systematically decomposes virtual image try-on into hierarchical, disentangled dimensions, each equipped with tailored test sets and evaluation criteria. VTBench exhibits three key advantages:1) Multi-Dimensional Evaluation Framework: The benchmark encompasses five critical dimensions for virtual try-on generation (e.g., overall image quality, texture preservation, complex background consistency, cross-category size adaptability, and hand-occlusion handling). Granular evaluation metrics of corresponding test sets pinpoint model capabilities and limitations across diverse, challenging scenarios.2) Human Alignment: Human preference annotations are provided for each test set, ensuring the benchmark's alignment with perceptual quality across all evaluation dimensions. (3) Valuable Insights: Beyond standard indoor settings, we analyze model performance variations across dimensions and investigate the disparity between indoor and real-world try-on scenarios. To foster the field of virtual try-on towards challenging real-world scenario, VTBench will be open-sourced, including all test sets, evaluation protocols, generated results, and human annotations.

Figures

Figures reproduced from arXiv: 2505.19571 by the authors.

Figure 1
Figure 1. Overview of VTBench. We propose VTBench, the first comprehensive benchmark suite designed to evaluate image-based virtual try-on models. To enable fine-grained and objective assess￾ment, we propose a comprehensive and hierarchical Evaluation Dimension Suite that systematically decomposes “image virtual try-on quality” into multiple well-defined dimensions. For each dimension and content category, we curate a dedicat… view at source ↗
Figure 2
Figure 2. VTBench Evaluation Results of SOTA Virtual try-on Models including GAN, UNet-based and DiT-based Diffusion. In Tab. 1 and [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗
Figure 3
Figure 3. Validate VTBench’s Human Alignment. Our experimental results demonstrate that VTBench evaluations across all dimensions exhibit a strong alignment with human perceptual judgments. Each plot illustrates the verification results for a specific VTBench dimension, where a single dot represents the human preference win rate (x-axis) and the VTBench evaluation win rate (y-axis) for a given virtual try-on generation model.… view at source ↗

Discussion (0). Sign in to comment.

Forward citations

Cited by 2 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. VTEdit-Bench: A Comprehensive Benchmark for Multi-Reference Image Editing Models in Virtual Try-On

    cs.CV 2026-03 conditional novelty 6.0 of 10

    VTEdit-Bench and VTEdit-QA show top universal multi-reference editors match specialized VTON models on standard tasks and transfer more stably to harder multi-person/multi-cloth settings, yet still fail under complex ...

  2. FeRA: Frequency-Energy Constrained Routing for Effective Diffusion Adaptation Fine-Tuning

    cs.CV 2025-11 conditional novelty 5.0 of 10

    A frequency-energy router that blends LoRA experts according to the latent's bandwise energy improves diffusion fine-tuning quality and style consistency across multiple backbones.

Reference graph

Works this paper leans on

49 extracted references · 32 canonical work pages · cited by 2 Pith papers

  1. [1]

    Towards multi-pose guided virtual try-on network

    Haoye Dong, Xiaodan Liang, Xiaohui Shen, Bochao Wang, Hanjiang Lai, Jia Zhu, Zhiting Hu, and Jian Yin. Towards multi-pose guided virtual try-on network. In Proceedings of the IEEE/CVF international conference on computer vision, pages 9026–9035, 2019

    work page 2019

  2. [2]

    Disentangled cycle consistency for highly-realistic virtual try-on

    Chongjian Ge, Yibing Song, Yuying Ge, Han Yang, Wei Liu, and Ping Luo. Disentangled cycle consistency for highly-realistic virtual try-on. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 16928–16937, 2021

    work page 2021

  3. [3]

    Do not mask what you do not need to mask: a parser-free virtual try-on

    Thibaut Issenhuth, Jérémie Mary, and Clément Calauzenes. Do not mask what you do not need to mask: a parser-free virtual try-on. In Computer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Proceedings, Part XX 16, pages 619–635. Springer, 2020

    work page 2020

  4. [4]

    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 Proceedings of the IEEE/CVF international conference on computer vision, pages 10471–10480, 2019

    work page 2019

  5. [5]

    Viton: An image-based virtual try-on network

    Xintong Han, Zuxuan Wu, Zhe Wu, Ruichi Yu, and Larry S Davis. Viton: An image-based virtual try-on network. In Proceedings of the IEEE conference on computer vision and pattern recognition , pages 7543–7552, 2018

    work page 2018

  6. [6]

    Style-based global appearance flow for virtual try-on

    Sen He, Yi-Zhe Song, and Tao Xiang. Style-based global appearance flow for virtual try-on. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 3470–3479, 2022

    work page 2022

  7. [7]

    Stableviton: Learning semantic correspondence with latent diffusion model for virtual try-on

    Jeongho Kim, Guojung Gu, Minho Park, Sunghyun Park, and Jaegul Choo. Stableviton: Learning semantic correspondence with latent diffusion model for virtual try-on. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 8176–8185, 2024

    work page 2024

  8. [8]

    Cp-vton+: Clothing shape and texture preserving image-based virtual try-on

    Matiur Rahman Minar, Thai Thanh Tuan, Heejune Ahn, Paul Rosin, and Yu-Kun Lai. Cp-vton+: Clothing shape and texture preserving image-based virtual try-on. In CVPR workshops, volume 3, pages 10–14, 2020

    work page 2020

Show all 49 references
  1. [9]

    Toward characteristic-preserving image-based virtual try-on network

    Bochao Wang, Huabin Zheng, Xiaodan Liang, Yimin Chen, Liang Lin, and Meng Yang. Toward characteristic-preserving image-based virtual try-on network. In Proceedings of the European conference on computer vision (ECCV), pages 589–604, 2018

    2018

  2. [10]

    Gp-vton: Towards general purpose virtual try-on via collaborative local-flow global-parsing learning

    Zhenyu Xie, Zaiyu Huang, Xin Dong, Fuwei Zhao, Haoye Dong, Xijin Zhang, Feida Zhu, and Xiaodan Liang. Gp-vton: Towards general purpose virtual try-on via collaborative local-flow global-parsing learning. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern ...

    2023

  3. [11]

    Towards photo- realistic virtual try-on by adaptively generating-preserving image content

    Han Yang, Ruimao Zhang, Xiaobao Guo, Wei Liu, Wangmeng Zuo, and Ping Luo. Towards photo- realistic virtual try-on by adaptively generating-preserving image content. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 7850–7859, 2020

    2020

  4. [12]

    Texture-preserving diffusion models for high-fidelity virtual try-on

    Xu Yang, Changxing Ding, Zhibin Hong, Junhao Huang, Jin Tao, and Xiangmin Xu. Texture-preserving diffusion models for high-fidelity virtual try-on. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 7017–7026, 2024

    2024

  5. [13]

    M&m vto: Multi-garment virtual try-on and editing

    Luyang Zhu, Yingwei Li, Nan Liu, Hao Peng, Dawei Yang, and Ira Kemelmacher-Shlizerman. M&m vto: Multi-garment virtual try-on and editing. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 1346–1356, 2024

    2024

  6. [14]

    Generative adversarial networks.Communications of the ACM, 63(11):139– 144, 2020

    Ian Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron Courville, and Yoshua Bengio. Generative adversarial networks.Communications of the ACM, 63(11):139– 144, 2020

    2020

  7. [15]

    High-resolution image synthesis with latent diffusion models

    Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Björn Ommer. High-resolution image synthesis with latent diffusion models. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 10684–10695, 2022

    2022

  8. [16]

    Image quality assessment: from error visibility to structural similarity

    Zhou Wang, Alan C Bovik, Hamid R Sheikh, and Eero P Simoncelli. Image quality assessment: from error visibility to structural similarity. IEEE transactions on image processing, 13(4):600–612, 2004

    2004

  9. [17]

    The unreasonable effectiveness of deep features as a perceptual metric

    Richard Zhang, Phillip Isola, Alexei A Efros, Eli Shechtman, and Oliver Wang. The unreasonable effectiveness of deep features as a perceptual metric. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 586–595, 2018. 10

    2018

  10. [18]

    On aliased resizing and surprising subtleties in gan evaluation

    Gaurav Parmar, Richard Zhang, and Jun-Yan Zhu. On aliased resizing and surprising subtleties in gan evaluation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 11410–11420, 2022

    2022

  11. [19]

    Demystifying mmd gans

    Mikołaj Bi´nkowski, Danica J Sutherland, Michael Arbel, and Arthur Gretton. Demystifying mmd gans. arXiv preprint arXiv:1801.01401, 2018

    2018 arXiv

  12. [20]

    Street tryon: Learning in-the-wild virtual try-on from unpaired person images

    Aiyu Cui, Jay Mahajan, Viraj Shah, Preeti Gomathinayagam, Chang Liu, and Svetlana Lazebnik. Street tryon: Learning in-the-wild virtual try-on from unpaired person images. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 8235–8239, 2024

    2024

  13. [21]

    Parser-free virtual try-on via distilling appearance flows

    Yuying Ge, Yibing Song, Ruimao Zhang, Chongjian Ge, Wei Liu, and Ping Luo. Parser-free virtual try-on via distilling appearance flows. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 8485–8493, 2021

    2021

  14. [22]

    High-resolution virtual try-on with misalignment and occlusion-handled conditions

    Sangyun Lee, Gyojung Gu, Sunghyun Park, Seunghwan Choi, and Jaegul Choo. High-resolution virtual try-on with misalignment and occlusion-handled conditions. In European Conference on Computer Vision, pages 204–219. Springer, 2022

    2022

  15. [23]

    Toward accurate and realistic outfits visualization with attention to details

    Kedan Li, Min Jin Chong, Jeffrey Zhang, and Jingen Liu. Toward accurate and realistic outfits visualization with attention to details. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 15546–15555, 2021

    2021

  16. [24]

    Occlumix: Towards de- occlusion virtual try-on by semantically-guided mixup

    Zhijing Yang, Junyang Chen, Yukai Shi, Hao Li, Tianshui Chen, and Liang Lin. Occlumix: Towards de- occlusion virtual try-on by semantically-guided mixup. IEEE Transactions on Multimedia, 25:1477–1488, 2023

    2023

  17. [25]

    Zflow: Gated appearance flow-based virtual try-on with 3d priors

    Ayush Chopra, Rishabh Jain, Mayur Hemani, and Balaji Krishnamurthy. Zflow: Gated appearance flow-based virtual try-on with 3d priors. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 5433–5442, 2021

    2021

  18. [26]

    Size does matter: Size-aware virtual try-on via clothing-oriented transformation try-on network

    Chieh-Yun Chen, Yi-Chung Chen, Hong-Han Shuai, and Wen-Huang Cheng. Size does matter: Size-aware virtual try-on via clothing-oriented transformation try-on network. In Proceedings of the IEEE/CVF international conference on computer vision, pages 7513–7522, 2023

    2023

  19. [27]

    Wear-any- way: Manipulable virtual try-on via sparse correspondence alignment

    Mengting Chen, Xi Chen, Zhonghua Zhai, Chen Ju, Xuewen Hong, Jinsong Lan, and Shuai Xiao. Wear-any- way: Manipulable virtual try-on via sparse correspondence alignment. arXiv preprint arXiv:2403.12965, 2024

    2024 arXiv

  20. [28]

    Vton-handfit: Virtual try-on for arbitrary hand pose guided by hand priors embedding

    Yujie Liang, Xiaobin Hu, Boyuan Jiang, Donghao Luo, Kai Wu, Wenhui Han, Taisong Jin, and Chengjie Wang. Vton-handfit: Virtual try-on for arbitrary hand pose guided by hand priors embedding. arXiv preprint arXiv:2408.12340, 2024

    2024 arXiv

  21. [29]

    Kolors: Effective training of diffusion model for photorealistic text-to-image synthesis

    Kolors Team. Kolors: Effective training of diffusion model for photorealistic text-to-image synthesis. arXiv preprint, 2024

    2024

  22. [30]

    Tryondiffusion: A tale of two unets

    Luyang Zhu, Dawei Yang, Tyler Zhu, Fitsum Reda, William Chan, Chitwan Saharia, Mohammad Norouzi, and Ira Kemelmacher-Shlizerman. Tryondiffusion: A tale of two unets. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 4606–4615, 2023

    2023

  23. [31]

    Improving diffusion models for virtual try-on

    Yisol Choi, Sangkyung Kwak, Kyungmin Lee, Hyungwon Choi, and Jinwoo Shin. Improving diffusion models for virtual try-on. ECCV, 2024

    2024

  24. [32]

    Anyfit: Controllable virtual try-on for any combination of attire across any scenario

    Yuhan Li, Hao Zhou, Wenxiang Shang, Ran Lin, Xuanhong Chen, and Bingbing Ni. Anyfit: Controllable virtual try-on for any combination of attire across any scenario. arXiv preprint arXiv:2405.18172, 2024

    2024 arXiv

  25. [33]

    Crossvton: Mimicking the logic reasoning on cross-category virtual try-on guided by tri-zone priors

    Donghao Luo, Yujie Liang, Xu Peng, Xiaobin Hu, Boyuan Jiang, Chengming Xu, Taisong Jin, Chengjie Wang, and Yanwei Fu. Crossvton: Mimicking the logic reasoning on cross-category virtual try-on guided by tri-zone priors. arXiv preprint arXiv:2502.14373, 2025

    2025 arXiv

  26. [34]

    Fitdit: Advancing the authentic garment details for high-fidelity virtual try-on

    Boyuan Jiang, Xiaobin Hu, Donghao Luo, Qingdong He, Chengming Xu, Jinlong Peng, Jiangning Zhang, Chengjie Wang, Yunsheng Wu, and Yanwei Fu. Fitdit: Advancing the authentic garment details for high-fidelity virtual try-on. arXiv preprint arXiv:2411.10499, 2024

    2024 arXiv

  27. [35]

    Itvton: Virtual try-on diffusion transformer model based on integrated image and text

    Haifeng Ni. Itvton: Virtual try-on diffusion transformer model based on integrated image and text. arXiv preprint arXiv:2501.16757, 2025

    2025

  28. [36]

    Mc-vton: Minimal control virtual try-on diffusion transformer

    Junsheng Luan, Guangyuan Li, Lei Zhao, and Wei Xing. Mc-vton: Minimal control virtual try-on diffusion transformer. arXiv preprint arXiv:2501.03630, 2025. 11

    2025 arXiv

  29. [37]

    Ted-viton: Transformer-empowered diffusion models for virtual try-on

    Zhenchen Wan, Yanwu Xu, Zhaoqing Wang, Feng Liu, Tongliang Liu, and Mingming Gong. Ted-viton: Transformer-empowered diffusion models for virtual try-on. arXiv preprint arXiv:2411.17017, 2024

    2024 arXiv

  30. [38]

    Viton-dit: Learning in-the-wild video try-on from human dance videos via diffusion transformers

    Jun Zheng, Fuwei Zhao, Youjiang Xu, Xin Dong, and Xiaodan Liang. Viton-dit: Learning in-the-wild video try-on from human dance videos via diffusion transformers. arXiv preprint arXiv:2405.18326, 2024

    2024 arXiv

  31. [39]

    Learning transferable visual models from natural language supervision

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. Learning transferable visual models from natural language supervision. In International conference on machine learning, pa...

    2021

  32. [40]

    Virtualmodel: Gen- erating object-id-retentive human-object interaction image by diffusion model for e-commerce marketing, 2024

    Binghui Chen, Chongyang Zhong, Wangmeng Xiang, Yifeng Geng, and Xuansong Xie. Virtualmodel: Gen- erating object-id-retentive human-object interaction image by diffusion model for e-commerce marketing, 2024

    2024

  33. [41]

    Pp-ocrv3: More attempts for the improvement of ultra lightweight ocr system

    Chenxia Li, Weiwei Liu, Ruoyu Guo, Xiaoting Yin, Kaitao Jiang, Yongkun Du, Yuning Du, Lingfeng Zhu, Baohua Lai, Xiaoguang Hu, et al. Pp-ocrv3: More attempts for the improvement of ultra lightweight ocr system. arXiv preprint arXiv:2206.03001, 2022

    2022 arXiv

  34. [42]

    Shuai Bai, Keqin Chen, Xuejing Liu, Jialin Wang, Wenbin Ge, Sibo Song, Kai Dang, Peng Wang, Shijie Wang, Jun Tang, et al. Qwen2. 5-vl technical report. arXiv preprint arXiv:2502.13923, 2025

    2025 arXiv

  35. [43]

    Reconstructing hands in 3d with transformers

    Georgios Pavlakos, Dandan Shan, Ilija Radosavovic, Angjoo Kanazawa, David Fouhey, and Jitendra Malik. Reconstructing hands in 3d with transformers. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 9826–9836, 2024

    2024

  36. [44]

    Single stage virtual try-on via deformable attention flows

    Shuai Bai, Huiling Zhou, Zhikang Li, Chang Zhou, and Hongxia Yang. Single stage virtual try-on via deformable attention flows. In European Conference on Computer Vision, pages 409–425. Springer, 2022

    2022

  37. [45]

    Towards squeezing-averse virtual try-on via sequential deformation

    Sang-Heon Shim, Jiwoo Chung, and Jae-Pil Heo. Towards squeezing-averse virtual try-on via sequential deformation. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 38, pages 4856– 4863, 2024

    2024

  38. [46]

    Ladi-vton: Latent diffusion textual-inversion enhanced virtual try-on

    Davide Morelli, Alberto Baldrati, Giuseppe Cartella, Marcella Cornia, Marco Bertini, and Rita Cucchiara. Ladi-vton: Latent diffusion textual-inversion enhanced virtual try-on. In Proceedings of the 31st ACM International Conference on Multimedia, pages 8580–8589, 2023

    2023

  39. [47]

    Cat-dm: Con- trollable accelerated virtual try-on with diffusion model

    Jianhao Zeng, Dan Song, Weizhi Nie, Hongshuo Tian, Tongtong Wang, and An-An Liu. Cat-dm: Con- trollable accelerated virtual try-on with diffusion model. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 8372–8382, 2024

    2024

  40. [48]

    Ootdiffusion: Outfitting fusion based latent diffusion for controllable virtual try-on

    Yuhao Xu, Tao Gu, Weifeng Chen, and Chengcai Chen. Ootdiffusion: Outfitting fusion based latent diffusion for controllable virtual try-on. arXiv preprint arXiv:2403.01779, 2024

    2024 arXiv

  41. [49]

    Catvton: Concatenation is all you need for virtual try-on with diffusion models

    Zheng Chong, Xiao Dong, Haoxiang Li, Shiyue Zhang, Wenqing Zhang, Xujie Zhang, Hanqing Zhao, and Xiaodan Liang. Catvton: Concatenation is all you need for virtual try-on with diffusion models. arXiv preprint arXiv:2407.15886, 2024. 12

    2024 arXiv

Pith tools

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