Pith. sign in

REVIEW 5 major objections 5 minor 62 references

Pursuing Temporal-Consistent Video Virtual Try-On via Dynamic Pose Interaction

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 A diffusion model that injects synchronized human and garment poses into spatial and temporal attention achieves state-of-the-art video virtual try-on, cutting VFID to 0.506 on VVT.

desk verdict A well-built video try-on system with a clean architecture and a big reported SOTA jump, but the paper never isolates the human–garment pose interaction that is its headline mechanism. read the letter →

arxiv 2505.16980 v1 pith:TYCH7KCG submitted 2025-05-22 cs.CV cs.MM

Dong Li , Wenqi Zhong , Wei Yu , Yingwei Pan , Dingwen Zhang , Ting Yao , Junwei Han , Tao Mei This is my paper · ORCID
classification cs.CVcs.MM
keywords videovirtualtry-ondiffusionmodelsposeinteractiontemporalconsistencypose-awareattentiongarmentestimationspatiotemporalgeneration
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

This paper argues that video virtual try-on fails to keep garments temporally consistent because existing methods ignore the dynamic interaction between the person's pose and the garment's pose across time. To fix that, it proposes Dynamic Pose Interaction Diffusion Models (DPIDM), which injects synchronized human and garment pose features into both spatial and temporal attention layers of a diffusion network, and adds a loss that regularizes self-attention maps across consecutive frames. The paper reports that DPIDM beats prior methods on VITON-HD, VVT, and ViViD, with a VFID of 0.506 on VVT, a 60.5% relative improvement over the strongest prior approach. If correct, the work shows that pose-interaction modeling, not just a larger temporal module, is what drives both garment fidelity and temporal coherence in video try-on.

What carries the argument

The central machinery is a hierarchical attention module inside the main U-Net of a latent diffusion model. Pose-aware spatial attention (PASA) injects human and garment pose embeddings into the self-attention that fuses garment and person features, so the garment deforms to fit the body within each frame. Pose-aware temporal attention (PATA) injects the human pose embedding into temporal attention to capture long-term pose dynamics across the whole video. Temporal-shift attention (TSA) reuses a 2D attention block with shifted frames to capture short-term motion cheaply. The temporal regularized attention (TRA) loss, computed on the last two decoder layers, minimizes the difference between self-attention maps of consecutive frames. A custom garment pose estimator trained on manually labeled landmarks produces the garment pose that makes the alignment possible.

What would settle it

Compare DPIDM's VFID where pose-aware attention is driven by the custom garment pose estimator against a variant driven by ground-truth manually verified garment landmarks; if the custom estimator's alignments are systematically faulty, the performance gain from pose-aware attention would shrink or disappear.

Watch

Extended reading notes

Core claim

The paper claims that modeling spatiotemporal pose interaction between human and garment is the missing ingredient for video virtual try-on. Its DPIDM uses a skeleton-based pose adapter to add pose embeddings to the queries, keys, and values of spatial self-attention (pose-aware spatial attention) and to the temporal attention that tracks long-term human pose dynamics (pose-aware temporal attention). A temporal-shift attention block handles short-term motion, and a temporal regularized attention loss forces self-attention maps of successive frames to stay similar. On the VVT dataset this yields a VFID of 0.506, a 60.5% improvement over GPD-VVTO, and consistent gains on ViViD and VITON-HD. The paper interprets these results as evidence that explicit pose alignment between garment and body, plus pose-conditioned temporal attention, is the mechanism that preserves garment identity while keeping videos temporally stable.

Load-bearing premise

The paper's central mechanism relies on the custom garment pose estimator producing landmark alignments accurate enough to guide generation, yet that estimator is trained on manually labeled data with no released protocol and no evaluated accuracy.

Editorial extensions

If this is right

  • Video try-on systems can treat pose interaction as first-class conditioning: injecting pose into attention layers, rather than as an extra input channel, yields measurable VFID reductions.
  • Combining short-range temporal shift attention with long-range pose-aware temporal attention covers both fast and slow motion, reducing flicker in generated garment fabric.
  • The temporal regularized attention loss is a training signal that directly targets attention-map stability, and could be transferred to other video generation tasks beyond try-on.
  • The same dual-branch diffusion architecture, when stripped of temporal modules, still produces competitive image try-on results on VITON-HD, showing that pose-aware spatial attention alone drives a large part of the improvement.

Reading between the lines

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

  • A reader should expect the reported 60.5% improvement to be sensitive to garment-pose landmark quality: because the garment pose estimator is trained on manually labeled data with no released protocol or accuracy numbers, the pose-aware attention's benefit is only as reliable as those labels.
  • The method's reliance on a custom garment pose estimator suggests a natural stress test: comparing DPIDM against a variant that uses only human pose, or a variant that uses ground-truth garment landmarks, would isolate whether the garment-pose channel truly causes the gains.
  • The temporal regularized attention loss may generalize beyond try-on: any video diffusion task with objects that should stay locally consistent across frames, such as face reenactment or object editing, could adopt the same 'stabilize self-attention across frames' objective.
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 proposes DPIDM, a latent diffusion framework for video virtual try-on that injects synchronized human and garment pose embeddings into spatial and temporal attention modules. The main components are pose-aware spatial attention (PASA), pose-aware temporal attention (PATA), temporal-shift attention (TSA), and a temporal regularized attention loss (TRA). The authors report state-of-the-art results on VITON-HD, VVT, and ViViD, including a large VFID improvement on VVT. The paper includes ablations on the VVT dataset showing that each proposed component improves the final metric.

Significance. If the claims hold, this would be a solid contribution to video virtual try-on, demonstrating that explicit pose-aware attention can improve temporal coherence and garment fidelity. The main strengths are the clean integration of pose conditioning into a diffusion backbone and a reasonable set of ablations that show the overall benefit of the proposed modules. However, the evaluation has important gaps: the specific contribution of garment-pose interaction is not isolated, the custom garment pose estimator is never validated, and the ViViD comparison uses a filtered test set against external baseline numbers. The lack of error bars and code/weights further limits confidence in the reported state-of-the-art results.

major comments (5)
  1. [Section 3.4 and Table 4] The central claim of human–garment pose interaction is not isolated. The 'PAA' ablation in Table 4 adds both PASA and PATA together, where PASA injects both human and garment pose embeddings and PATA uses human pose only. There is no ablation that removes the garment-pose branch from PASA while keeping human-pose conditioning. As a result, the large improvement from configuration (a) to (b) could be caused by human-pose conditioning alone, which prior methods such as PoseGuider already provide. Please add an ablation with human-pose-only PASA (and if feasible, garment-pose-only) to demonstrate that the human–garment alignment is actually load-bearing.
  2. [Section 3.3] The custom garment pose estimator is a load-bearing component but its accuracy is never reported. The paper states that ground-truth landmarks are manually labeled, but it does not describe the annotation protocol, the number of training images, the train/test split, or any quantitative metric such as PCK or endpoint error. Without an evaluation of this estimator, it is unclear whether the pose-aware attention operates on reliable alignments. Please report the estimator's accuracy on a held-out set and include qualitative failure cases.
  3. [Section 4.1 and Table 2] The ViViD test set is modified by excluding back-facing segments, but the baseline numbers in Table 2 are taken from the original papers that evaluated on the full test set. This makes the comparison unfair and could inflate the reported 74.2% improvement over ViViD. Please re-evaluate all baselines on the exact same filtered test set, or at minimum report the number of excluded clips and the precise filtering rule. If re-evaluation is impractical, this limitation should be stated explicitly.
  4. [Tables 1, 2, and 4] All results are reported as single runs without error bars or standard deviations. Since VFID is computed on generated video samples and thus has stochastic variability, the extremely large differences such as 0.506 vs 1.280 may be robust, but smaller differences (e.g., SSIM 0.930 vs 0.929 in Table 4) could be within noise. Please provide variance estimates over multiple runs or seeds, at least for the VVT and ViViD main results, to support the 'consistently outperforms' claim.
  5. [Section 4.3, Table 1] GPD-VVTO is initialized from SD v2.1, while DPIDM uses SD v1.5. The paper acknowledges this but does not discuss the potential confounding effect on the headline 60.5% VFID improvement. If the key advantage is pose interaction, the comparison should ideally control for the base model or at least analyze whether the backbone difference could explain part of the gap. Please address this in the discussion.
minor comments (5)
  1. [Figure 2 caption] The caption contains a typo: 'emplys' should be 'employs'.
  2. [Equation 6] The summation notation is non-standard: 'NX i' and 'TX j=2' should be written with proper limits, e.g., \(\sum_{i=1}^{N}\) and \(\sum_{j=2}^{T}\).
  3. [Section 3.3] The paper states that the number of garment landmarks varies by garment type but does not specify the counts for lower-body garments or dresses; please provide these details or a reference to the annotation scheme.
  4. [Section 3.6] The inference sliding-window length T is not specified; please state the window size and overlap used in the reported experiments.
  5. [General] No statement about code or data availability is provided. Adding a reproducibility statement would be helpful for the community.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: empirical benchmark comparison with independently trained components; no prediction reduces to fitted inputs by construction.

full rationale

This paper is an empirical systems paper, not a derivation. The claimed improvements are measured by standard metrics (SSIM, LPIPS, VFID) against published baselines on public datasets. No fitted parameter is renamed as a prediction, no quantity is defined in terms of the target metric, and no load-bearing result is imported solely from the authors' prior work. The custom garment pose estimator is not evaluated for accuracy, and the ablation does not isolate the garment-pose branch from the human-pose branch; these are experimental validation gaps, not circular reasoning. The temporal regularized attention loss and pose-aware attention modules are trained with the standard diffusion objective, and their contribution is assessed by ablations, albeit not perfectly controlled. Nothing in the paper's equations or citations exhibits a self-definitional or fitted-input-equals-prediction loop, so the circularity score is 0.

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

The system leans on pre-trained Stable Diffusion priors, a manually supervised garment landmark estimator, and the assumption that attention-map consistency is a good proxy for temporal coherence. No free physical constants or invented entities are introduced. Hyperparameters lambda, gamma_i, and the keypoint-drop probability are hand-chosen and not sensitivity-tested.

free parameters (3)
  • TRA loss weight lambda = 1e-3 for video datasets, 0 for image datasets
    Chosen by hand to balance MSE and temporal regularized attention loss (Section 3.6); no sensitivity analysis is reported.
  • Attention map weighting gamma_i = 0.5
    Fixed weight applied to the last two decoder layers when computing the TRA loss (Section 3.5); no sensitivity analysis is reported.
  • Keypoint drop probability = 0.05
    Condition dropping probability during training to make the model robust to pose estimation errors (Section 3.6); chosen by hand without ablation.
assumptions (4)
  • domain assumption Pre-trained Stable Diffusion v1.5 weights remain a valid feature prior when pose adapters and pose-aware attention are inserted.
    Section 3.2 and 4.2 initialize both U-Nets from SD v1.5; the entire architecture assumes this prior transfers to video try-on with the added pose conditioning.
  • domain assumption Manually labeled garment landmarks correspond to human body landmarks accurately enough to guide attention.
    Section 3.3 trains a garment pose estimator on manual labels; no accuracy evaluation, dataset release, or annotation protocol is provided, yet all pose-interaction modules depend on this alignment.
  • domain assumption Self-attention maps encode structural content, so regularizing them across frames improves temporal consistency.
    Section 3.5 motivates the TRA loss using prior attention-map works [3,29,42]; the paper validates the loss only indirectly on VVT and never measures whether attention-map stability is the actual cause.
  • domain assumption Removing back-facing test segments from ViViD does not bias the comparison against baselines.
    Section 4.1 notes the exclusion due to inaccurate cloth-agnostic masks but does not report how many clips were removed or re-evaluate baselines on the remaining split.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Pursuing Temporal-Consistent Video Virtual Try-On via Dynamic Pose Interaction." pith.science (2026). https://pith.science/paper/TYCH7KCG

@misc{pith2026250516980,
  author       = {Pith},
  title        = {Pith review of: Pursuing Temporal-Consistent Video Virtual Try-On via Dynamic Pose Interaction},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/TYCH7KCG}},
  note         = {Machine review of arXiv:2505.16980}
}
read the original abstract

Video virtual try-on aims to seamlessly dress a subject in a video with a specific garment. The primary challenge involves preserving the visual authenticity of the garment while dynamically adapting to the pose and physique of the subject. While existing methods have predominantly focused on image-based virtual try-on, extending these techniques directly to videos often results in temporal inconsistencies. Most current video virtual try-on approaches alleviate this challenge by incorporating temporal modules, yet still overlook the critical spatiotemporal pose interactions between human and garment. Effective pose interactions in videos should not only consider spatial alignment between human and garment poses in each frame but also account for the temporal dynamics of human poses throughout the entire video. With such motivation, we propose a new framework, namely Dynamic Pose Interaction Diffusion Models (DPIDM), to leverage diffusion models to delve into dynamic pose interactions for video virtual try-on. Technically, DPIDM introduces a skeleton-based pose adapter to integrate synchronized human and garment poses into the denoising network. A hierarchical attention module is then exquisitely designed to model intra-frame human-garment pose interactions and long-term human pose dynamics across frames through pose-aware spatial and temporal attention mechanisms. Moreover, DPIDM capitalizes on a temporal regularized attention loss between consecutive frames to enhance temporal consistency. Extensive experiments conducted on VITON-HD, VVT and ViViD datasets demonstrate the superiority of our DPIDM against the baseline methods. Notably, DPIDM achieves VFID score of 0.506 on VVT dataset, leading to 60.5% improvement over the state-of-the-art GPD-VVTO approach.

Figures

Figures reproduced from arXiv: 2505.16980 by the authors.

Figure 1
Figure 1. Given a garment and a person video, DPIDM generates a [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. (a) Overall architecture of DPIDM. DPIDM emplys a dual-branch architecture. The main U-Net processes a concatenated input [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Visualization of predicted human and garment poses. [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Qualitative comparison on the ViViD dataset. Our [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 5
Figure 5. Figure 5: Qualitative comparison on the ViViD dataset. Our DPIDM maintains temporal consistency even during substantial movements. [PITH_FULL_IMAGE:figures/full_fig_p008_5.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

62 extracted references · 47 canonical work pages

  1. [1]

    Latent-shift: Latent diffu- sion with temporal shift for efficient text-to-video genera- tion.arXiv preprint arXiv:2304.08477, 2023

    Jie An, Songyang Zhang, Harry Yang, Sonal Gupta, Jia-Bin Huang, Jiebo Luo, and Xi Yin. Latent-shift: Latent diffu- sion with temporal shift for efficient text-to-video genera- tion.arXiv preprint arXiv:2304.08477, 2023. 5

    arXiv 2023

  2. [2]

    Demystifying mmd gans

    Mikołaj Bi ´nkowski, Danica J Sutherland, Michael Arbel, and Arthur Gretton. Demystifying mmd gans. InICLR, 2018. 6

    work page 2018

  3. [3]

    Masactrl: Tuning-free mu- tual self-attention control for consistent image synthesis and editing

    Mingdeng Cao, Xintao Wang, Zhongang Qi, Ying Shan, Xi- aohu Qie, and Yinqiang Zheng. Masactrl: Tuning-free mu- tual self-attention control for consistent image synthesis and editing. InCVPR, 2023. 5

    work page 2023

  4. [4]

    Quo vadis, action recognition? a new model and the kinetics dataset

    Joao Carreira and Andrew Zisserman. Quo vadis, action recognition? a new model and the kinetics dataset. InCVPR,

    work page

  5. [5]

    Wear-any-way: Manip- ulable 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: Manip- ulable virtual try-on via sparse correspondence alignment. In ECCV, 2024. 1, 2, 7

    work page 2024

  6. [6]

    Text-to-3d using gaussian splatting

    Zilong Chen, Feng Wang, Yikai Wang, and Huaping Liu. Text-to-3d using gaussian splatting. InCVPR, 2024. 1

    work page 2024

  7. [7]

    Viton-hd: High-resolution virtual try-on via misalignment-aware normalization

    Seunghwan Choi, Sunghyun Park, Minsoo Lee, and Jaegul Choo. Viton-hd: High-resolution virtual try-on via misalignment-aware normalization. InCVPR, 2021. 1, 2, 6, 7

    work page 2021

  8. [8]

    Improving diffusion models for au- thentic virtual try-on in the wild

    Yisol Choi, Sangkyung Kwak, Kyungmin Lee, Hyungwon Choi, and Jinwoo Shin. Improving diffusion models for au- thentic virtual try-on in the wild. InECCV, 2024. 1, 3, 7

    work page 2024

Show all 62 references
  1. [9]

    Diffusion models in vision: A survey

    Florinel-Alin Croitoru, Vlad Hondru, Radu Tudor Ionescu, and Mubarak Shah. Diffusion models in vision: A survey. IEEE Transactions on Pattern Analysis and Machine Intelli- gence, 45(9):10850–10869, 2023. 1

    2023

  2. [10]

    Fw-gan: Flow-navigated warping gan for video virtual try-on

    Haoye Dong, Xiaodan Liang, Xiaohui Shen, Bowen Wu, Bing-Cheng Chen, and Jian Yin. Fw-gan: Flow-navigated warping gan for video virtual try-on. InCVPR, 2019. 2, 6, 7

    2019

  3. [11]

    Fash- ion editing with adversarial parsing learning

    Haoye Dong, Xiaodan Liang, Yixuan Zhang, Xujie Zhang, Xiaohui Shen, Zhenyu Xie, Bowen Wu, and Jian Yin. Fash- ion editing with adversarial parsing learning. InCVPR, 2020. 2

    2020

  4. [12]

    Vivid: Video virtual try-on using diffusion models

    Zixun Fang, Wei Zhai, Aimin Su, Hongliang Song, Kai Zhu, Mao Wang, Yu Chen, Zhiheng Liu, Yang Cao, and Zheng- Jun Zha. Vivid: Video virtual try-on using diffusion models. arXiv preprint arXiv:2405.11794, 2024. 2, 3, 4, 5, 6, 7

    2024 arXiv

  5. [13]

    Guess: Gradually enriching synthesis for text-driven human motion generation.IEEE Transactions on Visualization and Computer Graphics, 30 (12):7518–7530, 2024

    Xuehao Gao, Yang Yang, Zhenyu Xie, Shaoyi Du, Zhongqian Sun, and Yang Wu. Guess: Gradually enriching synthesis for text-driven human motion generation.IEEE Transactions on Visualization and Computer Graphics, 30 (12):7518–7530, 2024. 1

    2024

  6. [14]

    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. InCVPR, 2021. 6, 7

    2021

  7. [15]

    Generative adversarial nets.Advances in neural information processing systems, 27, 2014

    Ian J Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron Courville, and Yoshua Bengio. Generative adversarial nets.Advances in neural information processing systems, 27, 2014. 2

    2014

  8. [16]

    Taming the power of diffusion models for high-quality virtual try-on with appearance flow

    Junhong Gou, Siyu Sun, Jianfu Zhang, Jianlou Si, Chen Qian, and Liqing Zhang. Taming the power of diffusion models for high-quality virtual try-on with appearance flow. InACM MM, 2023. 7

    2023

  9. [17]

    Animatediff: Animate your personalized text-to-image diffusion models without specific tuning

    Yuwei Guo, Ceyuan Yang, Anyi Rao, Yaohui Wang, Yu Qiao, Dahua Lin, and Bo Dai. Animatediff: Animate your personalized text-to-image diffusion models without specific tuning. InICLR, 2024. 5

    2024

  10. [18]

    Generatect: text- conditional generation of 3d chest ct volumes

    Ibrahim Ethem Hamamci, Sezgin Er, Anjany Sekuboy- ina, Enis Simsar, Alperen Tezcan, Ayse Gulnihan Sim- sek, Sevval Nil Esirgun, Furkan Almas, Irem Do ˘gan, Muhammed Furkan Dasdelen, et al. Generatect: text- conditional generation of 3d chest ct volumes. InECCV,

  11. [19]

    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 CVPR, 2018. 1, 2

    2018

  12. [20]

    Can spatiotemporal 3d cnns retrace the history of 2d cnns and imagenet? InCVPR, 2018

    Kensho Hara, Hirokatsu Kataoka, and Yutaka Satoh. Can spatiotemporal 3d cnns retrace the history of 2d cnns and imagenet? InCVPR, 2018. 6

    2018

  13. [21]

    Wildvidfit: Video virtual try- on in the wild via image-based controlled diffusion models

    Zijian He, Peixin Chen, Guangrun Wang, Guanbin Li, Philip HS Torr, and Liang Lin. Wildvidfit: Video virtual try- on in the wild via image-based controlled diffusion models. InECCV, 2024. 3

    2024

  14. [22]

    Gans trained by a two time-scale update rule converge to a local nash equilib- rium.Advances in neural information processing systems, 30, 2017

    Martin Heusel, Hubert Ramsauer, Thomas Unterthiner, Bernhard Nessler, and Sepp Hochreiter. Gans trained by a two time-scale update rule converge to a local nash equilib- rium.Advances in neural information processing systems, 30, 2017. 6

    2017

  15. [23]

    Denoising dif- fusion probabilistic models.Advances in neural information processing systems, 33:6840–6851, 2020

    Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising dif- fusion probabilistic models.Advances in neural information processing systems, 33:6840–6851, 2020. 4

    2020

  16. [24]

    Animate anyone: Consistent and controllable image- to-video synthesis for character animation

    Li Hu. Animate anyone: Consistent and controllable image- to-video synthesis for character animation. InCVPR, 2024. 5

    2024

  17. [25]

    Training- free content injection using h-space in diffusion models

    Jaeseok Jeong, Mingi Kwon, and Youngjung Uh. Training- free content injection using h-space in diffusion models. In WACV, 2024. 5

    2024

  18. [26]

    Cloth- former: Taming video virtual try-on in all module

    Jianbin Jiang, Tan Wang, He Yan, and Junhui Liu. Cloth- former: Taming video virtual try-on in all module. InCVPR,

  19. [27]

    Adam: A method for stochastic opti- mization.arXiv preprint arXiv:1412.6980, 2014

    Diederik P Kingma. Adam: A method for stochastic opti- mization.arXiv preprint arXiv:1412.6980, 2014. 6

    2014 arXiv

  20. [28]

    Shineon: Illuminating design choices for prac- tical video-based virtual clothing try-on

    Gaurav Kuppa, Andrew Jong, Xin Liu, Ziwei Liu, and Teng- Sheng Moh. Shineon: Illuminating design choices for prac- tical video-based virtual clothing try-on. InWACV, 2021. 2

    2021

  21. [29]

    Harivo: Harnessing text-to-image models for video generation

    Mingi Kwon, Seoung Wug Oh, Yang Zhou, Difan Liu, Joon-Young Lee, Haoran Cai, Baqiao Liu, Feng Liu, and Youngjung Uh. Harivo: Harnessing text-to-image models for video generation. InECCV, 2024. 5

    2024

  22. [30]

    High-resolution virtual try-on with 9 misalignment and occlusion-handled conditions

    Sangyun Lee, Gyojung Gu, Sunghyun Park, Seunghwan Choi, and Jaegul Choo. High-resolution virtual try-on with 9 misalignment and occlusion-handled conditions. InECCV,

  23. [31]

    Virtual try-on with pose-garment keypoints guided inpaint- ing

    Zhi Li, Pengfei Wei, Xiang Yin, Zejun Ma, and Alex C Kot. Virtual try-on with pose-garment keypoints guided inpaint- ing. InCVPR, 2023. 2

    2023

  24. [32]

    Tsm: Temporal shift module for efficient video understanding

    Ji Lin, Chuang Gan, and Song Han. Tsm: Temporal shift module for efficient video understanding. InCVPR, 2019. 5

    2019

  25. [33]

    Dress code: High- resolution multi-category virtual try-on

    Davide Morelli, Matteo Fincato, Marcella Cornia, Federico Landi, Fabio Cesari, and Rita Cucchiara. Dress code: High- resolution multi-category virtual try-on. InCVPR, 2022. 2

    2022

  26. [34]

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

    Davide Morelli, Alberto Baldrati, Giuseppe Cartella, Mar- cella Cornia, Marco Bertini, and Rita Cucchiara. Ladi-vton: Latent diffusion textual-inversion enhanced virtual try-on. In ACM MM, 2023. 6, 7

    2023

  27. [35]

    Deep generative design: integration of topology optimization and generative models.Journal of Mechanical Design, 141(11):111405, 2019

    Sangeun Oh, Yongsu Jung, Seongsin Kim, Ikjin Lee, and Namwoo Kang. Deep generative design: integration of topology optimization and generative models.Journal of Mechanical Design, 141(11):111405, 2019. 1

    2019

  28. [36]

    To create what you tell: Generating videos from cap- tions

    Yingwei Pan, Zhaofan Qiu, Ting Yao, Houqiang Li, and Tao Mei. To create what you tell: Generating videos from cap- tions. InACM MM, 2017. 1

    2017

  29. [37]

    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. InICML, 2021. 4

    2021

  30. [38]

    You only look once: Unified, real-time object detection

    J Redmon. You only look once: Unified, real-time object detection. InCVPR, 2016. 6

    2016

  31. [39]

    High-resolution image syn- thesis with latent diffusion models

    Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Bj¨orn Ommer. High-resolution image syn- thesis with latent diffusion models. InCVPR, 2022. 1, 3

    2022

  32. [40]

    U-net: Convolutional networks for biomedical image segmentation

    Olaf Ronneberger, Philipp Fischer, and Thomas Brox. U-net: Convolutional networks for biomedical image segmentation. InMICCAI, 2015. 4

    2015

  33. [41]

    Denoising diffusion implicit models.arXiv preprint arXiv:2010.02502, 2020

    Jiaming Song, Chenlin Meng, and Stefano Ermon. Denoising diffusion implicit models.arXiv preprint arXiv:2010.02502, 2020. 4, 6

    2010 arXiv

  34. [42]

    Plug-and-play diffusion features for text-driven image-to-image translation

    Narek Tumanyan, Michal Geyer, Shai Bagon, and Tali Dekel. Plug-and-play diffusion features for text-driven image-to-image translation. InCVPR, 2023. 5

    2023

  35. [43]

    To- wards accurate generative models of video: A new metric & challenges.arXiv preprint arXiv:1812.01717, 2018

    Thomas Unterthiner, Sjoerd Van Steenkiste, Karol Kurach, Raphael Marinier, Marcin Michalski, and Sylvain Gelly. To- wards accurate generative models of video: A new metric & challenges.arXiv preprint arXiv:1812.01717, 2018. 6

    2018 arXiv

  36. [44]

    Improving virtual try-on with garment-focused diffusion models

    Siqi Wan, Yehao Li, Jingwen Chen, Yingwei Pan, Ting Yao, Yang Cao, and Tao Mei. Improving virtual try-on with garment-focused diffusion models. InECCV, 2024. 2

    2024

  37. [45]

    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. InECCV,

  38. [46]

    Gpd-vvto: Preserving garment details in video virtual try-on

    Yuanbin Wang, Weilun Dai, Long Chan, Huanyu Zhou, Aixi Zhang, and Si Liu. Gpd-vvto: Preserving garment details in video virtual try-on. InACM MM, 2024. 2, 3, 4, 5, 6, 7

    2024

  39. [47]

    Image quality assessment: from error visibility to structural similarity.IEEE transactions on image processing, 13(4):600–612, 2004

    Zhou Wang, Alan C Bovik, Hamid R Sheikh, and Eero P Si- moncelli. Image quality assessment: from error visibility to structural similarity.IEEE transactions on image processing, 13(4):600–612, 2004. 6

    2004

  40. [48]

    Gp- vton: Towards general purpose virtual try-on via collabora- tive 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 collabora- tive local-flow global-parsing learning. InCVPR, 2023. 7

    2023

  41. [49]

    Simda: Simple diffusion adapter for efficient video generation

    Zhen Xing, Qi Dai, Han Hu, Zuxuan Wu, and Yu-Gang Jiang. Simda: Simple diffusion adapter for efficient video generation. InCVPR, 2024. 5

    2024

  42. [50]

    Oot- diffusion: Outfitting fusion based latent diffusion for control- lable virtual try-on.arXiv preprint arXiv:2403.01779, 2024

    Yuhao Xu, Tao Gu, Weifeng Chen, and Chengcai Chen. Oot- diffusion: Outfitting fusion based latent diffusion for control- lable virtual try-on.arXiv preprint arXiv:2403.01779, 2024. 1

    2024 arXiv

  43. [51]

    Tunnel try-on: Excavating spatial-temporal tunnels for high-quality virtual try-on in videos

    Zhengze Xu, Mengting Chen, Zhao Wang, Linyu Xing, Zhonghua Zhai, Nong Sang, Jinsong Lan, Shuai Xiao, and Changxin Gao. Tunnel try-on: Excavating spatial-temporal tunnels for high-quality virtual try-on in videos. InACM MM, 2024. 2, 3, 4, 5, 6, 7

    2024

  44. [52]

    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. InCVPR, 2024. 2

    2024

  45. [53]

    Effec- tive whole-body pose estimation with two-stages distillation

    Zhendong Yang, Ailing Zeng, Chun Yuan, and Yu Li. Effec- tive whole-body pose estimation with two-stages distillation. InCVPR, 2023. 4

    2023

  46. [54]

    Cat-dm: Controllable acceler- ated virtual try-on with diffusion model

    Jianhao Zeng, Dan Song, Weizhi Nie, Hongshuo Tian, Tong- tong Wang, and An-An Liu. Cat-dm: Controllable acceler- ated virtual try-on with diffusion model. InCVPR, 2024. 2, 7

    2024

  47. [55]

    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. InCVPR, 2018. 6

    2018

  48. [56]

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

    Xuanpu Zhang, Dan Song, Pengxin Zhan, Qingguo Chen, Kuilong Liu, and Anan Liu. Stableviton: Learning semantic correspondence with latent diffusion model for virtual try- on. InCVPR, 2023. 1, 2, 6, 7

    2023

  49. [57]

    Mmtryon: Multi-modal multi-reference control for high-quality fashion generation.arXiv preprint arXiv:2405.00448, 2024

    Xujie Zhang, Ente Lin, Xiu Li, Yuxuan Luo, Michael Kampffmeyer, Xin Dong, and Xiaodan Liang. Mmtryon: Multi-modal multi-reference control for high-quality fashion generation.arXiv preprint arXiv:2405.00448, 2024. 2

    2024 arXiv

  50. [58]

    Towards consistent video edit- ing with text-to-image diffusion models.Advances in Neural Information Processing Systems, 36:58508–58519, 2023

    Zicheng Zhang, Bonan Li, Xuecheng Nie, Congying Han, Tiande Guo, and Luoqi Liu. Towards consistent video edit- ing with text-to-image diffusion models.Advances in Neural Information Processing Systems, 36:58508–58519, 2023. 5

    2023

  51. [59]

    Trip: Temporal residual learning with image noise prior for image-to-video diffusion models

    Zhongwei Zhang, Fuchen Long, Yingwei Pan, Zhaofan Qiu, Ting Yao, Yang Cao, and Tao Mei. Trip: Temporal residual learning with image noise prior for image-to-video diffusion models. InCVPR, 2024. 1

    2024

  52. [60]

    Viton-dit: Learning in-the-wild video try-on from human dance videos via diffusion transformers.arXiv preprint arXiv:2405.18326, 2024

    Jun Zheng, Fuwei Zhao, Youjiang Xu, Xin Dong, and Xi- aodan Liang. Viton-dit: Learning in-the-wild video try-on from human dance videos via diffusion transformers.arXiv preprint arXiv:2405.18326, 2024. 3, 6, 7

    2024 arXiv

  53. [61]

    Mv-ton: Memory-based video virtual try- on network

    Xiaojing Zhong, Zhonghua Wu, Taizhe Tan, Guosheng Lin, and Qingyao Wu. Mv-ton: Memory-based video virtual try- on network. InACM MM, 2021. 2

    2021

  54. [62]

    Tryondiffusion: A tale of two un- ets

    Luyang Zhu, Dawei Yang, Tyler Zhu, Fitsum Reda, William Chan, Chitwan Saharia, Mohammad Norouzi, and Ira Kemelmacher-Shlizerman. Tryondiffusion: A tale of two un- ets. InCVPR, 2023. 1, 2 10

    2023

Pith tools

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