Pith. sign in

REVIEW 4 major objections 6 minor 56 references

ChronoTailor: Harnessing Attention Guidance for Fine-Grained Video Virtual Try-On

T0 review · 4 major / 6 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 ChronoTailor claims that attention-guided diffusion with segmentation-based spatial guidance and pose-aligned multi-scale garment features yields temporally consistent video try-ons that preserve garment details and outperform previous…

desk verdict A useful dataset and a plausible video try-on pipeline, but the load-bearing loss in Eq. (1) is mis-specified as written and the evaluation does not control training data, so the SOTA claim is not yet supported. read the letter →

arxiv 2506.05858 v1 pith:PS5ZZALH submitted 2025-06-06 cs.CV

Jinjuan Wang , Wenzhang Sun , Ming Li , Yun Zheng , Fanyao Li , Zhulin Tao , Donglin Di , Hao Li
show 2 more authors
Wei Chen Xianglin Huang
This is my paper
classification cs.CV
keywords videovirtualtry-ondiffusionmodelattentionguidancetemporalconsistencygarmentdetailpreservationspatio-temporalStyleDressdatasetmulti-scalefeatureextraction
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

ChronoTailor is a diffusion-based video virtual try-on framework that replaces a person’s clothing in video while holding the rest of the frame intact. The paper argues that stable video try-on needs explicit steering of attention: segmentation masks push spatial attention toward the garment region, an asymmetric cross-attention layer fuses randomly sampled frames for temporal continuity, and multi-scale garment features aligned with pose projections preserve textures during motion. A reader should care because current video try-on methods tend to flicker, blur, or tear garment patterns when the body moves; ChronoTailor reports that its design reduces those artifacts and quantitatively outperforms previously published methods on the VVT and ViViD benchmarks. The paper also contributes StyleDress, a dataset of 14,258 images and 12,500 videos with diverse garments, poses, and backgrounds.

What carries the argument

The machinery is a pair of attention-guiding modules. Spatial-Temporal Attention Guidance contains Region-Aware Spatial Guidance (RASG), a loss $\mathcal{L}_{\text{RASG}} = \sum_{i\in N} \sum_{a\in A} \|M - A_i^{(a)}\|_2^2 + \lambda_N \sum_{i\in N} \sum_{a\in \bar A} \|1 - A_i^{(a)}\|_2^2$ that pushes attention maps toward the segmentation mask in target areas, and Attention-Driven Temporal Feature Fusion (ATFF), an asymmetric cross-attention layer using current-frame features as queries and randomly fused features from other frames as keys/values. The second pillar is Multi-scale Garment-Pose Feature Alignment: Adaptive Multi-scale Feature Extraction (AMFE) processes the VAE encoder’s last three downsampling feature maps through pixel-shuffle alignment, $1\times1$, $3\times3$, and $7\times7$ convolutions, normalization, and SiLU activation, then fuses them with learnable weights $\vec\alpha$; Garment-Pose Feature Alignment (GPFA) shares cross-attention weight matrices between reference and denoising U-Nets so that CLIP-encoded garment features and pose features are aligned. Together these direct where garment information enters the denoising process and how it moves across frames.

What would settle it

Inspect the training code’s RASG implementation and the attention maps it produces: if the code minimizes $\sum_{a \in \bar A} \|1 - A_i^{(a)}\|_2^2$ over non-target pixels, then attention is being driven to 1 in those regions, contradicting the stated claim that the term regularizes non-target attention; flipping the sign in an ablation would also reveal whether the gains attributed to RASG survive.

Watch

Extended reading notes

Core claim

On its own terms, ChronoTailor’s central discovery is that attention guidance, rather than explicit warping, is the right way to inject garment information into a video diffusion model. The framework keeps the standard latent diffusion denoiser, adds a reference U-Net for garment features, and trains two added mechanisms: Region-Aware Spatial Guidance, which uses segmentation masks to align spatial attention with the edit region, and Attention-Driven Temporal Feature Fusion, which uses current-frame queries against randomly fused other-frame keys/values to propagate motion. Multi-scale garment features from the VAE’s last downsampling layers are fused with learnable weights, and cross-attention weights are shared between reference and denoising U-Nets to align garment features with DensePose and CLIP pose features. On VVT the paper reports SSIM 0.978, LPIPS 0.049, and VFIDRes 4.065, which are the best among the compared methods on those metrics, and a user study on ViViD gives ChronoTailor majority preference on garment fidelity, background preservation, temporal consistency, and overall quality. The abstract summarizes the result as significantly outperforming previous methods.

Load-bearing premise

The load-bearing premise is that the attention-regularization loss in Eq. (1) works as described; the printed second term $\lambda_N \sum_{i\in N} \sum_{a\in \bar A} \|1 - A_i^{(a)}\|_2^2$ drives attention toward 1 in non-target regions, which is the opposite of the paper’s stated purpose, and no implementation note explains why the sign is not reversed.

Editorial extensions

If this is right

  • If the reported VVT numbers hold, ChronoTailor achieves the highest SSIM and lowest LPIPS among the methods compared, with SSIM 0.978 and LPIPS 0.049.
  • StyleDress would give the community a training set with balanced gender, varied body shapes, diverse poses, and indoor/outdoor scenes, covering categories that VVT and ViViD lack.
  • The ablation tables attribute gains to each module separately, so RASG, ATFF, AMFE, and GPFA can be adopted independently in other latent-diffusion try-on systems.
  • RASG is applied only during training, meaning the reported quality gains do not add inference-time compute per video frame.
  • The weight-sharing cross-attention between garment and pose features provides a mechanism for keeping garment texture aligned with body motion, which the paper identifies as a failure mode of flow-based and memory-based prior work.

Reading between the lines

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

  • Taken literally, the second term in Eq. (1) minimizes $\|1 - A_i^{(a)}\|_2^2$ over non-target locations, which would drive attention toward 1 in those regions; if the released implementation does not flip the sign, the reported RASG gains cannot be explained by the paper’s stated mechanism.
  • Because RASG is applied only during training, the same training-time attention-regularization recipe could transfer to other diffusion-based video editing tasks that need localized edits, though the paper does not test that.
  • The paper’s stated limitation that segmentation and pose estimation degrade under occlusion, low resolution, or high-dynamic motion suggests a concrete stress test: evaluate ChronoTailor on clips with noisy masks or DensePose outputs and compare against the clean-condition results.
  • StyleDress is announced but not yet released, so the claim that it surpasses existing datasets in diversity and quality can only be fully verified once the dataset is public and independently used.
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

4 major / 6 minor

Summary. ChronoTailor is a diffusion-based framework for video virtual try-on. It introduces two main components: Spatial-Temporal Attention Guidance, consisting of Region-Aware Spatial Guidance (RASG) and Attention-Driven Temporal Feature Fusion (ATFF), and Multi-scale Garment-Pose Feature Alignment, consisting of Adaptive Multi-scale Feature Extraction (AMFE) and Garment-Pose Feature Alignment (GPFA). The paper also introduces a new dataset, StyleDress, and evaluates the method on VVT and ViViD, reporting improved SSIM and LPIPS, competitive VFID scores, and a user study favoring ChronoTailor. The central claim is that the proposed attention-guidance mechanisms yield state-of-the-art fine-grained video virtual try-on with preserved garment details and temporal consistency.

Significance. If the architecture and loss are implemented as intended, the paper would make a meaningful contribution to video virtual try-on by addressing temporal continuity and fine-grained detail preservation without explicit warping. The modular design, the detailed ablation study, and the introduction of the StyleDress dataset are genuine strengths, as is the extensive qualitative comparison in the appendix. However, the significance is currently qualified by two concerns: the central loss in Eq. (1) appears to be mis-specified, and the quantitative comparisons do not control for training distribution or resolution. No code or checkpoints are released, and the loss hyperparameters are not reported, so the reported results cannot be independently reproduced. The paper is a useful step forward if these issues can be resolved, but the current manuscript does not fully support the headline claim of significantly outperforming prior methods across all reported metrics.

major comments (4)
  1. [Section 3.1.1, Eq. (1)] The second term of Eq. (1), λ_N Σ_{i∈N} Σ_{a∈Ā} ∥1−A_i^{(a)}∥_2^2, is inconsistent with the stated purpose of regularizing attention in non-target regions. Minimizing ∥1−A∥_2^2 drives A_i^{(a)} toward 1 for every non-target attention location, which would concentrate attention on the regions the method is supposed to de-emphasize. The stated goal would instead require a term such as ∥A∥_2^2, pushing non-target attention toward 0. This is not a cosmetic issue: Table 4 credits RASG with large improvements (SSIM 0.850→0.894, VFIDRes 0.630→0.535). If Eq. (1) is a typo, the correct formulation is absent from the manuscript; if it is literal, the reported ablation gains are not explained by the paper's own mathematics. The values of λ_N and λ_R are also not reported, so the loss cannot be reproduced as written.
  2. [Section 5, Table 2 and Appendix B.1.2] The comparison to prior work is not controlled for training distribution or resolution. Table 2 relies on numbers reported in other papers, while Appendix B.1.2 states that on StyleDress 'our approach demonstrates superior performance compared to methods trained on the VVT dataset.' The main experiments train ChronoTailor at 512×384, whereas VVT was originally 256×192 and ViViD 832×624. Under these conditions, the claim of 'significantly outperforming previous methods' may reflect training-set or resolution advantages rather than the proposed mechanisms. Please re-evaluate all baselines under a single training and evaluation protocol, or clearly restrict the claim to the settings used and disclose the mismatch.
  3. [Section 5.2, Table 2 and abstract] The abstract's claim of 'significantly outperforming previous methods' is not fully supported by Table 2. On VFID3D, ChronoTailor (3.721) is worse than Tunnel Try-On (3.345) and ViViD (3.405). The text in Section 5.2 does acknowledge 'competitive performance' on VFID3D, but the broader claim in the abstract and contribution list should be tempered or the metric-specific results should be reported more precisely.
  4. [Section 5.2 and Appendix B.3] The user study reporting is inconsistent. The main text reports responses from 40 users and 13 video results, while Appendix B.3 says 'A total of 50 evaluation samples are collected' with no clarification of whether these are 50 participants or 50 comparisons. No significance tests, confidence intervals, or per-participant variance are reported. The preference rates in Table 3 are presented as point estimates, so the robustness of the claimed user preference is unclear. Please reconcile the numbers and report full statistical details.
minor comments (6)
  1. [Section 3.1.2, Eq. (3)] The notation in Eq. (3) is unclear: 'd=-1' is described as a dimensional index for aggregation along the last dimension, but the displayed expression appears to be a summation over a single value. Please clarify the summation range and the meaning of the operator.
  2. [Section 3.1.2, Eq. (4)] The role of the weights w_j and w_k in Eq. (4) should be clarified. If they are scalar attention scores, the form z = GroupNorm(w_j·z_j + z_i + w_k·z_k) is unusual because no attention sum of all selected features is shown; if they are tensors, their shapes should be specified.
  3. [Section 2.2 and References [46, 47]] The text attributes 'Tunnel TryOn' to reference [47], but reference [47] is MagicAnimate; Tunnel Try-On is reference [46]. Please correct the citation.
  4. [Appendix A.2] The name 'AnimeDiff' appears to be a typo for 'AnimateDiff' (reference [17]). Please fix the spelling and ensure the cited work is the correct motion-module source.
  5. [Tables 2 and 4] The metric is inconsistently labeled 'VFIDResNeXt' in Table 2 and 'VFIDRES' in Table 4. Please use one consistent name and define the metric explicitly.
  6. [Section 4 and Table 1] The dataset statistics in the introduction (14,258 images and 12,500 videos) and Table 1 (26,758 I-V) should be reconciled, since the table appears to count image-video pairs rather than separate images and videos.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: ChronoTailor's claims are empirically evaluated against public benchmarks and do not reduce to fitted inputs or self-citations.

full rationale

The paper makes no formal derivation that reduces to its own inputs. The central claim is that the proposed modules—RASG, ATFF, AMFE, GPFA—improve video virtual try-on, supported by quantitative comparisons on VVT/ViViD and ablations. RASG loss Eq. (1) is a training objective, not a fitted prediction; ATFF, AMFE, and GPFA are architectural components trained with standard diffusion loss. The paper cites two prior works by overlapping authors (Moee, Uniavatar) only as general AIGC background, not as load-bearing support. The introduced StyleDress dataset is a contribution; evaluating the method on it and on public benchmarks does not make the evaluation tautological. The suspicious sign in Eq. (1), where minimizing ||1-A||^2 pushes non-target attention toward 1, opposite to the stated regularization, is a possible correctness or typo issue that undermines confidence in the RASG ablation, but it is not a circularity: the claim does not reduce to the equation by construction. No fitted parameter is renamed as a prediction, no uniqueness theorem is imported, and no ansatz is smuggled in solely via citation. Hence circularity score 0.

Assumptions & free parameters 3 free parameters · 5 assumptions · 0 invented entities

No new physical or conceptual entities are introduced. The method relies on standard generative priors, segmentation and pose extractors, and a random-frame fusion heuristic. The main free parameters are the loss weights (unreported) and the learned multi-scale fusion weights.

free parameters (3)
  • lambda_N (negative attention weight)
    Appears in Eq. (1) to balance positive and negative attention regularization. Value not reported.
  • lambda_R (RASG loss weight)
    Appears in Eq. (2) to weight L_RASG against L_LDM. Value not reported.
  • alpha = [alpha_1, alpha_2, alpha_3] (AMFE fusion weights) = learned, no fixed values
    Learnable multi-scale fusion weights in Eq. (7); trained with the model.
assumptions (5)
  • standard math Diffusion model (LDM/SD 1.5) provides a learnable generative prior
    The method builds on stable diffusion latent diffusion; the framework is assumed as background.
  • domain assumption Cross-attention maps can be steered by a loss on attention probabilities
    Used in RASG; known from Prompt-to-Prompt but still an assumption about how attention controls generation.
  • domain assumption Segmentation masks (SAM2) and DensePose pose are accurate enough for training
    The method relies on these external predictions; paper lists degradation under occlusion as limitation.
  • domain assumption CLIP image encoder captures global pose and garment semantics
    Used for f_p and f_g in GPFA.
  • ad hoc to paper Randomly fusing frame features provides useful temporal context
    ATFF's mechanism; no theoretical justification is given.

how reviews work

0 comments
Cite this review

Pith. "Pith review of ChronoTailor: Harnessing Attention Guidance for Fine-Grained Video Virtual Try-On." pith.science (2026). https://pith.science/paper/PS5ZZALH

@misc{pith2026250605858,
  author       = {Pith},
  title        = {Pith review of: ChronoTailor: Harnessing Attention Guidance for Fine-Grained Video Virtual Try-On},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/PS5ZZALH}},
  note         = {Machine review of arXiv:2506.05858}
}
read the original abstract

Video virtual try-on aims to seamlessly replace the clothing of a person in a source video with a target garment. Despite significant progress in this field, existing approaches still struggle to maintain continuity and reproduce garment details. In this paper, we introduce ChronoTailor, a diffusion-based framework that generates temporally consistent videos while preserving fine-grained garment details. By employing a precise spatio-temporal attention mechanism to guide the integration of fine-grained garment features, ChronoTailor achieves robust try-on performance. First, ChronoTailor leverages region-aware spatial guidance to steer the evolution of spatial attention and employs an attention-driven temporal feature fusion mechanism to generate more continuous temporal features. This dual approach not only enables fine-grained local editing but also effectively mitigates artifacts arising from video dynamics. Second, ChronoTailor integrates multi-scale garment features to preserve low-level visual details and incorporates a garment-pose feature alignment to ensure temporal continuity during dynamic motion. Additionally, we collect StyleDress, a new dataset featuring intricate garments, varied environments, and diverse poses, offering advantages over existing public datasets, and will be publicly available for research. Extensive experiments show that ChronoTailor maintains spatio-temporal continuity and preserves garment details during motion, significantly outperforming previous methods.

Figures

Figures reproduced from arXiv: 2506.05858 by the authors.

Figure 1
Figure 1. ChronoTailor excels at generating high-quality virtual try-ons for diverse garments, effec [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Architecture of the ChronoTailor. Spatial-Temporal Attention Guidance enables the acquisition of stable guidance for garment feature injection. Multi-scale Garment-Pose Feature Alignment, meanwhile, facilitates the capture of more precise garment information during motion. 3.1.1 Region-Aware Spatial Guidance Conventional cross-attention mechanisms typically lack explicit semantic guidance, resulting in diffuse atten… view at source ↗
Figure 3
Figure 3. Sample pairs from the StyleDress [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (14 more)
Figure 4
Figure 4. Figure 4: Qualitative comparison on the VVT dataset. The dashed box highlights an area with barely [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 5
Figure 5. Figure 5: Specifically, our method demonstrates enhanced preservation of garment texture details [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 6
Figure 6. Figure 6: ChronoTailor effectively preserves information in non - edited regions. Implementation Details. The experiments are conducted on two NVIDIA A800 GPUs. During the training process, all data is adjusted to a uniform resolution of 512×384. We set the batch size to 8 and t…
Figure 7
Figure 7. Figure 7: Qualitative results of our method on the StyleDress dataset. remarkable robustness, which can be attributed to our novel Spatial-Temporal Attention Guidance and Multi-scale Garment-Pose Feature Alignment mechanisms. In contrast to the ViViD method, our approach exhibit…
Figure 8
Figure 8. Figure 8: Ablation study of Region-Aware Spa￾tial Guidance. Our method balances attention between editing and non - editing regions. Input w/o ATFF w/ ATFF [PITH_FULL_IMAGE:figures/full_fig_p009_8.png]
Figure 10
Figure 10. Figure 10: Ablation study of Adaptive Multi-scale Feature Extraction. The upper garment try-on fully recovers two floral patterns, demonstrating ChronoTailor’s capability to preserve intricate tex￾tures and semantic details [PITH_FULL_IMAGE:figures/full_fig_p009_10.png]
Figure 12
Figure 12. Figure 12: Qualitative comparison on the ViViD dataset. Our method can generate more color [PITH_FULL_IMAGE:figures/full_fig_p016_12.png]
Figure 13
Figure 13. Figure 13: Qualitative comparison on the VVT dataset. Our method generates clearer texture details [PITH_FULL_IMAGE:figures/full_fig_p017_13.png]
Figure 14
Figure 14. Figure 14: Additional video try-on results on the StyleDress dataset are shown. Left column image [PITH_FULL_IMAGE:figures/full_fig_p018_14.png]
Figure 15
Figure 15. Figure 15: Qualitative comparison on the VITON-HD dataset. [PITH_FULL_IMAGE:figures/full_fig_p019_15.png]
Figure 16
Figure 16. Figure 16: Our StyleDress dataset offers additional examples with diverse clothing, challenging poses [PITH_FULL_IMAGE:figures/full_fig_p019_16.png]
Figure 17
Figure 17. Figure 17: ChronoTailor best preserves the texture details and appearance information of garments [PITH_FULL_IMAGE:figures/full_fig_p020_17.png]
Figure 18
Figure 18. Figure 18: Ablation study of Spatial-Temporal Attention Guidance. Our method not only achieves [PITH_FULL_IMAGE:figures/full_fig_p020_18.png]
Figure 19
Figure 19. Figure 19: Ablation study of Multi-scale Garment-Pose Feature Alignment. ChronoTailor achieves [PITH_FULL_IMAGE:figures/full_fig_p021_19.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

56 extracted references · 35 canonical work pages

  1. [1]

    Multimodal garment designer: Human-centric latent diffusion models for fashion image editing

    Alberto Baldrati, Davide Morelli, Giuseppe Cartella, Marcella Cornia, Marco Bertini, and Rita Cucchiara. Multimodal garment designer: Human-centric latent diffusion models for fashion image editing. In Proceedings of the IEEE/CVF international conference on computer vision, pages 23393–23402, 2023

    work page 2023

  2. [2]

    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. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 6299–6308, 2017

    work page 2017

  3. [3]

    Fashionmirror: Co-attention feature-remapping virtual try-on with sequential template poses

    Chieh-Yun Chen, Ling Lo, Pin-Jui Huang, Hong-Han Shuai, and Wen-Huang Cheng. Fashionmirror: Co-attention feature-remapping virtual try-on with sequential template poses. InProceedings of the IEEE/CVF International Conference on Computer Vision, pages 13809–13818, 2021

    work page 2021

  4. [4]

    Anydoor: Zero-shot object-level image customization

    Xi Chen, Lianghua Huang, Yu Liu, Yujun Shen, Deli Zhao, and Hengshuang Zhao. Anydoor: Zero-shot object-level image customization. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 6593–6602, 2024

    2024

  5. [5]

    Livephoto: Real image animation with text-guided motion control

    Xi Chen, Zhiheng Liu, Mengting Chen, Yutong Feng, Yu Liu, Yujun Shen, and Hengshuang Zhao. Livephoto: Real image animation with text-guided motion control. InEuropean Conference on Computer Vision, pages 475–491. Springer, 2024

    work page 2024

  6. [6]

    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. InProceedings of IEEE/CVF Conference on Computer Vision and Pattern Recognition, Jun 2021

    work page 2021

  7. [7]

    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. Mar 2024

    work page 2024

  8. [8]

    Catvton: Concatenation is all you need for virtual try-on with diffusion models.arXiv preprint arXiv:2407.15886, 2024

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

    arXiv 2024

Show all 56 references
  1. [9]

    Dressing in order: Recurrent person image generation for pose transfer, virtual try-on and outfit editing

    Aiyu Cui, Daniel McKee, and Svetlana Lazebnik. Dressing in order: Recurrent person image generation for pose transfer, virtual try-on and outfit editing. InProceedings of the IEEE/CVF international conference on computer vision, pages 14638–14647, 2021

    2021

  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. In2019 IEEE/CVF International Conference on Computer Vision (ICCV), Oct 2019

    2019

  3. [11]

    Structure and content-guided video synthesis with diffusion models

    Patrick Esser, Johnathan Chiu, Parmida Atighehchian, Jonathan Granskog, and Anastasis Germanidis. Structure and content-guided video synthesis with diffusion models. InProceedings of the IEEE/CVF international conference on computer vision, pages 7346–7356, 2023

    2023

  4. [12]

    Vivid: Video virtual try-on using diffusion models.arXiv preprint arXiv:2405.11794, 2024

    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

    2024 arXiv

  5. [13]

    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. InProceedings of IEEE/CVF Conference on Computer Vision and Pattern Recognition, Jun 2021

    2021

  6. [14]

    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

    2014

  7. [15]

    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. InProceedings of the 31st ACM International Conference on Multimedia, pages 7599–7607, 2023

    2023

  8. [16]

    Densepose: Dense human pose estimation in the wild

    Rıza Alp Güler, Natalia Neverova, and Iasonas Kokkinos. Densepose: Dense human pose estimation in the wild. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 7297–7306, 2018

    2018

  9. [17]

    Animatediff: Animate your personalized text-to-image diffusion models without specific tuning.arXiv preprint arXiv:2307.04725, 2023

    Yuwei Guo, Ceyuan Yang, Anyi Rao, Zhengyang Liang, Yaohui Wang, Yu Qiao, Maneesh Agrawala, Dahua Lin, and Bo Dai. Animatediff: Animate your personalized text-to-image diffusion models without specific tuning.arXiv preprint arXiv:2307.04725, 2023. 10

    2023 arXiv

  10. [18]

    Xintong Han, Zuxuan Wu, Zhe Wu, Ruichi Yu, and Larry S. Davis. Viton: An image-based virtual try-on network. InProceedings of IEEE/CVF Conference on Computer Vision and Pattern Recognition, Jun 2018

    2018

  11. [19]

    Efficient diffusion training via min-snr weighting strategy

    Tiankai Hang, Shuyang Gu, Chen Li, Jianmin Bao, Dong Chen, Han Hu, Xin Geng, and Baining Guo. Efficient diffusion training via min-snr weighting strategy. InProceedings of the IEEE/CVF international conference on computer vision, pages 7441–7451, 2023

    2023

  12. [20]

    Can spatiotemporal 3d cnns retrace the history of 2d cnns and imagenet? InProceedings of the IEEE/CVF conference on Computer Vision and Pattern Recognition, pages 6546–6555, 2018

    Kensho Hara, Hirokatsu Kataoka, and Yutaka Satoh. Can spatiotemporal 3d cnns retrace the history of 2d cnns and imagenet? InProceedings of the IEEE/CVF conference on Computer Vision and Pattern Recognition, pages 6546–6555, 2018

    2018

  13. [21]

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

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

    2020

  14. [22]

    Viton-gan: Virtual try-on image generator trained with adversarial loss.arXiv preprint arXiv:1911.07926, 2019

    Shion Honda. Viton-gan: Virtual try-on image generator trained with adversarial loss.arXiv preprint arXiv:1911.07926, 2019

    1911 arXiv

  15. [23]

    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. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 8153–8163, 2024

    2024

  16. [24]

    Make it move: controllable image-to-video generation with text descriptions

    Yaosi Hu, Chong Luo, and Zhenzhong Chen. Make it move: controllable image-to-video generation with text descriptions. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 18219–18228, 2022

    2022

  17. [25]

    Fitdit: Advancing the authentic garment details for high-fidelity virtual try-on.arXiv preprint arXiv:2411.10499, 2024

    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

  18. [26]

    Clothformer: Taming video virtual try-on in all module

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

  19. [27]

    Dreampose: Fashion image-to-video synthesis via stable diffusion

    Johanna Karras, Aleksander Holynski, Ting-Chun Wang, and Ira Kemelmacher-Shlizerman. Dreampose: Fashion image-to-video synthesis via stable diffusion. In2023 IEEE/CVF International Conference on Computer Vision (ICCV), pages 22623–22633. IEEE, 2023

    2023

  20. [28]

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

    Jeongho Kim, Gyojung Gu, Minho Park, Sunghyun Park, and Jaegul Choo. Stableviton: Learning semantic correspondence with latent diffusion model for virtual try-on. Dec 2023

    2023

  21. [29]

    Shineon: Illuminating de- sign choices for practical video-based virtual clothing try-on

    Gaurav Kuppa, Andrew Jong, Xin Liu, Ziwei Liu, and Teng-Sheng Moh. Shineon: Illuminating de- sign choices for practical video-based virtual clothing try-on. InProceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, pages 191–200, 2021

    2021

  22. [30]

    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. InEuropean Conference on Computer Vision, pages 204–219. Springer, 2022

    2022

  23. [31]

    Blip-2: Bootstrapping language-image pre-training with frozen image encoders and large language models

    Junnan Li, Dongxu Li, Silvio Savarese, and Steven Hoi. Blip-2: Bootstrapping language-image pre-training with frozen image encoders and large language models. InInternational conference on machine learning, pages 19730–19742. PMLR, 2023

    2023

  24. [32]

    Diffcloth: Differentiable cloth simulation with dry frictional contact.ACM Transactions on Graphics (TOG), 42(1):1–20, 2022

    Yifei Li, Tao Du, Kui Wu, Jie Xu, and Wojciech Matusik. Diffcloth: Differentiable cloth simulation with dry frictional contact.ACM Transactions on Graphics (TOG), 42(1):1–20, 2022

    2022

  25. [33]

    Moee: Mixture of emotion experts for audio-driven portrait animation.arXiv preprint arXiv:2501.01808, 2025

    Huaize Liu, Wenzhang Sun, Donglin Di, Shibo Sun, Jiahui Yang, Changqing Zou, and Hujun Bao. Moee: Mixture of emotion experts for audio-driven portrait animation.arXiv preprint arXiv:2501.01808, 2025

    2025 arXiv

  26. [34]

    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. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 2231–2235, 2022

    2022

  27. [35]

    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. InProceedings of the 31st ACM international conference on multimedia, pages 8580–8589, 2023

    2023

  28. [36]

    Swifttry: Fast and consistent video virtual try-on with diffusion models.arXiv preprint arXiv:2412.10178, 2024

    Hung Nguyen, Quang Qui-Vinh Nguyen, Khoi Nguyen, and Rang Nguyen. Swifttry: Fast and consistent video virtual try-on with diffusion models.arXiv preprint arXiv:2412.10178, 2024. 11

    2024 arXiv

  29. [37]

    Sam 2: Segment anything in images and videos.arXiv preprint arXiv:2408.00714, 2024

    Nikhila Ravi, Valentin Gabeur, Yuan-Ting Hu, Ronghang Hu, Chaitanya Ryali, Tengyu Ma, Haitham Khedr, Roman Rädle, Chloe Rolland, Laura Gustafson, et al. Sam 2: Segment anything in images and videos.arXiv preprint arXiv:2408.00714, 2024

    2024 arXiv

  30. [38]

    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. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 10684–10695, 2022

    2022

  31. [39]

    U-net: Convolutional networks for biomedical image segmentation

    Olaf Ronneberger, Philipp Fischer, and Thomas Brox. U-net: Convolutional networks for biomedical image segmentation. InMedical image computing and computer-assisted intervention–MICCAI 2015: 18th international conference, Munich, Germany, October 5-9, 2015, proceedings, part I...

    2015

  32. [40]

    Uniavatar: Taming lifelike audio-driven talking head generation with comprehensive motion and lighting control.arXiv preprint arXiv:2412.19860, 2024

    Wenzhang Sun, Xiang Li, Donglin Di, Zhuding Liang, Qiyuan Zhang, Hao Li, Wei Chen, and Jianxun Cui. Uniavatar: Taming lifelike audio-driven talking head generation with comprehensive motion and lighting control.arXiv preprint arXiv:2412.19860, 2024

    2024 arXiv

  33. [41]

    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. InProceedings of the European conference on computer vision (ECCV), pages 589–604, 2018

    2018

  34. [42]

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

    2004

  35. [43]

    3dv-ton: Textured 3d-guided consistent video try-on via diffusion models.arXiv preprint arXiv:2504.17414, 2025

    Min Wei, Chaohui Yu, Jingkai Zhou, and Fan Wang. 3dv-ton: Textured 3d-guided consistent video try-on via diffusion models.arXiv preprint arXiv:2504.17414, 2025

    2025 arXiv

  36. [44]

    Sam2-unet: Segment anything 2 makes strong encoder for natural and medical image segmentation.arXiv preprint arXiv:2408.08870, 2024

    Xinyu Xiong, Zihuang Wu, Shuangyi Tan, Wenxue Li, Feilong Tang, Ying Chen, Siying Li, Jie Ma, and Guanbin Li. Sam2-unet: Segment anything 2 makes strong encoder for natural and medical image segmentation.arXiv preprint arXiv:2408.08870, 2024

    2024

  37. [45]

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

    Yuhao Xu, Tao Gu, Weifeng Chen, and Arlene Chen. Ootdiffusion: Outfitting fusion based latent diffusion for controllable virtual try-on. InProceedings of the AAAI Conference on Artificial Intelligence, volume 39, pages 8996–9004, 2025

    2025

  38. [46]

    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. InProceedings of the 32nd ACM International Conference on Multim...

    2024

  39. [47]

    Magicanimate: Temporally consistent human image animation using diffusion model

    Zhongcong Xu, Jianfeng Zhang, JunHao Liew, Hanshu Yan, Jia-Wei Liu, Chenxu Zhang, Jiashi Feng, and MikeZheng Shou. Magicanimate: Temporally consistent human image animation using diffusion model

  40. [48]

    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. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 8372–8382, 2024

    2024

  41. [49]

    Semantic latent motion for portrait video generation, 2025

    Qiyuan Zhang, Chenyu Wu, Wenzhang Sun, Huaize Liu, Donglin Di, Wei Chen, and Changqing Zou. Semantic latent motion for portrait video generation, 2025

    2025

  42. [50]

    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. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 586–595, 2018

    2018

  43. [51]

    Improved adam optimizer for deep neural networks

    Zijun Zhang. Improved adam optimizer for deep neural networks. In2018 IEEE/ACM 26th international symposium on quality of service (IWQoS), pages 1–2. Ieee, 2018

    2018

  44. [52]

    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. InProceedings of the 29th ACM International Conference on Multimedia, Oct 2021

    2021

  45. [53]

    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. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 4606–4615, 2023

    2023

  46. [54]

    Champ: Controllable and consistent human image animation with 3d parametric guidance

    Shenhao Zhu, Junming Leo Chen, Zuozhuo Dai, Zilong Dong, Yinghui Xu, Xun Cao, Yao Yao, Hao Zhu, and Siyu Zhu. Champ: Controllable and consistent human image animation with 3d parametric guidance. InEuropean Conference on Computer Vision, pages 145–162. Springer, 2024. 12 A APP...

    2024

  47. [55]

    w/o STAG

    and CatVTON [8] produce clothing category errors—for example, transforming shorts into long 13 pants or skirts. ViViD not only shows color deviations but also introduces spurious patterns absent from the input. Qualitative results on the VVT dataset.The experimental results on...

  48. [56]

    18 Input StableVITON Anydoor OOTDiffusion Ours CatVTON CP-VTON Figure 15: Qualitative comparison on the VITON-HD dataset

    spatiotemporal texture consistency during walking/turning motions (third row); 3) background integrity preservation during clothing replacement (fourth row). 18 Input StableVITON Anydoor OOTDiffusion Ours CatVTON CP-VTON Figure 15: Qualitative comparison on the VITON-HD datase...

Pith tools

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