Video Virtual Try-on with Conditional Diffusion Transformer Inpainter

Bolei Xu; Cheng Zou; Dandan Zheng; Jingdong Chen; Ming Yang; Senlin Cheng; Xiaobo Li

REVIEW 4 major objections 5 minor 46 references

Video Virtual Try-on with Conditional Diffusion Transformer Inpainter

T0 review · 4 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read ViTI reformulates video virtual try-on as a conditional video inpainting problem solved with a full 3D attention Diffusion Transformer, reporting the lowest VFID on VVT.

desk verdict A genuinely new video-try-on formulation with big reported gains on VVT, but the headline comparison is confounded by an extra fine-tuning stage on Vivid; needs matched-protocol evaluation before the architectural claim holds. read the letter →

arxiv 2506.21270 v1 pith:R2KISD2R submitted 2025-06-26 cs.CV

Cheng Zou , Senlin Cheng , Bolei Xu , Dandan Zheng , Xiaobo Li , Jingdong Chen , Ming Yang This is my paper

classification cs.CV

keywords videovirtualtry-onconditionalinpaintingdiffusiontransformerfull3Dattentiontemporalconsistencygarmentgenerationmodel

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

Video virtual try-on has usually been built by taking an image-based try-on model and adding a temporal attention layer, which the paper argues mainly smooths frames rather than modeling motion. ViTI instead starts from a video-generation problem: a Diffusion Transformer with full 3D spatial-temporal attention is first trained as a video inpainter, then progressively adapted to fill masked garment regions, and finally conditioned on a reference garment image. On the VVT benchmark the method reports VFID(I3D) of 2.121, compared with 3.345 for Tunnel Try-on and 3.405 for ViViD, together with improved SSIM and LPIPS. The paper also contributes a 51,278-clip human-centric dataset for video try-on pretraining. The practical stake is that online try-on videos could show a garment consistently across poses and viewpoints instead of flickering between frames.

What carries the argument

The carrying mechanism is a Diffusion Transformer with full 3D spatial-temporal attention, meaning self-attention is computed over the entire token sequence of all frames at once rather than over spatial tokens frame-by-frame. On top of this backbone, ViTI layers a masked inpainting objective, a mask reshaper that interpolates pixel-space masks into latent space, a curriculum of four masking strategies across three training stages, and a garment adapter that injects reference garment features through cross-attention in parallel with text cross-attention. A DensePose-based pose encoder adds spatial priors, and a temporal consistency loss $L_{\text{temporal}} = \sum_i \|\epsilon_\theta(z_t,t,c)_i - \epsilon_\theta(z_t,t,c)_{i+1}\|_2^2$ constrains consecutive latent frames. The pretraining dataset VTP supplies 51,278 clips whose garment masks come from an automatic segmentation model.

What would settle it

Take the VTP clips used for pretraining, replace the automatic garment masks with manually corrected, frame-consistent masks, retrain ViTI, and measure VFID(I3D) on VVT; if the score does not improve and garment-region pixel error is unchanged, the mask-accuracy assumption is not what carries the result.

Watch

Extended reading notes

Core claim

The central claim is that reformulating video virtual try-on as conditional video inpainting gives better spatial-temporal consistency than the first-image-then-video approach. The authors build a video inpainter on a Diffusion Transformer whose self-attention operates jointly over space and time, train it with a curriculum of masks (static boxes, moving boxes, instance masks, garment masks), and then add a garment encoder and a DensePose pose encoder as conditions. A temporal consistency loss penalizes differences between predicted noise latents in consecutive frames. The authors report that this design achieves the best quantitative results on VVT, most notably VFID(I3D) 2.121 versus 3.345 for Tunnel Try-on and 3.405 for ViViD, and they interpret this as evidence that starting from a video generation problem, rather than decorating an image model with temporal smoothing, is the right route.

Load-bearing premise

The VTP pretraining masks, produced automatically by a segmentation model, must be accurate and stable across frames; if they drift or include body pixels, the model learns to fill the wrong regions and temporal consistency suffers.

Editorial extensions

If this is right

Video try-on models can be built on video-generation backbones, so advances in video diffusion transformers transfer directly to the try-on task.
The 51,278-clip VTP dataset gives the community a pretraining resource that reduces reliance on the two small public try-on datasets.
The masking curriculum shows a route to train garment-specific inpainting when precisely labeled garment videos are scarce.
Explicit temporal consistency loss on predicted noise latents is a general stabilizer for diffusion video models trained on limited high-quality data.

Reading between the lines

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

The same conditional-video-inpainting recipe could apply to other region-editing tasks, such as replacing an object or background in video with a reference appearance, where a reference image conditions the inpainted area.
Because the paper does not report inference-time cost, a natural next test is whether the full 3D attention can be replaced by windowed or factorized attention without losing the VFID advantage.
The dependence on automatically generated garment masks suggests mask quality, rather than model capacity, may become the next scaling bottleneck; a sensitivity study that corrupts masks by varying amounts would test this directly.

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.

Desk Editor's Note

read the letter

Let me give you the short version. This paper does something new: it treats video virtual try-on as conditional video inpainting with a full-3D-attention diffusion transformer, rather than the usual image-then-temporally-inflate recipe. That framing is a real departure, and the ablations are honest about what each component contributes. The reported VFID(I3D) of 2.121 on VVT is a large jump over the 3.3 range of prior methods, and the paper ships a useful multi-stage training recipe plus a 51k-clip pretraining dataset (VTP) described in enough detail to be reconstructable.

The soft spot is the one the stress-test flagged, and it is load-bearing. The full ViTI is fine-tuned on Vivid, while the baselines in Table 1 are cited numbers from papers that were not retrained under the same protocol. So the 37% relative VFID gain is not attributable to the 3D-DiT design alone; extra in-domain fine-tuning could explain a large chunk of it. The reader's worry about VTP mask quality is milder — Table 4 shows the w/o VTP version still beats all baselines (2.935 vs 3.345), so even imperfect masks do not sink the headline. But the missing matched-protocol comparison does. There are also no error bars or significance tests anywhere, and the abstract's 'superior' overstates the case since SSIM (0.938) is below Vivid's 0.949. That is a minor overclaim, but it signals a bit of spin.

What the paper does well: the ablation study is genuinely informative — the garment encoder comparison (VAE vs +CLIP vs +DINOv2), the pose and temporal-loss ablations, and the stagewise VBench numbers all go in the expected direction and give a clear picture of where the gains come from. The masking-strategy taxonomy is a practical contribution. The writing is direct and the method section is understandable.

Bottom line: this deserves a serious referee, but the referee should press on a controlled comparison. Either retrain the main baselines on the same data (at least fine-tune Vivid and Tunnel Try-on on Vivid as well) or show an ablation that removes the Vivid fine-tuning from ViTI. Without that, the architectural conclusion is plausible but not demonstrated. I'd take it to reading group and would cite it for the task formulation and the dataset description, but I would not put the claimed SOTA in a paper until the confound is resolved.

Referee Report

4 major / 5 minor

Summary. ViTI formulates video virtual try-on as a conditional video inpainting task, built on a Diffusion Transformer with full 3D spatial-temporal attention. The method is trained in multiple stages with different masking strategies, on a newly collected 51,278-clip VTP dataset, and then fine-tuned on the Vivid dataset with a garment encoder, a DensePose pose encoder, and a temporal consistency loss. On the VVT benchmark, the paper reports a VFID(I3D) of 2.121 versus 3.345 for Tunnel Try-on and 3.405 for ViViD, and claims state-of-the-art performance.

Significance. If the results hold, the formulation of video try-on as conditional video inpainting is a meaningful departure from the prevailing 'first-image-then-temporal-attention' approach, and the VTP pretraining dataset is a potentially useful community resource. The paper also provides useful ablations (Tables 2--5) for garment encoder choice, pose conditioning, temporal loss, and pretraining data. However, the empirical evidence does not currently isolate the contribution of the proposed architecture from the extra fine-tuning data, and the headline claim of superiority is not uniformly supported by the reported metrics.

major comments (4)

[Sec. 3.3, Sec. 4.3, Table 1] The state-of-the-art comparison is not matched-protocol. Section 3.3 states that ViTI is optimized with L_total on the Vivid dataset, while the baseline numbers in Table 1 are taken from previous publications and were not retrained or fine-tuned with the same data protocol; image-based baselines such as OOTDiffusion and StableVITON were certainly not fine-tuned on Vivid. The 37% relative VFID improvement over Tunnel Try-on is therefore confounded by additional in-domain training data. The authors' Table 4 ablation (ViTI without VTP) does not resolve this confound, because that model still undergoes the Vivid optimization. Please provide a controlled comparison: either fine-tune the baselines on the same data and report their numbers, or ablate ViTI with the Vivid fine-tuning removed, including comparable training iterations.
[Abstract, Sec. 4.3, Table 1] The abstract claims ViTI is 'superior to previous works' on both quantitative and qualitative results, but Table 1 shows SSIM of 0.938, which is lower than ViViD's 0.949. The claim should be tempered to the metrics where ViTI actually leads (LPIPS and VFID), and the discrepancy should be discussed, since SSIM is one of the two single-frame quality metrics reported.
[Sec. 3.1, Sec. 4.4] The central architectural claim -- that full 3D spatial-temporal attention yields better temporal consistency than separated spatial/temporal attention -- is not directly tested. The paper motivates this in Sec. 3.1 and uses it in the final model, but the ablations in Sec. 4.4 vary other components (garment encoder, pose encoder, temporal loss, VTP data) and do not compare against a Latte-style stacked spatial/temporal attention backbone under the same training protocol. Without this ablation, the observed gains could stem from other design choices or from the additional Vivid fine-tuning. Please add such an ablation or rephrase the claim as a hypothesis.
[Sec. 4.3, Table 1] All quantitative results are reported as point estimates on a single benchmark with no error bars, multiple seeds, or significance tests. For example, the LPIPS advantage over Tunnel Try-on (0.042 vs. 0.054) and the SSIM deficit relative to ViViD (0.938 vs. 0.949) are small enough that run-to-run variance could change the ranking. Please report standard deviations over at least three seeds for the main comparison and ablations, or otherwise discuss the stability of the results.

minor comments (5)

[Sec. 3.2, Eq. (2)--(3)] Equation (2) defines a pixel-space reconstruction loss, but the actual DiT implementation uses the noise-domain loss in Eq. (3). The relationship between the two is not explained, and the notation xi_p in Eq. (2) is never used afterward. Please clarify whether Eq. (2) is motivated only or actually optimized in some stage.
[Sec. 3.2, Eq. (2)--(3)] The symbol 'P I' in Eqs. (2) and (3) is never defined. If it denotes an indicator function, please define it explicitly or use standard notation such as 1[cond].
[Sec. 3.2, Stage 3] The paper says the Stage 3 text prompt is 'associated with the garment mask area,' but it does not describe how these prompts are generated. For reproducibility, please specify the prompt-construction procedure or state that the prompt is a fixed template.
[Throughout] The method name is inconsistently written: 'Vivid' and 'ViViD' are both used (e.g., Sec. 3.3 vs. Table 1). Please use one spelling consistently, especially since Vivid is also the name of the fine-tuning dataset.
[Sec. 5] There is a typo in the Conclusions: 'diffsion transformer' should be 'diffusion transformer.'

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: evaluation is on held-out VVT with standard losses and no self-citation chain.

full rationale

The central claims are empirical and benchmarked on held-out data: ViTI is fine-tuned on Vivid/VTP and compared on the public VVT set against published baselines. There is no equation-level self-reduction: the temporal consistency regularizer (Eq. 4) constrains consecutive latent-frame noise predictions, whereas the reported VFID(I3D) metric compares I3D feature distributions of full generated videos, so the training loss is not the evaluation metric in disguise. No parameter is fitted to VFID and then reported as a prediction. The VTP garment masks are a training-data detail, not a derived result; Table 4 explicitly shows the model still beats baselines without VTP, so the mask pipeline is not load-bearing for the headline. No uniqueness theorem or central premise is imported from the authors' own prior work, and the reference list contains no self-citations. Unequal training protocols across baselines are a comparison-protocol concern for correctness, not circularity. Therefore no circular step can be exhibited.

Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

The central claim relies on standard diffusion assumptions plus two domain-specific assumptions (VAE fidelity and mask accuracy) and one stated hypothesis (DensePose continuity). No new physical or conceptual entities are introduced beyond the VTP dataset, which is a resource rather than an entity.

free parameters (2)

alpha (temporal consistency loss weight) = 0.1
Chosen by hand in Eq. (5); no sensitivity analysis or ablation of alpha is provided, and the trade-off between reconstruction and temporal smoothness is not explored.
garment cross-attention guidance scale = not reported
Mentioned in Sec 3.3 as a scale factor controlling the garment condition strength; no value or sensitivity study is given.

assumptions (4)

domain assumption The causal video VAE provides a latent space where the masked-noise reconstruction loss is a valid proxy for pixel-level inpainting quality.
Sec 3.2, uses Open-Sora-Plan VAE; no analysis of compression artifacts.
domain assumption The Sapiens segmentation model produces accurate garment masks for VTP videos.
Sec 4.1, VTP construction; mask accuracy not quantified.
ad hoc to paper DensePose depth is continuous in space-time and provides a useful pose prior for try-on.
Sec 3.3, stated as a hypothesis; ablation shows benefit on VVT but no external validation.
ad hoc to paper Stage-3 text prompts are semantically aligned with the masked garment region.
Sec 3.2 mentions 'text prompt is associated with the garment mask area' but generation procedure is not described.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Video Virtual Try-on with Conditional Diffusion Transformer Inpainter." pith.science (2026). https://pith.science/paper/R2KISD2R

@misc{pith2026250621270,
  author       = {Pith},
  title        = {Pith review of: Video Virtual Try-on with Conditional Diffusion Transformer Inpainter},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/R2KISD2R}},
  note         = {Machine review of arXiv:2506.21270}
}

read the original abstract

Video virtual try-on aims to naturally fit a garment to a target person in consecutive video frames. It is a challenging task, on the one hand, the output video should be in good spatial-temporal consistency, on the other hand, the details of the given garment need to be preserved well in all the frames. Naively using image-based try-on methods frame by frame can get poor results due to severe inconsistency. Recent diffusion-based video try-on methods, though very few, happen to coincide with a similar solution: inserting temporal attention into image-based try-on model to adapt it for video try-on task, which have shown improvements but there still exist inconsistency problems. In this paper, we propose ViTI (Video Try-on Inpainter), formulate and implement video virtual try-on as a conditional video inpainting task, which is different from previous methods. In this way, we start with a video generation problem instead of an image-based try-on problem, which from the beginning has a better spatial-temporal consistency. Specifically, at first we build a video inpainting framework based on Diffusion Transformer with full 3D spatial-temporal attention, and then we progressively adapt it for video garment inpainting, with a collection of masking strategies and multi-stage training. After these steps, the model can inpaint the masked garment area with appropriate garment pixels according to the prompt with good spatial-temporal consistency. Finally, as other try-on methods, garment condition is added to the model to make sure the inpainted garment appearance and details are as expected. Both quantitative and qualitative experimental results show that ViTI is superior to previous works.

Figures

Figures reproduced from arXiv: 2506.21270 by the authors.

**Figure 1.** Results of ViTI (Video Try-on Inpainter). The first two columns provide the input. ViTI performs well on garment detail [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗

**Figure 2.** Overall architecture. The main framework of ViTI (Video Try-on Inpainter) consists of two parts, the video inpainting part, and [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗

**Figure 4.** Masking strategies for garment inpainting model. These [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗

Figures from the paper (2 more)

**Figure 5.** Figure 5: Visualization of an example of the proposed dataset VTP [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]

**Figure 6.** Figure 6: Qualitative comparisons with other methods on the pub [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

46 extracted references · 33 canonical work pages

[1]

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

Alberto Baldrati, Davide Morelli, Giuseppe Cartella, Mar- cella Cornia, Marco Bertini, and Rita Cucchiara. Multimodal garment designer: Human-centric latent diffusion models for fashion image editing. In Proceedings of the IEEE/CVF In- ternational Conference on Computer Vision , pages 23393– 23402, 2023. 1, 3

work page 2023
[2]

The 2019 davis challenge on vos: Unsupervised multi-object seg- mentation

Sergi Caelles, Jordi Pont-Tuset, Federico Perazzi, Alberto Montes, Kevis-Kokitsi Maninis, and Luc Van Gool. The 2019 davis challenge on vos: Unsupervised multi-object seg- mentation. arXiv:1905.00737, 2019. 5

arXiv 2019
[3]

Quo vadis, action recognition? a new model and the kinetics dataset, 2018

Joao Carreira and Andrew Zisserman. Quo vadis, action recognition? a new model and the kinetics dataset, 2018. 7

work page 2018
[4]

Infusion: internal diffusion for inpainting of dynamic textures and complex motion

Nicolas Cherel, Andr ´es Almansa, Yann Gousseau, and Alas- dair Newson. Infusion: Internal diffusion for video inpaint- ing. arXiv preprint arXiv:2311.01090, 2023. 3

work page Pith review arXiv 2023
[5]

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. In Proc. of the IEEE conference on computer vision and pattern recognition (CVPR), 2021. 1, 3

work page 2021
[6]

Mevis: A large-scale benchmark for video segmentation with motion expressions, 2023

Henghui Ding, Chang Liu, Shuting He, Xudong Jiang, and Chen Change Loy. Mevis: A large-scale benchmark for video segmentation with motion expressions, 2023. 5

work page 2023
[7]

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. In Proceedings of the IEEE/CVF international conference on computer vision, pages 1161–1170, 2019. 3, 7

work page 2019
[8]

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

work page

Show all 46 references

[9]

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 In Proceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition , page 8485–8493, 2021. 7

2021
[10]

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 con- ference on computer vision and pattern recognition , pages 8485–8493, 2021. 1, 3

2021
[11]

Parser-free virtual try-on via distilling appearance flows, 2021

Yuying Ge, Yibing Song, Ruimao Zhang, Chongjian Ge, Wei Liu, and Ping Luo. Parser-free virtual try-on via distilling appearance flows, 2021. 7

2021
[12]

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. arXiv preprint arXiv:2308.06101, 2023. 1, 3

2023 arXiv
[13]

Se- mantically consistent video inpainting with conditional dif- fusion models

Dylan Green, William Harvey, Saeid Naderiparizi, Matthew Niedoba, Yunpeng Liu, Xiaoxuan Liang, Jonathan Laving- ton, Ke Zhang, Vasileios Lioutas, Setareh Dabiri, et al. Se- mantically consistent video inpainting with conditional dif- fusion models. arXiv preprint arXiv:2405.00...

2024 arXiv
[14]

Flow-guided diffusion for video inpainting

Bohai Gu, Yongsheng Yu, Heng Fan, and Libo Zhang. Flow-guided diffusion for video inpainting. arXiv preprint arXiv:2311.15368, 2023. 3

2023 arXiv
[15]

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

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. Interna- tional Conference on Learning Representations, 2024. 3

2024
[16]

Densepose: Dense human pose estimation in the wild, 2018

Rıza Alp G ¨uler, Natalia Neverova, and Iasonas Kokkinos. Densepose: Dense human pose estimation in the wild, 2018. 6

2018
[17]

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, 3

2018
[18]

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

Kensho Hara, Hirokatsu Kataoka, and Yutaka Satoh. Can spatiotemporal 3d cnns retrace the history of 2d cnns and imagenet?, 2018. 7

2018
[19]

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. arXiv preprint arXiv:2407.10625, 2024. 7

2024 arXiv
[20]

Lvos: A benchmark for large-scale long-term video object segmentation, 2024

Lingyi Hong, Zhongying Liu, Wenchao Chen, Chenzhi Tan, Yuang Feng, Xinyu Zhou, Pinxue Guo, Jinglun Li, Zhaoyu Chen, Shuyong Gao, Wei Zhang, and Wenqiang Zhang. Lvos: A benchmark for large-scale long-term video object segmentation, 2024. 5

2024
[21]

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

Thibaut Issenhuth, J ´er´emie Mary, and Cl ´ement Calauzenes. Do not mask what you do not need to mask: a parser-free virtual try-on. In Computer Vision–ECCV 2020: 16th Euro- pean Conference, Glasgow, UK, August 23–28, 2020, Pro- ceedings, Part XX 16, pages 619–635. Springer, ...

2020
[22]

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. In Pro- ceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2022. 1, 3, 6, 7

2022
[23]

Sapiens: Foundation for human vision mod- els

Rawal Khirodkar, Timur Bagautdinov, Julieta Martinez, Su Zhaoen, Austin James, Peter Selednik, Stuart Anderson, and Shunsuke Saito. Sapiens: Foundation for human vision mod- els. arXiv preprint arXiv:2408.12569, 2024. 2

2024 arXiv
[24]

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

Jeongho Kim, Gyojung Gu, Minho Park, Sunghyun Park, and Jaegul Choo. Stableviton: Learning semantic correspon- dence with latent diffusion model for virtual try-on. arXiv preprint arXiv:2312.01725, 2023. 1, 3, 7

2023 arXiv
[25]

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

Jeongho Kim, Guojung Gu, Minho Park, Sunghyun Park, and Jaegul Choo. Stableviton: Learning semantic correspon- dence with latent diffusion model for virtual try-on. In Pro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 8176–8185, 2024. 7

2024
[26]

Open-sora-plan, 2024

PKU-Yuan Lab and Tuzhan AI etc. Open-sora-plan, 2024. 2, 5

2024
[27]

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 Eu- ropean Conference on Computer Vision , pages 204–219. Springer, 2022. 1, 3

2022
[28]

Latte: La- tent diffusion transformer for video generation

Xin Ma, Yaohui Wang, Gengyun Jia, Xinyuan Chen, Ziwei Liu, Yuan-Fang Li, Cunjian Chen, and Yu Qiao. Latte: La- tent diffusion transformer for video generation. 2024. 4

2024
[29]

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. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 2231–2235, 2022. 1, 3

2022
[30]

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 Proceedings of the ACM International Confer- ence on Multimedia, 2023. 1, 3

2023
[31]

Dinov2: Learning robust visual features with- out supervision, 2024

Maxime Oquab, Timoth ´ee Darcet, Th ´eo Moutakanni, Huy V o, Marc Szafraniec, Vasil Khalidov, Pierre Fernandez, Daniel Haziza, Francisco Massa, Alaaeldin El-Nouby, Mah- moud Assran, Nicolas Ballas, Wojciech Galuba, Russell Howes, Po-Yao Huang, Shang-Wen Li, Ishan Misra, Michae...

2024
[32]

The 2017 davis challenge on video object segmentation, 2018

Jordi Pont-Tuset, Federico Perazzi, Sergi Caelles, Pablo Ar- bel´aez, Alex Sorkine-Hornung, and Luc Van Gool. The 2017 davis challenge on video object segmentation, 2018. 5

2017
[33]

Colin Raffel, Noam Shazeer, Adam Roberts, Katherine Lee, Sharan Narang, Michael Matena, Yanqi Zhou, Wei Li, and Peter J. Liu. Exploring the limits of transfer learning with a unified text-to-text transformer, 2023. 5

2023
[34]

To- wards accurate generative models of video: A new metric & challenges

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. 7

2018 arXiv
[35]

Toward characteristic- preserving image-based virtual try-on network, 2018

Bochao Wang, Huabin Zheng, Xiaodan Liang, Yimin Chen, Liang Lin, and Meng Yang. Toward characteristic- preserving image-based virtual try-on network, 2018. 7

2018
[36]

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 Proceed- ings of the European conference on computer vision (ECCV), pages 589–604, 2018. 1, 3, 7

2018
[37]

Image quality assessment: from error visibility to structural similarity

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. 7

2004
[38]

Oot- diffusion: Outfitting fusion based latent diffusion for control- lable virtual try-on

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. 3, 7

2024 arXiv
[39]

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. arXiv preprint, 2024. 1, 3, 7

2024
[40]

Video instance seg- mentation, 2019

Linjie Yang, Yuchen Fan, and Ning Xu. Video instance seg- mentation, 2019. 5

2019
[41]

Ip- adapter: Text compatible image prompt adapter for text-to- image diffusion models

Hu Ye, Jun Zhang, Sibo Liu, Xiao Han, and Wei Yang. Ip- adapter: Text compatible image prompt adapter for text-to- image diffusion models. InarXiv preprint arxiv:2308.06721,

arXiv
[42]

Efros, Eli Shecht- man, and Oliver Wang

Richard Zhang, Phillip Isola, Alexei A. Efros, Eli Shecht- man, and Oliver Wang. The unreasonable effectiveness of deep features as a perceptual metric, 2018. 7

2018
[43]

Avid: Any-length video inpainting with dif- fusion model

Zhixing Zhang, Bichen Wu, Xiaoyan Wang, Yaqiao Luo, Luxin Zhang, Yinan Zhao, Peter Vajda, Dimitris Metaxas, and Licheng Yu. Avid: Any-length video inpainting with dif- fusion model. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 71...

2024
[44]

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. In Proceedings of the 29th ACM International Conference on Multimedia, pages 908–916, 2021. 3

2021
[45]

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 Com- puter Vision and Pattern Recognition , pages 4606–4615,
[46]

Designing a better asymmetric vqgan for stablediffusion, 2023

Zixin Zhu, Xuelu Feng, Dongdong Chen, Jianmin Bao, Le Wang, Yinpeng Chen, Lu Yuan, and Gang Hua. Designing a better asymmetric vqgan for stablediffusion, 2023. 6

2023

Pith tools

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

[1] [1]

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

Alberto Baldrati, Davide Morelli, Giuseppe Cartella, Mar- cella Cornia, Marco Bertini, and Rita Cucchiara. Multimodal garment designer: Human-centric latent diffusion models for fashion image editing. In Proceedings of the IEEE/CVF In- ternational Conference on Computer Vision , pages 23393– 23402, 2023. 1, 3

work page 2023

[2] [2]

The 2019 davis challenge on vos: Unsupervised multi-object seg- mentation

Sergi Caelles, Jordi Pont-Tuset, Federico Perazzi, Alberto Montes, Kevis-Kokitsi Maninis, and Luc Van Gool. The 2019 davis challenge on vos: Unsupervised multi-object seg- mentation. arXiv:1905.00737, 2019. 5

arXiv 2019

[3] [3]

Quo vadis, action recognition? a new model and the kinetics dataset, 2018

Joao Carreira and Andrew Zisserman. Quo vadis, action recognition? a new model and the kinetics dataset, 2018. 7

work page 2018

[4] [4]

Infusion: internal diffusion for inpainting of dynamic textures and complex motion

Nicolas Cherel, Andr ´es Almansa, Yann Gousseau, and Alas- dair Newson. Infusion: Internal diffusion for video inpaint- ing. arXiv preprint arXiv:2311.01090, 2023. 3

work page Pith review arXiv 2023

[5] [5]

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. In Proc. of the IEEE conference on computer vision and pattern recognition (CVPR), 2021. 1, 3

work page 2021

[6] [6]

Mevis: A large-scale benchmark for video segmentation with motion expressions, 2023

Henghui Ding, Chang Liu, Shuting He, Xudong Jiang, and Chen Change Loy. Mevis: A large-scale benchmark for video segmentation with motion expressions, 2023. 5

work page 2023

[7] [7]

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. In Proceedings of the IEEE/CVF international conference on computer vision, pages 1161–1170, 2019. 3, 7

work page 2019

[8] [8]

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

work page