Pith. sign in

REVIEW 3 major objections 5 minor 41 references

Enhancing Virtual Try-On with Synthetic Pairs and Error-Aware Noise Scheduling

T0 review · 3 major / 5 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read This paper claims that virtual try-on quality can be pushed past prior baselines by augmenting training with ~21,000 synthetic (person, garment) pairs and refining base-model outputs with an error-aware Schrödinger bridge.

desk verdict The synthetic-pair augmentation idea is practical and well-ablated, but EARSB's sampling equations as written do not invert the bridge forward process, so the central claim needs correction or code verification. read the letter →

arxiv 2501.04666 v3 pith:4DRZAPAQ submitted 2025-01-08 cs.CV

Nannan Li , Kevin J. Shih , Bryan A. Plummer This is my paper · ORCID
classification cs.CV
keywords virtualtry-onsyntheticdataaugmentationhuman-to-garmentextractionSchrödingerbridgeerror-awarenoiseschedulingweakly-supervisederrorclassificationdiffusionrefinementimagegeneration
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 tries to establish that two low-cost ingredients can push virtual try-on quality past prior GAN- and diffusion-based methods: synthetic training pairs, made by a human-to-garment model that turns photos of clothed people into canonical product-view garment images, and a refinement stage that repairs only the flawed parts of a base model's output. The refinement stage, called EARSB, uses a weakly-supervised error classifier to build a heatmap of artifacts and then feeds that heatmap into a Schrödinger bridge's noise schedule, so noise is added precisely where errors are. On VITON-HD and DressCode-Upper the paper reports better FID, KID, and SSIM than seven prior baselines, and a user study prefers its outputs in 59% of pairwise comparisons. If the paper is right, existing try-on models can be improved without collecting new paired product images or retraining them from scratch.

What carries the argument

Two mechanisms carry the argument. The human-to-garment model is a ClothFlow-style UNet with flow-like feature warping that maps the segmented clothing region of a person photo onto the canonical product view, trained with an L1 reconstruction plus adversarial loss on existing paired data such as VITON-HD. The refinement mechanism, EARSB, extends the I2SB Schrödinger bridge: the noise in the forward process is replaced by εr = M·ε, where M = WSC(x1, C) is the sigmoid error heatmap from a weakly-supervised two-encoder classifier trained with a few hours of bounding-box labels; training minimizes the mean squared error between the predicted reweighted noise and εr, and sampling re-injects noise and guidance only where M is high. The error map thus acts as a per-pixel noise scheduler: pixels judged correct are copied through nearly untouched, and pixels judged artifact-ridden are re-synthesized.

What would settle it

Take a toy problem with known endpoints — for example, two simple image classes or Gaussian point clouds where the true conditional distribution of x0 given x1 is computable — and run the masked schedule of Eqs. (3)-(8) with a deliberately imperfect error map. If the masked process still transports samples to the correct x0 distribution within the I2SB objective's tolerance, the bridge claim holds; if the chain shows systematic bias in low-M regions, the error-aware success is a property of this specific schedule rather than of Schrödinger bridge theory.

Watch

Extended reading notes

Core claim

On its own terms, the paper's central claim is that virtual try-on errors are best fixed by knowing where they are. The human-to-garment model — a flow-based UNet trained on existing paired data — extracts a canonical garment view from a single clothed-person image, and after filtering the resulting pairs by background cleanliness, frontal pose, and LPIPS reconstruction error, roughly 21,000 synthetic (human, garment) pairs from DeepFashion2 and UPT improve multiple base try-on models when mixed into training at a 15% batch ratio with a real/synthetic conditioning flag. The second claim is that the image-to-image Schrödinger bridge of I2SB becomes a sharper refiner when the noise is reweighted pointwise by the error map, εr = M·ε, so clean regions are copied through nearly noiseless while artifact regions receive more noise, with the same error map masking the classifier guidance term during sampling. The full pipeline, EARSB with synthetic augmentation, reports the best FID, KID, and SSIM on both benchmarks, and the ablations trace the gain to the error map itself: uniform noise, random maps, inpainting of flagged regions, and removed guidance all do worse.

Load-bearing premise

The load-bearing premise is that multiplying the bridge noise pointwise by an arbitrary error map, εr = M·ε, still yields a valid Schrödinger bridge that can be trained with the standard I2SB objective; the paper does not derive that the masked noise preserves the required forward-backward marginals or that the reverse-time sampler remains consistent.

Editorial extensions

If this is right

  • Any existing virtual try-on model, GAN- or diffusion-based, can be wrapped by EARSB and improved without modifying the base model's weights.
  • Synthetic pairs substitute for scarce copyrighted product photos: roughly 21,000 filtered pairs sourced from unlabeled human images suffice to improve several published try-on models.
  • The error-aware schedule keeps image quality at fewer sampling steps, so refinement can be run cheaply when the initial image is mostly good.
  • Because the error classifier is trained on a specific base model's artifacts, retargeting the refiner to a new base model costs only a few hours of bounding-box labeling.
  • The real/synthetic conditioning flag matters: mixing synthetic pairs without the label hurts performance, while labeled mixing beats pretrain-then-finetune.

Reading between the lines

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

  • The error-masked process εr = M·ε is asserted rather than derived; a natural test is whether the masked forward-backward process still converges to the true endpoints on a toy distribution, and if not, the gains may depend on the specific error-map statistics rather than a general bridge principle.
  • The same refine-where-flagged recipe should transfer to other image-to-image tasks with cheap error localizers, such as face restoration or text rendering in edited images, though the paper only demonstrates try-on.
  • The synthetic-pair gain may scale further with 3D-aware garment extraction; the paper's own limitation note reports residual texture deformations because the generator operates in image space without 3D information.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. This paper proposes two complementary improvements to virtual try-on. First, it trains a human-to-garment model that converts a single image of a clothed person into a canonical product-view garment image, creating (human, synthetic garment) pairs that are filtered and used to augment try-on training under a real/synthetic flag to mitigate the domain gap. Second, it introduces EARSB, a refinement module built on the Image-to-Image Schrödinger Bridge (I2SB): a weakly supervised classifier (WSC) predicts a spatial error map from an initial try-on output, and that map is used to reweight the diffusion noise, condition the denoiser, and guide the reverse sampler. Experiments on VITON-HD and DressCode-Upper report improved FID, KID, SSIM, and LPIPS over GAN and diffusion baselines, and a user study prefers EARSB in 59% of cases on average.

Significance. If the results hold, the synthetic-data contribution is practical and well ablated: Table 3 shows that H2G-UH improves CAT-DM and StableVITON, and Table 4 isolates the benefit of flag-conditioned augmentation over plain mixing and pretraining. The error-map idea is also interesting; Table 5 demonstrates that a meaningful map outperforms a random map and an unweighted I2SB baseline. The paper also ships code, which is valuable for reproducing the experiments. The main weakness is mathematical: the sampling equations in §4.2 do not, as written, invert the forward process, which undermines the central claim that EARSB is a valid Schrödinger bridge. The quantitative results are reported as single numbers without error bars or significance tests, which makes some of the smaller reported gains difficult to evaluate.

major comments (3)
  1. [Sec. 4.2, Eqs. (5)-(8)] The proposed sampler does not invert the forward process in Eqs. (2)-(4). If the noise prediction is perfect (ε^r_θ = ε^r), Eq. (5) gives \hat{x}_0 = μ_t(x0,x1) = (σ̄_t²/(σ̄_t²+σ_t²)) x0 + (σ_t²/(σ̄_t²+σ_t²)) x1, not x0. The correct denoising estimate must remove the known x1 term and the normalization, e.g. \hat{x}_0 = (x_t − (σ_t²/(σ̄_t²+σ_t²)) x1 − √Σ_t ε^r_θ) / (σ̄_t²/(σ̄_t²+σ_t²)). Consequently Eq. (7) conditions the Brownian-bridge posterior on the wrong value, and the sampler remains anchored to the artifact image x1. This is not a consequence of the mask M; it already occurs when M is all ones, so the statement that EARSB reverts to I2SB when M=1 is not supported. Please correct the equations or clearly report the actual sampler used in the released code; as written, the theoretical framing of EARSB as a Schrödinger bridge is invalid.
  2. [Sec. 4.2, Eqs. (3)-(4) and (6)] The paper introduces the spatially reweighted noise ε^r = M·ε and a reverse update that again injects M·√Σ_t ε, but it provides no derivation that this reweighted stochastic process is a valid Schrödinger bridge or that the reverse update is consistent with the forward marginals of Eqs. (3)-(4). Since the central contribution is an "Error-Aware ... Schrödinger Bridge," the authors should either supply the supporting derivation or explicitly characterize the M-reweighting as a heuristic modification of I2SB and evaluate it on that basis.
  3. [Tables 1-5 and Fig. 6] All quantitative results are single measurements without error bars, confidence intervals, or significance tests. Several of the key gains that support the synthetic-data claim are small (e.g., Table 3: StableVITON unpaired FID 8.25→8.17; Table 5: EARSB w.o. M vs. full, unpaired FID 8.48 vs. 8.42 with identical SSIM/LPIPS to three decimals), so it is difficult to know whether these differences are reliable. Please report variance over multiple seeds or a paired significance test for the central comparisons.
minor comments (5)
  1. [Sec. 4.1, Eq. (1)] The final loss is written as "Lins + Lpat," which appears to be a typo for "Limg + Lpat"; the patch-level loss also omits an explicit averaging or summation over spatial locations.
  2. [Sec. 4.2, after Eq. (8)] The sentence "When M is all zeros (i.e., no error), x1 is believed to be perfect x1 does not need to be refined" has a missing comma and reads incorrectly; please rephrase.
  3. [Table 1] The GP-VTON row reports KID 0.86 for VITON-HD unpaired but 0.07 for paired, which is atypical compared with the other baselines; please double-check the KID computation for that row.
  4. [Sec. 5.1 and Table 3] The main table uses 25 sampling steps for all diffusion models, while Table 3 uses each method's original step count (2 for CAT-DM, 50 for StableVITON, 25 for EARSB); this makes cross-table comparisons inconsistent and should be clarified.
  5. [Sec. 3.2] The filtering criterion (c) uses LPIPS reconstruction error from a try-on model; since that try-on model is trained on the same benchmark used for evaluation, the potential selection bias or data leakage in synthetic-pair construction should be discussed.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: the synthetic-data filter is a selection heuristic, the WSC is independently trained, and the EARSB gains are measured against external benchmarks; the few self-citations are not load-bearing.

full rationale

The paper's main claims are not circular by construction. The synthetic-pair augmentation in Sec. 3.2 filters generated garments using criteria that include a try-on reconstruction loss, but this is a data-quality selection step, not a fitted parameter renamed as a prediction: the selected pairs are then used to train different try-on models and EARSB, and the reported gains are on held-out VITON-HD and DressCode-Upper test sets rather than on the filtering loss itself. The weakly-supervised classifier WSC is trained from bounding-box annotations and real/fake image labels, independently of the EARSB sampling objective, so using its error map M to reweight the noise schedule does not reduce to the classifier's own training signal. The EARSB formulation is built on the external I2SB framework and is evaluated against GAN and diffusion baselines, including ablated versions (w.o. M, rand(M), w.o. CG), which provide a controlled test of the error-map contribution. The only self-citations are [22], used to obtain the WVTON out-of-domain test set, and [29], cited as an alternative pretraining strategy that the paper tests and finds inferior to its flag-conditioning approach; neither is load-bearing for the central claim. The known concern that Eq. (5) may omit the x1 term in the clean-image estimate is a mathematical correctness issue, not a circularity, so it is left out of this score.

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

The central claims rest on a learned error map and a modified diffusion process; no new physical entities are introduced. The main free choices are the error map reweighting, hyperparameters (guidance scale, batch ratio, sampling steps), and the synthetic data filtering, none of which are fit to the test set. The axioms are the validity of the reweighted Schrödinger bridge, the accuracy of the error classifier, the reasonableness of the initial image, and the usefulness of the filtered synthetic distribution.

free parameters (3)
  • synthetic data batch ratio = 15%
    Chosen by hand for EARSB training (Table 6); the proportion of synthetic pairs in each batch affects the balance between real and synthetic learning.
  • classifier guidance scale = 6
    Applied to Eq. (10) and clamped to [-0.3, 0.3] during inference (Appendix A); this scale controls the strength of the WSC guidance and was tuned without report of a sweep.
  • number of sampling steps = 25
    Used for EARSB and baselines in Table 1; Fig. 4 shows that performance varies with the number of steps, so this choice affects the reported results.
assumptions (4)
  • ad hoc to paper The error-map-reweighted noise process εr = M·ε is a valid image-to-image Schrödinger bridge with the same reverse-time denoising equations as I2SB.
    Stated in Sec. 4.2 (Eqs. 3-9) without proof; the reverse sampler in Eqs. (5)-(8) is assumed to remain correct when the noise covariance is spatially modulated by M.
  • domain assumption The weakly supervised error map M reliably localizes all regions that need refinement.
    Used throughout Sec. 4; if M misses artifacts, EARSB will not correct them, and if M marks good regions, unnecessary noise may degrade them.
  • domain assumption The initial image x1 from the base try-on model is of reasonable quality and requires only partial, localized refinement.
    Explicitly acknowledged in Sec. F (Limitations): very poor initial images cannot be fully regenerated.
  • domain assumption Filtering synthetic pairs by LPIPS reconstruction error preserves the usefulness of the synthetic distribution for training.
    Sec. 3.2 criterion (c); this selection can bias the synthetic set toward easy examples and may not represent the harder cases that virtual try-on needs.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Enhancing Virtual Try-On with Synthetic Pairs and Error-Aware Noise Scheduling." pith.science (2026). https://pith.science/paper/4DRZAPAQ

@misc{pith2026250104666,
  author       = {Pith},
  title        = {Pith review of: Enhancing Virtual Try-On with Synthetic Pairs and Error-Aware Noise Scheduling},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/4DRZAPAQ}},
  note         = {Machine review of arXiv:2501.04666}
}
read the original abstract

Given an isolated garment image in a canonical product view and a separate image of a person, the virtual try-on task aims to generate a new image of the person wearing the target garment. Prior virtual try-on works face two major challenges in achieving this goal: a) the paired (human, garment) training data has limited availability; b) generating textures on the human that perfectly match that of the prompted garment is difficult, often resulting in distorted text and faded textures. Our work explores ways to tackle these issues through both synthetic data as well as model refinement. We introduce a garment extraction model that generates (human, synthetic garment) pairs from a single image of a clothed individual. The synthetic pairs can then be used to augment the training of virtual try-on. We also propose an Error-Aware Refinement-based Schr\"odinger Bridge (EARSB) that surgically targets localized generation errors for correcting the output of a base virtual try-on model. To identify likely errors, we propose a weakly-supervised error classifier that localizes regions for refinement, subsequently augmenting the Schr\"odinger Bridge's noise schedule with its confidence heatmap. Experiments on VITON-HD and DressCode-Upper demonstrate that our synthetic data augmentation enhances the performance of prior work, while EARSB improves the overall image quality. In user studies, our model is preferred by the users in an average of 59% of cases.

Figures

Figures reproduced from arXiv: 2501.04666 by the authors.

Figure 1
Figure 1. Example of our proposed Error-Aware Refinement [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. (a) Our human-to-garment model, which is explained in [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. The diffusion process in our refinement-based EARSB. We first preprocess the input image, then use a base try-on model that [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (12 more)
Figure 4
Figure 4. Figure 4: Results on VITON-HD at 5, 10, 25, 50, and 100 sampling steps. Our method consistently improves our baseline starting model [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 5
Figure 5. Figure 5: Visualizations on VITON-HD (top row) and DressCode (bottom row). Our EARSB+H2G-UH and EARSB(SD)+H2G-UH better [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 6
Figure 6. Figure 6: The precision-recall curve for retrieving annotated errors [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]
Figure 8
Figure 8. Figure 8: Architecture of our UNet in EARSB [PITH_FULL_IMAGE:figures/full_fig_p011_8.png]
Figure 9
Figure 9. Figure 9: Architecture of our UNet in the human-to-garment [PITH_FULL_IMAGE:figures/full_fig_p011_9.png]
Figure 10
Figure 10. Figure 10: Results on different time steps. Our error map focuses on low-quality regions and maintains the quality of the sufficiently good [PITH_FULL_IMAGE:figures/full_fig_p012_10.png]
Figure 11
Figure 11. Figure 11: Failure cases on VITON-HD where the initial image has a poor-quality. [PITH_FULL_IMAGE:figures/full_fig_p013_11.png]
Figure 12
Figure 12. Figure 12: Visualization of the generated images in WVTON. [PITH_FULL_IMAGE:figures/full_fig_p014_12.png]
Figure 13
Figure 13. Figure 13: Visualized examples of the (human, synthetic garment) pairs on our proposed H2G-UH. [PITH_FULL_IMAGE:figures/full_fig_p015_13.png]
Figure 14
Figure 14. Figure 14: Visualized examples of the (human, synthetic garment) pairs on our proposed H2G-FH. [PITH_FULL_IMAGE:figures/full_fig_p016_14.png]
Figure 15
Figure 15. Figure 15: Visualized examples on VITON-HD. Our EARSB and EARSB+H2G-UH better recovers the intricate textures in the garment. [PITH_FULL_IMAGE:figures/full_fig_p017_15.png]
Figure 16
Figure 16. Figure 16: Visualized examples on DressCode-Upper. Our EARSB and EARSB+H2G-UH better reconstructs the texts and graphics in the [PITH_FULL_IMAGE:figures/full_fig_p018_16.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

41 extracted references · 33 canonical work pages

  1. [1]

    Self- improving diffusion models with synthetic data

    Sina Alemohammad, Ahmed Imtiaz Humayun, Shruti Agar- wal, John Collomosse, and Richard Baraniuk. Self- improving diffusion models with synthetic data. arXiv preprint arXiv:2408.16333, 2024. 2

    arXiv 2024

  2. [2]

    eDiff-I: Text-to-image diffusion models with ensemble of expert denoisers

    Yogesh Balaji, Seungjun Nah, Xun Huang, Arash Vahdat, Jiaming Song, Qinsheng Zhang, Karsten Kreis, Miika Ait- tala, Timo Aila, Samuli Laine, Bryan Catanzaro, Tero Kar- ras, and Ming-Yu Liu. eDiff-I: Text-to-image diffusion models with ensemble of expert denoisers. arXiv preprint arXiv:2211.01324, 2022. 5, 11

    arXiv 2022

  3. [3]

    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 CVPR, 2023. 2

    work page 2023

  4. [4]

    Demystifying MMD GANs

    Mikołaj Bi ´nkowski, Danica J Sutherland, Michael Arbel, and Arthur Gretton. Demystifying MMD GANs. In ICLR, 2018. 6

    work page 2018

  5. [5]

    In- structpix2pix: Learning to follow image editing instructions

    Tim Brooks, Aleksander Holynski, and Alexei A Efros. In- structpix2pix: Learning to follow image editing instructions. In CVPR, 2023. 2

    2023

  6. [6]

    Improving diffusion models for vir- tual try-on

    Yisol Choi, Sangkyung Kwak, Kyungmin Lee, Hyungwon Choi, and Jinwoo Shin. Improving diffusion models for vir- tual try-on. In ECCV, 2024. 6

    work page 2024

  7. [7]

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

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

    arXiv

  8. [8]

    Diffusion posterior sampling for general noisy inverse problems

    Hyungjin Chung, Jeongsol Kim, Michael T Mccann, Marc L Klasky, and Jong Chul Ye. Diffusion posterior sampling for general noisy inverse problems. In ICLR, 2023. 5

    work page 2023

Show all 41 references
  1. [9]

    Learn- ing garment densepose for robust warping in virtual try-on

    Aiyu Cui, Sen He, Tao Xiang, and Antoine Toisoul. Learn- ing garment densepose for robust warping in virtual try-on. arXiv preprint arXiv:2303.17688, 2023. 14

    2023 arXiv

  2. [10]

    Deepsmote: Fusing deep learning and smote for imbalanced data

    Damien Dablain, Bartosz Krawczyk, and Nitesh V Chawla. Deepsmote: Fusing deep learning and smote for imbalanced data. IEEE Transactions on Neural Networks and Learning Systems, 34(9):6390–6404, 2022. 2

    2022

  3. [11]

    Diffusion models beat gans on image synthesis

    Prafulla Dhariwal and Alexander Nichol. Diffusion models beat gans on image synthesis. NeurIPS, 2021. 2, 5

    2021

  4. [12]

    Stylegan-human: A data-centric odyssey of human genera- tion

    Jianglin Fu, Shikai Li, Yuming Jiang, Kwan-Yee Lin, Chen Qian, Chen Change Loy, Wayne Wu, and Ziwei Liu. Stylegan-human: A data-centric odyssey of human genera- tion. In European Conference on Computer Vision, 2022. 2, 3

    2022

  5. [13]

    Implicit diffusion models for continuous super-resolution

    Sicheng Gao, Xuhui Liu, Bohan Zeng, Sheng Xu, Yan- jing Li, Xiaoyan Luo, Jianzhuang Liu, Xiantong Zhen, and Baochang Zhang. Implicit diffusion models for continuous super-resolution. In CVPR, 2023. 2

    2023

  6. [14]

    Densepose: Dense human pose estimation in the wild

    Rıza Alp G ¨uler, Natalia Neverova, and Iasonas Kokkinos. Densepose: Dense human pose estimation in the wild. In CVPR, 2018. 3

    2018

  7. [15]

    Clothflow: A flow-based model for clothed person generation

    Xintong Han, Xiaojun Hu, Weilin Huang, and Matthew R Scott. Clothflow: A flow-based model for clothed person generation. In CVPR, 2019. 2, 3, 5, 11

    2019

  8. [16]

    GANs trained by a two time-scale update rule converge to a local nash equi- librium

    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 equi- librium. In NeurIPS, 2017. 6

    2017

  9. [17]

    Distribu- tion augmentation for generative modeling

    Heewoo Jun, Rewon Child, Mark Chen, John Schulman, Aditya Ramesh, Alec Radford, and Ilya Sutskever. Distribu- tion augmentation for generative modeling. In ICML, 2020. 3

    2020

  10. [18]

    Training generative adver- sarial networks with limited data

    Tero Karras, Miika Aittala, Janne Hellsten, Samuli Laine, Jaakko Lehtinen, and Timo Aila. Training generative adver- sarial networks with limited data. NeurIPS, 2020. 2

    2020

  11. [19]

    StableVITON: Learning semantic corre- spondence with latent diffusion model for virtual try-on

    Jeongho Kim, Guojung Gu, Minho Park, Sunghyun Park, and Jaegul Choo. StableVITON: Learning semantic corre- spondence with latent diffusion model for virtual try-on. In CVPR, 2024. 1, 2, 5, 6, 7, 13

    2024

  12. [20]

    Fine-tuning can distort pre- trained features and underperform out-of-distribution

    Ananya Kumar, Aditi Raghunathan, Robbie Matthew Jones, Tengyu Ma, and Percy Liang. Fine-tuning can distort pre- trained features and underperform out-of-distribution. In ICLR, 2022. 3

    2022

  13. [21]

    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 ECCV,

  14. [22]

    UniHu- man: A unified model for editing human images in the wild

    Nannan Li, Qing Liu, Krishna Kumar Singh, Yilin Wang, Jianming Zhang, Bryan A Plummer, and Zhe Lin. UniHu- man: A unified model for editing human images in the wild. In CVPR, 2024. 2, 3, 14

    2024

  15. [23]

    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. In ICCV, 2023. 1, 2

    2023

  16. [24]

    I2sb: image-to-image schr¨odinger bridge

    Guan-Horng Liu, Arash Vahdat, De-An Huang, Evange- los A Theodorou, Weili Nie, and Anima Anandkumar. I2sb: image-to-image schr¨odinger bridge. In ICML, 2023. 2, 3, 4, 5

    2023

  17. [25]

    Deepfashion: Powering robust clothes recognition and retrieval with rich annotations

    Ziwei Liu, Ping Luo, Shi Qiu, Xiaogang Wang, and Xiaoou Tang. Deepfashion: Powering robust clothes recognition and retrieval with rich annotations. In CVPR, 2016. 2, 3

    2016

  18. [26]

    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 CVPR, 2022. 2, 6

    2022

  19. [27]

    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 Multimedia, 2023. 5, 6

    2023

  20. [28]

    Picture: Photorealistic virtual try-on from unconstrained designs

    Shuliang Ning, Duomin Wang, Yipeng Qin, Zirong Jin, Baoyuan Wang, and Xiaoguang Han. Picture: Photorealistic virtual try-on from unconstrained designs. In CVPR, 2024. 2 9

    2024

  21. [29]

    Maan Qraitem, Kate Saenko, and Bryan A. Plummer. From fake to real: Pretraining on balanced synthetic images to pre- vent spurious correlations in image recognition. In The Eu- ropean Conference on Computer Vision (ECCV), 2024. 8

    2024

  22. [30]

    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. In CVPR, 2022. 11

    2022

  23. [31]

    Towards squeezing-averse virtual try-on via sequential deformation

    Sang-Heon Shim, Jiwoo Chung, and Jae-Pil Heo. Towards squeezing-averse virtual try-on via sequential deformation. In AAAI, 2024. 1, 2, 3, 6, 11, 13

    2024

  24. [32]

    Semantic data augmenta- tion with generative models

    C Shivashankar and Shane Miller. Semantic data augmenta- tion with generative models. In CVPR, 2023. 2

    2023

  25. [33]

    Generative modeling by es- timating gradients of the data distribution

    Yang Song and Stefano Ermon. Generative modeling by es- timating gradients of the data distribution. NeurIPS, 2019. 2

    2019

  26. [34]

    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 Process- ing, 13(4):600–612, 2004. 6

    2004

  27. [35]

    Paint by inpaint: Learning to add image objects by removing them first

    Navve Wasserman, Noam Rotstein, Roy Ganz, and Ron Kimmel. Paint by inpaint: Learning to add image objects by removing them first. arXiv preprint arXiv:2404.18212 ,

    arXiv

  28. [36]

    Towards scalable unpaired virtual try-on via patch-routed spatially- adaptive gan

    Zhenyu Xie, Zaiyu Huang, Fuwei Zhao, Haoye Dong, Michael Kampffmeyer, and Xiaodan Liang. Towards scalable unpaired virtual try-on via patch-routed spatially- adaptive gan. NeurIPS, 2021. 2, 3

    2021

  29. [37]

    GP- VTON: Towards general purpose virtual try-on via collabo- rative 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 collabo- rative local-flow global-parsing learning. In CVPR, 2023. 1, 2, 6, 11, 13

    2023

  30. [38]

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

    Xu Yang, Changxing Ding, Zhibin Hong, Junhao Huang, Jin Tao, and Xiangmin Xu. Texture-preserving diffusion models for high-fidelity virtual try-on. In CVPR, 2024. 1, 2, 5, 6

    2024

  31. [39]

    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. In CVPR, 2024. 3, 5, 6, 7, 13

    2024

  32. [40]

    The unreasonable effectiveness of deep features as a perceptual metric

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

    2018

  33. [41]

    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 CVPR, 2023. 1 10 Appendices Parameter Value Batch Ratio of Synthetic Data 15% Batch Size 32 Image Size 512x5...

    2023

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.