Pith. sign in

REVIEW 3 major objections 6 minor 91 references

Virtual try-on is faster and sharper when structure is built in latent space and garment detail is copied in pixel space.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · grok-4.5

2026-07-14 06:06 UTC pith:YJRYZLDO

load-bearing objection Solid first VAR-based VTON with a clean structure/detail split and real speed numbers; the 4× claim and plug-in refiner hold up in the tables, with the matching reliability caveat already flagged by the authors. the 3 major comments →

arxiv 2607.11233 v1 pith:YJRYZLDO submitted 2026-07-13 cs.CV

Structure-Detail Decoupled Autoregressive Generation for Fast and High-Fidelity Virtual Try-On

Lu Yang , Xiaonan Hu , Yanan Li , Daqi Liu , Xiang Bai , Hao Lu This is my paper
classification cs.CV
keywords virtual try-onvisual autoregressive modelsstructure-detail decouplingdense image matchingpixel-space refinementgarment detail preservationlatent-space generation
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

Virtual try-on must keep the person looking like themselves while warping a new garment onto them and preserving logos, text, and fabric patterns. Diffusion models do this in one compressed latent process, which is slow and blurs fine detail. This paper claims the jobs should be split: use a visual autoregressive model to lay out person pose and garment shape quickly in latent space, then use dense image matching to transfer real high-frequency texture from the source garment in pixel space. The resulting two-stage system, STAR-VTON, is at least four times faster than typical diffusion try-on methods while matching or beating their overall quality, and the pixel-space refiner can be bolted onto other try-on models to improve detail. A sympathetic reader cares because the work both demonstrates that autoregressive generation can handle bi-conditional try-on and offers a practical split of representation spaces that recovers details latent compression normally destroys.

Core claim

Different virtual try-on sub-tasks belong in different representation spaces: structural synthesis (garment warping and person layout) is suited to latent space, while fine-grained detail recovery should be done in pixel space via matching-informed transfer. Building on that split, VAR-VTON performs fast latent-space try-on and STAR-VTON's second-stage refiner restores logos, text, and patterns that latent generators lose, yielding a better efficiency-fidelity trade-off.

What carries the argument

STAR-VTON: a two-stage pipeline whose first stage is VAR-VTON (decoupled garment-prefix conditioning plus scale-wise garment-agnostic embedding) and whose second stage is a matching-informed refiner that warps selected high-density garment patches via dense correspondences and injects them through a gated dual-encoder U-Net.

Load-bearing premise

Dense matching between the source garment and the stage-one try-on image stays accurate enough for reliable warping even when the generated image has already lost high-frequency texture.

What would settle it

On a held-out set of logo- or text-heavy garments, measure whether GIM correspondences remain dense and correct after stage-one generation; if matching density collapses or the gated refiner cannot raise CLIP-I on garment crops without pasting artifacts, the structure-detail split fails.

Watch this falsifier — get emailed when new claim-graph text bears on it.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 6 minor

Summary. The paper proposes STAR-VTON, a two-stage virtual try-on framework that decouples latent-space structural synthesis from pixel-space detail recovery. Stage one (VAR-VTON) adapts visual autoregressive next-scale prediction to bi-conditional VTON via (i) decoupled garment-prefix conditioning in self-attention and (ii) scale- and block-wise embedding of garment-agnostic person cues (mask, masked person, dense pose). Stage two uses a matching-informed refiner: GIM dense correspondences warp high-density garment patches into a lightweight dual-encoder U-Net with a learned spatial gate that injects pixel-space textures into the stage-one output. On VITON-HD and DressCode, VAR-VTON reports substantially lower latency/TFLOPs than recent diffusion VTON baselines with competitive FID/KID, while the refiner raises garment-region CLIP-I and is shown as a plug-and-play module on several existing backbones (Tables 1, 8).

Significance. If the reported efficiency–fidelity trade-off holds under transparent evaluation, the work is a useful contribution to VTON: it is among the first systematic applications of next-scale VAR to bi-conditional try-on, supplies a concrete hybrid conditioning recipe (prefix garment + scale-wise structure), and argues for assigning structure vs. texture to latent vs. pixel spaces rather than forcing both through a VAE bottleneck. The plug-and-play refiner results (Table 8) and conditioning ablations (Tables 2–5) are practically valuable. Strengths include clear architectural description, multi-dataset quantitative comparison against recent baselines, wall-clock latency on a fixed GPU, and an explicit limitations discussion of matching fragility (Appendix A.3).

major comments (3)
  1. [Sec. 4.1, Table 1] Sec. 4.1 and Table 1 (CLIP-I*): The primary detail-fidelity metric is CLIP-I on author-constructed “detail-rich subsets” with garment-region crops, but the manuscript does not state subset size, selection criteria, or whether the same crops are used for all methods. Because the Abstract and strongest claim rest on superior detail preservation (and STAR-VTON’s large CLIP-I* jump), this protocol must be fully specified and, ideally, released or re-evaluated on a fixed public protocol so that the efficiency–fidelity claim is reproducible.
  2. [Sec. 3.4, Appendix A.3] Sec. 3.4 and Appendix A.3: The refiner’s validity depends on the claim that GIM still yields accurate dense matches when high-frequency textures are missing from ˆx_tr—an “interesting phenomenon” later qualified by the admission that correspondence density drops when stage-one detail is poor and refinement becomes sensitive to patch count P and thresholds. A load-bearing quantitative analysis is missing (e.g., match density / inlier rate vs. stage-one quality, failure cases, sensitivity of CLIP-I* to P and density threshold). Without this, both the reliability of pixel-space recovery and the plug-and-play narrative (Table 8) remain only partially supported.
  3. [Abstract, Table 1, Table 6] Abstract / Table 1 wording of “without degrading quality”: VAR-VTON improves FID/KID relative to most baselines, but paired CLIP-I* on VITON-HD is below IDM-VTON (0.881 vs. 0.897) before refinement; SSIM/LPIPS in Table 6 are mixed. The claim should be restated more carefully (e.g., competitive overall quality with large latency gains; detail gains primarily from the refiner) so that the efficiency–fidelity trade-off is not overstated for the latent stage alone.
minor comments (6)
  1. [Global] Throughout the extracted text, “VAR” is frequently split as “V AR”; fix typesetting so model names and citations are consistent.
  2. [Fig. 1] Fig. 1 bubble sizes (TFLOPs) and axes should be defined in the caption with the exact measurement protocol (resolution, precision, batch size) matching Table 1.
  3. [Sec. 4.1] Sec. 4.1: clarify how lower-resolution baselines are upscaled (interpolation vs. Real-ESRGAN) and whether that choice affects FID/KID fairness at 1024×768.
  4. [Sec. 3.4] Eqs. (5)–(8): define the Warp operator and how multi-patch compositing handles overlaps; a short pseudocode block would help reproducibility.
  5. [Appendix A.1] Appendix A.1: report the exact number of refiner training epochs/steps and whether the 2,000 detail-rich pairs overlap the VITON-HD test set.
  6. [Table 7] Table 7 vs. Table 1: note that MobileVTON TFLOPs/latency are missing at 512×384; fill or mark N/A consistently.

Circularity Check

0 steps flagged

No circularity: empirical VTON method paper with external metrics and third-party baselines; no derivation that reduces a claimed prediction to its own fitted inputs or self-citation by construction.

full rationale

STAR-VTON / VAR-VTON is a standard empirical computer-vision methods paper. Its load-bearing claims (efficiency–fidelity trade-off, 4× speed, detail recovery, plug-and-play refiner) are supported by wall-clock latency, TFLOPs, FID/KID, CLIP-I*, SSIM and LPIPS measured on public VITON-HD and DressCode splits against third-party diffusion baselines (StableVITON, IDM-VTON, CatVTON, etc.). The architecture (decoupled garment-prefix conditioning, scale-wise garment-agnostic embedding, matching-informed U-Net refiner using GIM) is a design choice trained end-to-end; nothing is defined in terms of the reported metrics, no parameter is fitted on a subset and then “predicted” on a closely related quantity, and no uniqueness theorem or ansatz is imported from the present authors’ prior work. Citations to Infinity (Han et al.) and GIM (Shen et al.) are ordinary tooling references with non-overlapping authorship. Evaluation-subset construction and the matching-reliability assumption are potential correctness or robustness concerns, not circular reductions of the form “Eq. X = Eq. Y by construction.” Hence the derivation chain is self-contained against external benchmarks and circularity score is zero.

Axiom & Free-Parameter Ledger

6 free parameters · 5 axioms · 3 invented entities

The central claim rests on standard generative-model tooling (VQ-VAE latents, next-scale VAR, GIM matching, dual-encoder U-Net) plus several modeling choices and hyperparameters that are not derived from first principles. The structure–detail space split and the reliability of matching on detail-degraded try-on images are the main paper-specific premises; free parameters include training schedules, loss weights, and unspecified matching/patch thresholds that control the refiner.

free parameters (6)
  • AdamW learning rate = 1e-5
    Set to 1e-5 for both stages; not derived, chosen for training stability.
  • Refiner loss weights (λ_mse, λ_p, λ_adv, λ_fft) = 1, 1, 1, 0.01
    Hand-set combination of MSE, perceptual, adversarial, and DFT losses that defines what “detail recovery” optimizes.
  • VAR training schedule (steps and resolutions) = 10k + 15k steps
    10k steps at 512×384 then 15k at 1024×768 on 4×A6000; schedule is empirical.
  • Refiner training set size and selection = 2000 pairs
    2,000 “detail-rich” VITON-HD pairs selected by authors; selection rule not fully formalized.
  • Matching density threshold and patch count P
    Control which garment regions are warped and how they are partitioned; numerical values not reported but load-bearing for refinement quality.
  • CFG on/off choice = w/o CFG (default)
    Authors disable classifier-free guidance by default for speed after a small quality–latency trade-off table.
axioms (5)
  • domain assumption VAE/VQ latent compression irreversibly discards high-frequency garment detail that multi-step latent generators cannot fully restore.
    Stated in Abstract, Fig. 2, and Sec. 1 as the motivation for pixel-space recovery; standard but not proven for every VTON case.
  • ad hoc to paper Structural synthesis (warping, pose, layout) is better handled in latent space while fine texture recovery should be done in pixel space.
    Core design thesis of STAR-VTON (Abstract, Sec. 1); motivates the two-stage split rather than a single end-to-end latent model.
  • ad hoc to paper Dense matching between source garment and detail-degraded try-on image remains accurate enough to produce precise warps.
    Explicitly invoked in Sec. 3.4 as an “interesting phenomenon”; load-bearing for the refiner; partially contradicted by the limitations appendix when synthesis is poor.
  • domain assumption GIM (Shen et al., 2024) provides generalizable dense correspondences suitable for non-rigid garment transfer.
    Used as the correspondence engine without re-deriving matching theory (Sec. 3.4).
  • domain assumption Next-scale visual autoregressive prediction (Infinity/VAR) is a valid backbone for bi-conditional try-on when hybrid garment prefix and scale-wise structural embeddings are added.
    Sec. 3.2–3.3; relies on prior VAR generation quality claims.
invented entities (3)
  • VAR-VTON bi-conditioning stack (decoupled garment prefix + scale-wise garment-agnostic MLPs) no independent evidence
    purpose: Enable garment–person interaction and structural consistency inside multi-scale VAR decoding for VTON.
    New architectural combination relative to cited ControlAR/SCALAR-style controls; validated only by this paper’s ablations.
  • Matching-informed gated pixel-space refiner no independent evidence
    purpose: Selectively inject warped high-frequency garment patches into stage-one outputs via learned spatial gates and match masks.
    Central second-stage module; plug-in results on other models provide some external support but the exact gate design is paper-specific.
  • STAR-VTON two-stage structure–detail decoupled pipeline no independent evidence
    purpose: Assign structure synthesis to latent VAR and detail recovery to pixel matching rather than one joint latent generator.
    Framing and system that the paper’s main claims evaluate as a whole.

pith-pipeline@v1.1.0-grok45 · 19433 in / 3919 out tokens · 45920 ms · 2026-07-14T06:06:46.921654+00:00 · methodology

0 comments
read the original abstract

Virtual try-on (VTON) is a bi-conditional image generation problem that requires not only accurate person preservation but also faithful garment deformation and detail synthesis. Diffusion-based VTON methods can jointly model these factors in a compressed latent space, but suffer from high-frequency detail loss due to inherent latent compression, even with costly multi-step denoising. Recent visual autoregressive (VAR) models offer a promising alternative for high-quality generation with faster inference, yet remain unexplored for VTON due to the lack of effective bi-conditioning mechanisms. To bridge this gap, we first introduce VAR-VTON, a VAR-based VTON model that incorporates garment conditioning and structural guidance for efficient latent-space VTON. Despite its efficacy, latent-space generation still struggles to preserve fine-grained garment details. We argue that different VTON sub-tasks should be addressed in different representation spaces: structural synthesis such as garment warping and person layout is suited to the latent space, whereas fine-grained detail recovery should be tackled in the pixel space. Motivated by this insight, we further propose STAR-VTON, a Two-Stage AutoRegressive framework that builds upon VAR-VTON by decoupling latent-space structural synthesis from pixel-space detail recovery. Our idea is to resort to a matching-informed refiner to establish dense correspondences between the stage-one generation and the source garment to directly map fine-grained pixel-space details. Extensive experiments show that STAR-VTON achieves an impressive efficiency-fidelity trade-off: VAR-VTON runs at least $4\times$ faster than diffusion-based counterparts without degrading quality, and the pixel-space refiner effectively restores fine details and acts as a plug-and-play module that can benefit existing VTON approaches.

Figures

Figures reproduced from arXiv: 2607.11233 by Daqi Liu, Hao Lu, Lu Yang, Xiang Bai, Xiaonan Hu, Yanan Li.

Figure 1
Figure 1. Figure 1: Comparison with the state of the art in terms of detail fidelity and efficiency. Bubble size denotes computational cost (TFLOPs). We evaluate detail preservation using CLIP-I, with higher values indicating better fidelity. Our so￾lutions achieve better fidelity-efficiency trade-off. Given a person image and a garment image, virtual try-on (VTON) aims to synthesize a photorealistic image in which the person… view at source ↗
Figure 2
Figure 2. Figure 2: Comparison between latent-space VTON and our structure-detail decoupled gener￾ation framework. (a) Existing VTON methods rely on compressed latent representations, where high-frequency garment details are inevitably lost during VAE encoding. Consequently, (b) latent￾space generation struggles to faithfully preserve fine-grained garment textures. (c) Our framework leverages the VAR model for efficient holis… view at source ↗
Figure 3
Figure 3. Figure 3: Overview of STAR-VTON. Our method adopts a two-stage framework. The first stage consists of two components: i) decoupled garment prefix conditioning, which fuses garment tokens and image tokens within self-attention while maintaining separate processing pathways thereafter; and ii) scale-wise garment-agnostic representation embedding, where multi-scale garment-agnostic features are progressively integrated… view at source ↗
Figure 4
Figure 4. Figure 4: Qualitative comparison with baselines. Our method faithfully preserves fine-grained gar [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Ablation on the gating strategy in the refiner. Direct concatenation causes noticeable color shifts, while mask blending introduces obvious pasted artifacts. In contrast, the proposed gat￾ing mechanism produces more natural and coherent garment-detail recovery. Zoom in to compare. Ablation on Garment Encoding Module [PITH_FULL_IMAGE:figures/full_fig_p009_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Refiner results under challenging poses with partial occlusion. global structure generated by the original models. We further provide quantitative results in [PITH_FULL_IMAGE:figures/full_fig_p014_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Qualitative comparison on the DressCode dataset. Our method produces realistic gar￾ment deformation across diverse garment categories [PITH_FULL_IMAGE:figures/full_fig_p015_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Compatibility of the proposed refiner with different baseline methods. The proposed refiner consistently improves the preservation of fine-grained garment textures, while maintaining the overall appearance generated by the original models. 15 [PITH_FULL_IMAGE:figures/full_fig_p015_8.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

91 extracted references · 15 linked inside Pith

  1. [1]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Viton: An image-based virtual try-on network , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  2. [2]

    Proceedings of the European Conference on Computer Vision (ECCV) , pages=

    Toward characteristic-preserving image-based virtual try-on network , author=. Proceedings of the European Conference on Computer Vision (ECCV) , pages=

  3. [3]

    Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

    Vtnfp: An image-based virtual try-on network with body and clothing feature preservation , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

  4. [4]

    Proceedings of the IEEE/CVF Iinternational Conference on Computer Vision , pages=

    Clothflow: A flow-based model for clothed person generation , author=. Proceedings of the IEEE/CVF Iinternational Conference on Computer Vision , pages=

  5. [5]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Viton-hd: High-resolution virtual try-on via misalignment-aware normalization , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  6. [6]

    Proceedings of the European Conference on Computer Vision (ECCV) , pages=

    High-resolution virtual try-on with misalignment and occlusion-handled conditions , author=. Proceedings of the European Conference on Computer Vision (ECCV) , pages=

  7. [7]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Towards photo-realistic virtual try-on by adaptively generating-preserving image content , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  8. [8]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Gp-vton: Towards general purpose virtual try-on via collaborative local-flow global-parsing learning , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  9. [9]

    Advances in Neural Information Processing Systems , volume=

    Towards scalable unpaired virtual try-on via patch-routed spatially-adaptive gan , author=. Advances in Neural Information Processing Systems , volume=

  10. [10]

    Communications of the ACM , volume=

    Generative adversarial networks , author=. Communications of the ACM , volume=. 2020 , publisher=

    2020

  11. [11]

    Advances in Neural Information Processing Systems , volume=

    Denoising diffusion probabilistic models , author=. Advances in Neural Information Processing Systems , volume=

  12. [12]

    Advances in Neural Information Processing Systems , volume=

    Diffusion models beat gans on image synthesis , author=. Advances in Neural Information Processing Systems , volume=

  13. [13]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Tryondiffusion: A tale of two unets , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  14. [14]

    Proceedings of the 31st ACM International Conference on Multimedia , pages=

    Taming the power of diffusion models for high-quality virtual try-on with appearance flow , author=. Proceedings of the 31st ACM International Conference on Multimedia , pages=

  15. [15]

    Proceedings of the 31st ACM International Conference on Multimedia , pages=

    Ladi-vton: Latent diffusion textual-inversion enhanced virtual try-on , author=. Proceedings of the 31st ACM International Conference on Multimedia , pages=

  16. [16]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Stableviton: Learning semantic correspondence with latent diffusion model for virtual try-on , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  17. [17]

    Proceedings of the European Conference on Computer Vision (ECCV) , pages=

    Improving diffusion models for authentic virtual try-on in the wild , author=. Proceedings of the European Conference on Computer Vision (ECCV) , pages=. 2024 , organization=

    2024

  18. [18]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    Ootdiffusion: Outfitting fusion based latent diffusion for controllable virtual try-on , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

  19. [19]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Learning flow fields in attention for controllable person image generation , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  20. [20]

    arXiv preprint arXiv:2208.01618 , year=

    An image is worth one word: Personalizing text-to-image generation using textual inversion , author=. arXiv preprint arXiv:2208.01618 , year=

    Pith/arXiv arXiv

  21. [21]

    International Conference on Learning Representations , year=

    CatVTON: Concatenation Is All You Need for Virtual Try-On with Diffusion Models , author=. International Conference on Learning Representations , year=

  22. [22]

    arXiv preprint arXiv:2411.10499 , year=

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

    Pith/arXiv arXiv

  23. [23]

    Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

    Scalable diffusion models with transformers , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

  24. [24]

    International Conference on Learning Representations , year=

    Incorporating Visual Correspondence into Diffusion Model for Virtual Try-On , author=. International Conference on Learning Representations , year=

  25. [25]

    Proceedings of the European Conference on Computer Vision (ECCV) , pages=

    Wear-any-way: Manipulable virtual try-on via sparse correspondence alignment , author=. Proceedings of the European Conference on Computer Vision (ECCV) , pages=. 2024 , organization=

    2024

  26. [26]

    Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

    Adding conditional control to text-to-image diffusion models , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

  27. [27]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    Imagdressing-v1: Customizable virtual dressing , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

  28. [28]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Linking garment with person via semantically associated landmarks for virtual try-on , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  29. [29]

    Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

    Size does matter: Size-aware virtual try-on via clothing-oriented transformation try-on network , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

  30. [30]

    Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision , pages=

    Controlling virtual try-on pipeline through rendering policies , author=. Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision , pages=

  31. [31]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    M&m vto: Multi-garment virtual try-on and editing , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  32. [32]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    High-resolution image synthesis with latent diffusion models , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  33. [33]

    arXiv preprint arXiv:2112.10741 , year=

    Glide: Towards photorealistic image generation and editing with text-guided diffusion models , author=. arXiv preprint arXiv:2112.10741 , year=

    Pith/arXiv arXiv

  34. [34]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    T2i-adapter: Learning adapters to dig out more controllable ability for text-to-image diffusion models , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

  35. [35]

    arXiv preprint arXiv:2308.06721 , year=

    Ip-adapter: Text compatible image prompt adapter for text-to-image diffusion models , author=. arXiv preprint arXiv:2308.06721 , year=

    Pith/arXiv arXiv

  36. [36]

    Controlnet++: Improving conditional controls with efficient consistency feedback: Project page: liming-ai. github. io/controlnet\_plus\_plus , author=. Proceedings of the European Conference on Computer Vision (ECCV) , pages=. 2024 , organization=

    2024

  37. [37]

    Advances in Neural Information Processing Systems , volume=

    UniControl: A Unified Diffusion Model for Controllable Visual Generation In the Wild , author=. Advances in Neural Information Processing Systems , volume=

  38. [38]

    Advances in Neural Information Processing Systems , volume=

    Uni-controlnet: All-in-one control to text-to-image diffusion models , author=. Advances in Neural Information Processing Systems , volume=

  39. [39]

    International Conference on Machine Learning , year=

    Scaling rectified flow transformers for high-resolution image synthesis , author=. International Conference on Machine Learning , year=

  40. [40]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Picture: Photorealistic virtual try-on from unconstrained designs , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  41. [41]

    Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

    Towards multi-pose guided virtual try-on network , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

  42. [42]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Diffusion autoencoders: Toward a meaningful and decodable representation , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  43. [43]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Dress code: High-resolution multi-category virtual try-on , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  44. [44]

    IEEE Transactions on Pattern Analysis and Machine Intelligence , volume=

    Comparing images using the Hausdorff distance , author=. IEEE Transactions on Pattern Analysis and Machine Intelligence , volume=. 1993 , publisher=

    1993

  45. [45]

    IRE Transactions on Information Theory , volume=

    Visual pattern recognition by moment invariants , author=. IRE Transactions on Information Theory , volume=. 1962 , publisher=

    1962

  46. [46]

    Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

    Effective whole-body pose estimation with two-stages distillation , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

  47. [47]

    Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition , pages=

    Densepose: Dense human pose estimation in the wild , author=. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition , pages=

  48. [48]

    Proceedings of the European Conference on Computer Vision (ECCV) , pages=

    Sapiens: Foundation for human vision models , author=. Proceedings of the European Conference on Computer Vision (ECCV) , pages=

  49. [49]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Boow-vton: Boosting in-the-wild virtual try-on via mask-free pseudo data training , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  50. [50]

    Advances in Neural Information Processing Systems , volume=

    Gans trained by a two time-scale update rule converge to a local nash equilibrium , author=. Advances in Neural Information Processing Systems , volume=

  51. [51]

    arXiv preprint arXiv:1801.01401 , year=

    Demystifying mmd gans , author=. arXiv preprint arXiv:1801.01401 , year=

    Pith/arXiv arXiv

  52. [52]

    IEEE Transactions on Image Processing , volume=

    Image quality assessment: from error visibility to structural similarity , author=. IEEE Transactions on Image Processing , volume=. 2004 , publisher=

    2004

  53. [53]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    The unreasonable effectiveness of deep features as a perceptual metric , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  54. [54]

    International Conference on Machine Learning , pages=

    Learning transferable visual models from natural language supervision , author=. International Conference on Machine Learning , pages=

  55. [55]

    arXiv preprint arXiv:1312.6114 , year=

    Auto-encoding variational bayes , author=. arXiv preprint arXiv:1312.6114 , year=

    Pith/arXiv arXiv

  56. [56]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    Film: Visual reasoning with a general conditioning layer , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

  57. [57]

    International Conference on Machine Learning , pages=

    Zero-shot text-to-image generation , author=. International Conference on Machine Learning , pages=

  58. [58]

    arXiv preprint arXiv:2507.15037 , year=

    OmniVTON: Training-Free Universal Virtual Try-On , author=. arXiv preprint arXiv:2507.15037 , year=

    Pith/arXiv arXiv

  59. [59]

    arXiv preprint arXiv:2412.16978 , year=

    PromptDresser: Improving the Quality and Controllability of Virtual Try-On via Generative Textual Prompt and Prompt-aware Mask , author=. arXiv preprint arXiv:2412.16978 , year=

    Pith/arXiv arXiv

  60. [60]

    International Conference on Medical Image Computing and Computer-assisted Intervention , pages=

    U-net: Convolutional networks for biomedical image segmentation , author=. International Conference on Medical Image Computing and Computer-assisted Intervention , pages=. 2015 , organization=

    2015

  61. [61]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Texture-preserving diffusion models for high-fidelity virtual try-on , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  62. [62]

    2017 , eprint=

    Adam: A Method for Stochastic Optimization , author=. 2017 , eprint=

    2017

  63. [63]

    Black Forest Labs , title=

  64. [64]

    arXiv preprint arXiv:2307.01952 , year=

    Sdxl: Improving latent diffusion models for high-resolution image synthesis , author=. arXiv preprint arXiv:2307.01952 , year=

    Pith/arXiv arXiv

  65. [65]

    arXiv preprint arXiv:1711.05101 , year=

    Decoupled weight decay regularization , author=. arXiv preprint arXiv:1711.05101 , year=

    Pith/arXiv arXiv

  66. [66]

    2021 , publisher=

    Fashion, Performance, and Performativity: The Complex Spaces of Fashion , author=. 2021 , publisher=

    2021

  67. [67]

    Proceedings of the IEEE International Conference on Computer Vision , pages=

    Learning the latent" look": Unsupervised discovery of a style-coherent embedding from fashion images , author=. Proceedings of the IEEE International Conference on Computer Vision , pages=

  68. [68]

    2004 , publisher=

    Clothing appearance and fit: Science and technology , author=. 2004 , publisher=

    2004

  69. [69]

    Annual Review of Psychology , volume=

    Visual aesthetics and human preference , author=. Annual Review of Psychology , volume=. 2013 , publisher=

    2013

  70. [70]

    arXiv preprint arXiv:2405.18172 , year=

    AnyFit: Controllable Virtual Try-on for Any Combination of Attire Across Any Scenario , author=. arXiv preprint arXiv:2405.18172 , year=

    Pith/arXiv arXiv

  71. [71]

    International Conference on Learning Representations , year=

    CtrLoRA: An Extensible and Efficient Framework for Controllable Image Generation , author=. International Conference on Learning Representations , year=

  72. [72]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Mobile-VTON: High-Fidelity On-Device Virtual Try-On , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  73. [73]

    Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

    Any2anytryon: Leveraging adaptive position embeddings for versatile virtual clothing tasks , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

  74. [74]

    Advances in Neural Information Processing Systems , volume=

    Visual autoregressive modeling: Scalable image generation via next-scale prediction , author=. Advances in Neural Information Processing Systems , volume=

  75. [75]

    arXiv preprint arXiv:2406.09750 , year=

    Controlvar: Exploring controllable visual autoregressive modeling , author=. arXiv preprint arXiv:2406.09750 , year=

    Pith/arXiv arXiv

  76. [76]

    arXiv preprint arXiv:2410.04671 , year=

    Car: Controllable autoregressive modeling for visual generation , author=. arXiv preprint arXiv:2410.04671 , year=

    Pith/arXiv arXiv

  77. [77]

    arXiv preprint arXiv:2501.18993 , year=

    Visual autoregressive modeling for image super-resolution , author=. arXiv preprint arXiv:2501.18993 , year=

    Pith/arXiv arXiv

  78. [78]

    International Conference on Learning Representations , year=

    EchoGen: Generating Visual Echoes in Any Scene via Feed-Forward Subject-Driven Auto-Regressive Model , author=. International Conference on Learning Representations , year=

  79. [79]

    International Conference on Learning Representations , year=

    ControlAR: Controllable Image Generation with Autoregressive Models , author=. International Conference on Learning Representations , year=

  80. [80]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    Scalar: Scale-wise controllable visual autoregressive learning , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

Showing first 80 references.