Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

AlignVTOFF: Texture-Spatial Feature Alignment for High-Fidelity Virtual Try-Off

Yihan Zhu, Mengying Ge

TL;DR

AlignVTOFF is a novel parallel U-Net framework built upon a Reference U-Net and Texture-Spatial Feature Alignment, enabling robust modeling of deformation while retaining complex structured patterns and improving structural realism and high-frequency detail fidelity.

Abstract

Virtual Try-Off (VTOFF) is a challenging multimodal image generation task that aims to synthesize high-fidelity flat-lay garments under complex geometric deformation and rich high-frequency textures. Existing methods often rely on lightweight modules for fast feature extraction, which struggles to preserve structured patterns and fine-grained details, leading to texture attenuation during generation.To address these issues, we propose AlignVTOFF, a novel parallel U-Net framework built upon a Reference U-Net and Texture-Spatial Feature Alignment (TSFA). The Reference U-Net performs multi-scale feature extraction and enhances geometric fidelity, enabling robust modeling of deformation while retaining complex structured patterns. TSFA then injects the reference garment features into a frozen denoising U-Net via a hybrid attention design, consisting of a trainable cross-attention module and a frozen self-attention module. This design explicitly aligns texture and spatial cues and alleviates the loss of high-frequency information during the denoising process.Extensive experiments across multiple settings demonstrate that AlignVTOFF consistently outperforms state-of-the-art methods, producing flat-lay garment results with improved structural realism and high-frequency detail fidelity.

AlignVTOFF: Texture-Spatial Feature Alignment for High-Fidelity Virtual Try-Off

TL;DR

AlignVTOFF is a novel parallel U-Net framework built upon a Reference U-Net and Texture-Spatial Feature Alignment, enabling robust modeling of deformation while retaining complex structured patterns and improving structural realism and high-frequency detail fidelity.

Abstract

Virtual Try-Off (VTOFF) is a challenging multimodal image generation task that aims to synthesize high-fidelity flat-lay garments under complex geometric deformation and rich high-frequency textures. Existing methods often rely on lightweight modules for fast feature extraction, which struggles to preserve structured patterns and fine-grained details, leading to texture attenuation during generation.To address these issues, we propose AlignVTOFF, a novel parallel U-Net framework built upon a Reference U-Net and Texture-Spatial Feature Alignment (TSFA). The Reference U-Net performs multi-scale feature extraction and enhances geometric fidelity, enabling robust modeling of deformation while retaining complex structured patterns. TSFA then injects the reference garment features into a frozen denoising U-Net via a hybrid attention design, consisting of a trainable cross-attention module and a frozen self-attention module. This design explicitly aligns texture and spatial cues and alleviates the loss of high-frequency information during the denoising process.Extensive experiments across multiple settings demonstrate that AlignVTOFF consistently outperforms state-of-the-art methods, producing flat-lay garment results with improved structural realism and high-frequency detail fidelity.
Paper Structure (29 sections, 4 equations, 7 figures, 3 tables)

This paper contains 29 sections, 4 equations, 7 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Virtual Try-On and Virtual Try-Off
  4. Latent Diffusion Models
  5. Proposed Method
  6. Reference U-Net
  7. Denoising U-Net
  8. Texture-Spatial Feature Alignment
  9. Loss Function
  10. Comparison with Existing Architectures
  11. Experiment and Analysis
  12. Datasets
  13. VITON-HD
  14. DressCode
  15. Metrics
  16. ...and 14 more sections

Figures (7)

  • Figure 1: VTOFF results on (a) common garment and (b) garment with complex pattern. Existing methods struggle to preserve the complex structured patterns, while ours ensures consistent and faithful generation.
  • Figure 2: Overview of the AlignVTOFF framework. It mainly consist of a trainable Reference U-Net and a frozen Denoising U-Net. The Reference U-Net extract features from the clothed-person image and inject it into the Denoising U-Net via TSFA.
  • Figure 3: User study results on VITON-HD in terms of R2G, G2R and Jab metric. Higher values in these three metrics indicate better performance.
  • Figure 4: Qualitative comparison on results with different coefficient $\lambda$.
  • Figure 5: Quantitative analysis of the Denoising U-Net freezing strategy on VITON-HD.
  • ...and 2 more figures