Skip to content

Repository files navigation

VTON Demo

Gradio Interactive app for virtual try-on

To run this app locally, clone repo and do the following:

conda create --name vton python==3.8
pip install -r requirements.txt
python app.py

Performance Comparison

Method SSIM (↑) FID (↓)
Wardrobe-GAN 0.740 47.34
VITON 0.783 55.71
CP-VTON 0.745 24.43
ACGPN 0.845 16.64
Ours 0.886 13.46

Evaluation Metrics

SSIM (Structural Similarity Index Measure)

SSIM measures the structural similarity between the generated try-on image and the ground truth image. In virtual try-on systems:

  • Range: 0 to 1 (higher is better)
  • Purpose: Evaluates how well the generated image preserves structural details, textures, and overall visual quality
  • Importance: Critical for VTON as it ensures the clothing item maintains its realistic appearance and proper fit on the person
  • What it captures: Luminance, contrast, and structural information between images

FID (Fréchet Inception Distance)

FID measures the quality and diversity of generated images by comparing feature distributions between real and generated images:

  • Range: 0 to ∞ (lower is better)
  • Purpose: Evaluates the overall realism and quality of the generated try-on results
  • Importance: Ensures that the virtual try-on images are indistinguishable from real photos and maintains natural appearance
  • What it captures: Feature-level similarity using deep neural network representations, capturing both quality and diversity

Why These Metrics Matter for VTON

  • SSIM ensures the clothing details, patterns, and textures are preserved accurately during the virtual fitting process
  • FID guarantees that the final try-on image looks realistic and natural, avoiding artifacts or unnatural distortions
  • Together, they provide a comprehensive evaluation of both structural fidelity and perceptual quality in virtual try-on systems

Our method achieves state-of-the-art performance with the highest SSIM (0.886) and lowest FID (13.46), demonstrating superior quality in both structural preservation and realistic image generation.

About

No description, website, or topics provided.

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages