Outfit Anyone in the Wild addresses the limitations of current virtual try-on methods, which struggle with diverse poses and complex environments found in street photography. It offers high-quality try-on results for real-world scenarios, balancing user identity retention with clothing detail accuracy, targeting users and developers in e-commerce and fashion tech.

How It Works

The method models both the user and garment using a pre-trained human body reconstruction large model. It then deforms the human body representation in a parameter space to match the user's pose and figure. Clothing appearance is integrated into this parametric model, deforming naturally with the body and adhering to physical laws for a harmonious fit. A final detect-and-refine network addresses any discordant elements in the generated image.

Quick Start & Requirements

• Install/Run: Via ModelScope or OpenXLab.
• Prerequisites: Likely requires significant GPU resources and specific deep learning frameworks (e.g., PyTorch, TensorFlow) based on common practices for such models. Specific dependencies are not detailed in the README.
• Resources: Training or running this model likely demands substantial VRAM and processing power.
• Links:
  • ModelScope: https://www.modelscope.cn/studios/selfitcamera/OutfitAnyoneInTheWild/summary
  • OpenXLab: https://openxlab.org.cn/apps/detail/jiangxiaoguo/OutfitAnyone-in-the-Wild
  • Website: heybeauty.com (mentioned for free try-on)

Highlighted Details

• Claims to solve hand generation issues, resulting in non-misshapen hands.
• Incorporates skin color matching for improved inclusivity.
• Focuses on user facial ID consistency and robustness across different poses.
• Mentions a large-scale 3D human body dataset collection costing $750,000.

Maintenance & Community

• An API for fast try-on has been released.
• A Discord group is available for API usage guidance.
• The project acknowledges contributions from various research areas and tools like HuggingFace and ModelScope.

Licensing & Compatibility

• License details are not explicitly stated in the provided README excerpt. Compatibility for commercial use or closed-source linking is therefore undetermined.

Limitations & Caveats

The README does not detail specific limitations, unsupported platforms, or known bugs. The project appears to be actively developed with recent updates addressing key features like hand generation and skin tone matching.