FabricDiffusion – SIGGRAPH Asia 2024

About the Project

FabricDiffusion introduces a method for transferring fabric textures from a single clothing image to 3D garments of arbitrary shapes. Leveraging a denoising diffusion model trained on a large-scale synthetic dataset, the system generates distortion-free, tileable textures that integrate seamlessly into Physically-Based Rendering (PBR) pipelines.

The method focuses on preserving fine texture details even under occlusions, distortions, or complex poses, enabling realistic relighting and visualization of garments under various lighting conditions.

For more technical details, visit the official project page:
https://humansensinglab.github.io/fabric-diffusion/

Project Video

My Role

I contributed to this project by creating and optimizing the entire visual content used in the publication and website presentation for SIGGRAPH Asia 2024. My specific responsibilities included:

• Adapting and optimizing texture images provided by the research team for real-time use in Unreal Engine 5.
• Creating master materials and material instances for consistent rendering.
• Building the real-time visualization scene inside Unreal Engine 5.
• Rendering high-quality images and videos for the official website and publication.
• Balancing visual fidelity and performance for efficient visualization and presentation.