TAMF-VTON: Token-Wise Sparse Adaptation for Mask-Free Virtual Try-On

TAMF-VTON is a texture-aware, mask-free framework for virtual try-on that performs high-fidelity image synthesis under unconstrained conditions. It achieves this by using a lightweight Mixture-of-Experts adaptation to inject try-on expertise, a frequency-domain supervision loss to preserve fine textures, and an adaptive inpainting strategy for creating high-quality training data.

Open the full episode

More from AI Radio