sangyun884/HR-VITON

Implements a unified try-on condition generator that jointly performs clothing warping and segmentation generation with feature fusion, eliminating misalignment artifacts and pixel-squeezing from occluded regions. The two-stage pipeline combines a condition generator with discriminator rejection filtering, followed by a separate image synthesis module trained with mixed-precision (FP16) optimization. Built for high-resolution synthesis on the VITON-HD dataset using PyTorch with modular checkpoints for inference and training.

912 stars. No commits in the last 6 months.