Overview
HR-VITON is an open-source PyTorch implementation for high-resolution virtual try-on, addressing misalignment and occlusion issues. It proposes a novel try-on condition generator as a unified module for warping and segmentation generation. A feature fusion block facilitates information exchange, avoiding misalignment and pixel-squeezing artifacts. Discriminator rejection filters incorrect segmentation map predictions. The model is trained and evaluated using the VITON-HD dataset, demonstrating superior performance in handling misalignment and occlusion compared to baselines. Its architecture facilitates synthesizing realistic images of individuals wearing different clothing items, solving issues in traditional methods that separate warping and segmentation stages.