The idea of online garment fitting has been revolutionized by garment transfer technology, which digitally ‘dresses’ a shopper with a selected garment. However, the reliance on self-supervised learning due to the absence of ground-truth data has limited the accuracy of existing methods.

• This paper introduces a novel approach, training garment transfer supervised by knowledge distilled from virtual try-on model performance.
• A multi-phase transfer parsing reasoning model shapes the garment fitting process, learning from try-on models and ground truth inputs.
• The subsequent warping knowledge enables precise garment adjustment according to body shape and content correspondence.
• A supplemental ‘arm regrowth’ task ingeniously fills in skin areas left exposed by garment fitting, enhancing the authenticity of the transfer.

The method presented couples the strengths of virtual try-on models with garment transfer, seemingly bridging the gap between theoretical modeling and practical application. Its implications for online shopping experiences are significant, allowing users to better visualize how clothes might look on them without physical trial, potentially reshaping e-commerce in the fashion industry. Read more.