When a shopper chooses garments online, garment transfer technology wears the garment from the model image onto the shopper's image, allowing the shopper to decide whether the garment is suitable for them. As garment transfer leverages wild and cheap person image as garment condition, it has attracted tremendous community attention and holds vast commercial potential. However, since the ground truth of garment transfer is almost unavailable in reality, previous studies have treated garment transfer as either pose transfer or garment-pose disentanglement, and trained garment transfer in self-supervised learning, yet do not cover garment transfer intentions completely. Therefore, the training supervising the garment transfer is a rock-hard issue. Notably, virtual try-on technology has exhibited superior performance using self-supervised learning. We supervise the garment transfer training via knowledge distillation from virtual try-on. Specifically, we first train the transfer parsing reasoning model at multi-phases to provide shape guidance for downstream tasks. The transfer parsing reasoning model learns the response and feature knowledge from the try-on parsing reasoning model and absorbs the hard knowledge from the ground truth. By leveraging the warping knowledge from virtual try-on, we estimate a progressive flow to precisely warp the garment by learning the shape and content correspondence. To enhance transfer realism, we propose a well-designed arm regrowth task to infer exposed skin pixel content. Experiments demonstrate that our method has state-of-the-art performance in transferring garments between person compared with other virtual try-on and garment transfer methods.

利用虚拟试穿技术和自我监督学习，我们提出了一种新的服装转换方法，通过知识蒸馏从虚拟试穿中提供形状引导，并利用逐步流来准确地转换服装，同时使用一个设计良好的手臂再生任务来增强转换的逼真度。实验证明，与其他虚拟试穿和服装转换方法相比，我们的方法在人际服装转换方面表现出优越的性能。

一种由虚拟试穿模型的精简知识监督的新型服装转移方法