Garment De-Warping for Virtual Try-on in the Wild

Image-based virtual try-on aims to replace the clothing on a model in a target image with specified garments. However, most existing virtual try-on techniques are designed based on flat-laying clothing images, making it challenging to adapt to the complexities of clothing in real-world environments. To address this issue, we propose a new task—garment dewarping, which aims to convert distorted garment images in natural scenes into flat-laying images. By integrating this task with existing virtual try-on methods, we can achieve virtual try-on in real-world environments. To this end, we design a Garment De-warping Network (GDN) based on a dual U-Net structure diffusion model. Experimental results demonstrate that our proposed GDN is highly compatible with existing virtual try-on methods, significantly enhancing their ability in the wild.