Regarding Virtual Try-on

[Deepak2405]

hi, does this technique for Virtual Tryon where we give input person and reference cloth? Further, can it work on complicated textures for Virtual tryon?

[EndyWon]

Sorry for the late reply. For Virtual Tryon, we use the semantic maps to filter the results so as to better preserve the background areas. Specifically, you can transfer the texture (reference cloth) to the region of content_sem (mask of the input person's cloth). Then, use content_sem as the mask to paste the generated result to the corresponding masked region of the content image (input person). Our method is able to preserve the main structure/style of the complicated textures, but may not work very well for geometric textures with strict structure requirements.