Virtual TryOn related questions

[whiterose199187]

Hello,

Thanks for this amazing project. I've been trying Virtual Try-On. Few questions:

1. What is the recommended size of input images ref image/mask and target image/mask for best results?
2. I notice that sometimes face of the subject gets slightly altered even though it's not part of target masked image. Why would that happen?

Any other tips to get best results?

Thanks

[XavierCHEN34]

1. The inference size should match the training size, 512x512 for the generated image and human image and 224x224 for DINOv2. Our code has automatically make the crop the resize.
2. Although the faces are not included in the inpainting region, the pixels are still regenerated under the guidance of the given background. If you want the face to be exactly unchanged. Maybe you could use a mask to blend the generation results with the given image.

[XavierCHEN34]

"Other tips to get the best results" might be making some Tryon-related modifications 😂

1. Replacing the HF map with the warped garment.
2. Adding the pose or dense pose of the human to the model.
3. Finetuning on tryon datasets.

You could have a try by yourself, and we are expected to release the tryon specific models in the future.

[tanuki-create]

Hello. Thank you for developing this great tool.

I would like to finetune the model, but any instruction? I mean what I should contain the dataset. The garment pics only?

[XavierCHEN34]

you could observe the dataset like VITON-HD or Dresscode to prepare image pairs of single garment+ try on image

[Siraj-HM]

The face always has imperfections after the swap. How can I leave it unchanged? Also, I found that it has issues with people with dark-colored skin. Have you noticed it too? @XavierCHEN34

[chen-rn]

@XavierCHEN34 How "precise" should the mask be for both the target and reference? Ie. If i'm trying to put a big puffer jacker on a person, what should the mask on the person be like?