colab result: Two legs are merged like wearing a dress :-)

[bigboss97]

I'm experimenting on colab with different photos. This is my recent result: You can see in the photos that the legs are spread out. But looks like it can't handle pants when they are too wide.

[piaozhx]

It seems that the cloth_smpl_link mode of digital_type failed in your case (which is used for the skirt case). you can close it by following steps:

1. openassets/configs/deploy.toml
2. change the value of digital_type to smpl (default value is cloth_smpl_link)
3. delete all results and rerun the code

[bigboss97]

Wow, problem fixed! Thank you so much. I downloaded deploy.toml, modified it and transfered it back. It would be great if there's an option in the colab.

[bigboss97]

For some reason there's a black mark flashing around my left arm (right arm when I turn around) all the time: I've tried the same reference video with other source images, with tight sleeves. The result was very clean.

[piaozhx]

The sleeves in your image are too wide compared to the body shape of the estimated SMPL model, so the deep network cannot be trained very well in the step of personalization, which causes the problem of "black mark".

[bigboss97]

I fixed the problem by modifying the sleeves smaller, thanks. Another question, sometimes I got more "cloudy" edges than other even I was using the same input images: Is there a parameter I can adjust?