Testing on a different garment template (different topology)

[Amna-Elsaqa]

Hello,

Is there a way to make the model work on a different 3D garment mesh? If yes, does it have to share the same topology as the ones used in this repo? When we tried it on our own templates, it gave a very weird output.

Thank you.

[dancasas]

Hi, unfortunately SNUG is train for specific garment (and topology). If you want to use a different garment or topology, it would need to be retrained. You can check out follow-up works such as HOOD for topology agnostic methods.

[Amna-Elsaqa]

@dancasas Thank you so much for your help! I'll take a look at HOOD as you advised. However, if we were to retrain SNUG to work on a different topology, could you please explain to me how we should go about that process? I'd really appreciate your help and thanks again.

[dancasas]

We released the training losses used to train SNUG but, unfortunately, we were not allowed to release the full training script. Nonetheless, some people have re-implemented it e.g., https://github.com/zhiqiang-git/pytorch-snug or HOOD, but their results differ slightly to what we report.