NoW Challenge

[jylovec]

I want to evaluate your results on NoW Challenge. But i meet some problem about getting the 7 3d landmarks which are simply abtained by using index of 68 landmarks from 3d vertex. I don't if it‘s properly.

[wang-zidu]

Hello, I am not familiar with the NoW Challenge. As you mentioned, if it requires some landmarks as initial registration, then extracting directly from existing annotations (such as 68 landmarks) is definitely the easiest way. However, based on my experience, the selection of initial landmarks is actually very important and can affect the final result. Same reconstruction with different mesh topologies might also make a difference, such as using the full head region or just the face region (see issue).

Additionally, I recall you previously mentioned wanting to obtain some annotations for expression dataset. Recently, I have been attempting to deform 3D meshes to existing expression faces to construct a dataset similar to dad-3dheads. However, as you can see, some issues also require video demos and I am also working on some other projects. These all take time to handle, and I cannot guarantee that I will be able to do everything. If there are any updates, I will let you know.

[jylovec]

Thanks for your reply.Since you are busy with many other works, don't know if you have plan to release new benchmark in your paper.

[jylovec]

I have tested your results on NoW Validation set using both face region model and full head model, which shows that full head model has much lower reconstrction error.

[jimydavis]

@jylovec do you mind publishing your validation results on the NoW challenge? Also did you try the non-metrical evaluation or the metrical evaluation?

Thank you

[jimydavis]

@wang-zidu reference: https://now.is.tue.mpg.de/