Closed Nisekoixmy closed 2 years ago
Hi Nisekoi, data.skin_input loads _skin.txt here. The _skin.txt files are generated here.
Specifically, suppose we consider K nearest bones and V vertices, skin_input has shape V*8K. Each bone-vertex pair has 8-d input, including start position (3-d), end position (3-d) of the bone, 1/geodesic_distance (1-d) and whether this bone is a "virtual bone" (1-d. "Virtual bone" is created for each leaf joint since we want the leaf joints to have skinning weights as well.)
Hi Nisekoi, data.skin_input loads _skin.txt here. The _skin.txt files are generated here.
Specifically, suppose we consider K nearest bones and V vertices, skin_input has shape V*8K. Each bone-vertex pair has 8-d input, including start position (3-d), end position (3-d) of the bone, 1/geodesic_distance (1-d) and whether this bone is a "virtual bone" (1-d. "Virtual bone" is created for each leaf joint since we want the leaf joints to have skinning weights as well.)
Thank you!After diving into your cide, i got the idea of it. But thank you for the quick reply. I have another question which is related to the drop out of edges. I read your paper and found that when training the skinnet, you mentioned about dropping out edges. However, after investigation, i can't find it. May i ask you to point it out for me.?I'm really interested in your work and want to build some knowledge in this field. Thank you again.
Edge dropout is implemented off-line. When generating geodesic edges here, we randomly choose a subset from the range.
Thank you! I will close this issue. It is clear to me now.
Hi Zhan Xu,
Thanks for providing this impressive work! May I know what is the data.skin_input represents? From what I understood, I think it is composed by (1/geodesic_distance). Is that correct? Could you please explain it a little more?
Best regards, Nisekoi.