search

rubenvillegas / cvpr2018nkn

Tensorflow implementation of the CVPR 2018 paper: Neural Kinematic Networks for Unsupervised Motion Retargetting
https://sites.google.com/umich.edu/nik
Other
240 stars 42 forks source link

Is it necessary to train for motion reatarget #15

Closed wuxiaolianggit closed 4 years ago

wuxiaolianggit commented 4 years ago

The number of bone points is the same, but the proportion is not the same. I can directly copy the displacement and rotation vector of character a to the corresponding node of character B. in this way, the same effect can be achieved without using deep learning. Do you think my understanding is correct? @jimeiyang

rubenvillegas commented 4 years ago

If I understand correctly, you may be referring to the "Copy input quaternions and velocities" baseline in Table 1, and in the supplementary video. Please let me know if this is right.

wuxiaolianggit commented 4 years ago

Your understanding is correct, but I read your paper and found that the effect of hand reproduction is not very good. How do you think it should be improved? @jimeiyang

rubenvillegas commented 4 years ago

You can probably see in the paper that we do not handle full hands. We merely handle the joint named hand which does not include fingers. Therefore, hand is a leave joint and there is nothing really constraining it other than the twist term we use. If you want more accurate hand movements, I assume you would have to handle everything up to the fingers.