CRISTAL-3DSAM/Sparse2Dense

about the training lable of motion3dgan

Opened this issue · 3 comments

UchihaBarry commented

after the third part of Readme.3) transform landmarks to SRVF data with SRVF_Process/create_SRVF_COMA.m function.
we can get the srvf coding of all groups, i wonder how can i get the lable of them while the training of SRVF_WGAN not only needs the srvf, but also the lable.
i wonder if the intensity of create_SRVF_COMA.m function. should be the lable?

UchihaBarry commented

i wonder why the D_err and G_err went nan after 9 epoches of training

pixiaonuo commented

i wonder why the D_err and G_err went nan after 9 epoches of training

I meet the same problem. Start running D_ Err and G_ Err is nan. Have you solved this problem yet？

UchihaBarry commented

i just start training for 5times and finally trained the model. However, the model cant generate good srvf coding that can be decoded by the S2D decoder