search

yfeng95 / DECA

DECA: Detailed Expression Capture and Animation (SIGGRAPH 2021)
Other
2.06k stars 417 forks source link

[Question] The magnitude and change of loss convergence during the training process. #184

Open JSHZT opened 1 year ago

JSHZT commented 1 year ago

Excellent work! your work inspires me. Now I am trying to train this model. I am confused during the rough training stage. The calculation of identity loss has been maintained at a value of around 0.15 without dropping. According to your paper, this loss can help learn shape, and the result of my training It's the shape that I didn't learn successfully. So I want to ask, what is the normal behavior of this loss during training. BTW, What is the magnitude of the convergence of each loss to the final?

lyingers commented 1 year ago

可以加个联系方式吗？我也在复现这项工作。

JSHZT commented 1 year ago

wechat: hzt3016