search

HongwenZhang / PyMAF-X

[TPAMI 2023] PyMAF-X: Towards Well-aligned Full-body Model Regression from Monocular Images
https://www.liuyebin.com/pymaf-x
Other
210 stars 28 forks source link

hi, why the model is large to 1 GB? #1

Closed lucasjinreal closed 2 years ago

lucasjinreal commented 2 years ago

I look inside the code, seems didn't use transformer, why the ckpt so big?

HongwenZhang commented 2 years ago

Hi, the model checkpoint contains three subnetworks for body, hand, and face reconstruction. Moreover, each subnetwork also contains a self-attention module (which includes lots of parameters).

lucasjinreal commented 2 years ago

@HongwenZhang does the attention module using bert as encoder?

HongwenZhang commented 2 years ago

yes