search

SaltedSlark / Fast-3D-Talking-Face

Drive your metahuman to speak within 1 second.
19 stars 5 forks source link

what is the difference from SaiD? #2

Closed SunnyLi2015 closed 2 weeks ago

SunnyLi2015 commented 2 weeks ago

Hi

May I know the diff between this project and https://github.com/yunik1004/SAiD, from quality side?

Thanks

SaltedSlark commented 2 weeks ago

Better quality to drive metahuman.

SunnyLi2015 commented 2 weeks ago

Better quality to drive metahuman.

ok, so the said model topology is the same, but has been re-trained with better quality, right?

SaltedSlark commented 2 weeks ago

Better quality to drive metahuman.

ok, so the said model topology is the same, but has been re-trained with better quality, right?

What is your meaning of model topology? network backbone or 3d dataset topology?

SaltedSlark commented 2 weeks ago

Better quality to drive metahuman.

ok, so the said model topology is the same, but has been re-trained with better quality, right?

In fact, the data topology is the same as SaiD and the network is more close to https://github.com/psyai-net/SelfTalk_release

SunnyLi2015 commented 2 weeks ago

Better quality to drive metahuman.

ok, so the said model topology is the same, but has been re-trained with better quality, right?

What is your meaning of model topology? network backbone or 3d dataset topology?

thanks for reply, I mean the network backbone. But would like to know if any work is done to improve the quality from other side.

SunnyLi2015 commented 2 weeks ago

Better quality to drive metahuman.

ok, so the said model topology is the same, but has been re-trained with better quality, right?

In fact, the data topology is the same as SaiD and the network is more close to https://github.com/psyai-net/SelfTalk_release

got you, thank you!

SaltedSlark commented 2 weeks ago

Better quality to drive metahuman.

ok, so the said model topology is the same, but has been re-trained with better quality, right?

What is your meaning of model topology? network backbone or 3d dataset topology?

thanks for reply, I mean the network backbone. But would like to know if any work is done to improve the quality from other side.

In my point，to improve the quality you need more 3d data with certain language. And I am working on generating more data by 3d reconstruction on 2d videos.

SaltedSlark commented 2 weeks ago

Better quality to drive metahuman.

ok, so the said model topology is the same, but has been re-trained with better quality, right?

What is your meaning of model topology? network backbone or 3d dataset topology?

thanks for reply, I mean the network backbone. But would like to know if any work is done to improve the quality from other side.

Current work is just combining the to project (said data + selftalk network) to realize faster driving efficiency on metahuman.

SunnyLi2015 commented 2 weeks ago

got it, thanks for the explanation!