search

taotaonice / FaceShifter

Try to reproduce FaceShifter
250 stars 62 forks source link

Any training details? #3

Open ainrichman opened 4 years ago

ainrichman commented 4 years ago

How many epochs does it cost to obtain reasonable results?

taotaonice commented 4 years ago

How many epochs does it cost to obtain reasonable results?

I am not sure about that, because I trained it on celeb-a-HQ only, with many modifications during the training process. I guess about 20 epochs.

ainrichman commented 4 years ago

Could you please share me your well trained weights? victorzuo@outlook.com

huboqiang commented 4 years ago

How many epochs does it cost to obtain reasonable results?

I am not sure about that, because I trained it on celeb-a-HQ only, with many modifications during the training process. I guess about 20 epochs.

Hi, I found if I use the current master branch for training, the source face would not be swapped into target after 20 epochs using celeb-a-HQ.

As I found there are many history updates for the code, focusing on revising samp_prob, lambda_id, I just want to know some detailed information about training. For example, which commit do you used for training? And any suggestion about avoiding no-swap? Thanks.

You could also reply to my e-mail, huboqiang@gmail.com . Thanks.

xiefeiwhu commented 4 years ago

How many epochs does it cost to obtain reasonable results?

I am not sure about that, because I trained it on celeb-a-HQ only, with many modifications during the training process. I guess about 20 epochs.

Hi, I found if I use the current master branch for training, the source face would not be swapped into target after 20 epochs using celeb-a-HQ.

As I found there are many history updates for the code, focusing on revising samp_prob, lambda_id, I just want to know some detailed information about training. For example, which commit do you used for training? And any suggestion about avoiding no-swap? Thanks.

You could also reply to my e-mail, huboqiang@gmail.com . Thanks.

I meet the same question, swap face did not work. Any method to solve this problem?

by321 commented 4 years ago

Hi,

Is this project still actively being worked on ? I'd like to see a guide on training. My email is by321@hotmail.com.

Thanks, Bo

TheshowN commented 4 years ago

How many epochs does it cost to obtain reasonable results?

I am not sure about that, because I trained it on celeb-a-HQ only, with many modifications during the training process. I guess about 20 epochs.

Hi, I found if I use the current master branch for training, the source face would not be swapped into target after 20 epochs using celeb-a-HQ. As I found there are many history updates for the code, focusing on revising samp_prob, lambda_id, I just want to know some detailed information about training. For example, which commit do you used for training? And any suggestion about avoiding no-swap? Thanks. You could also reply to my e-mail, huboqiang@gmail.com . Thanks.

I meet the same question, swap face did not work. Any method to solve this problem?

I meet the same question, too. Did you solve it？And how？

TheshowN commented 4 years ago

How many epochs does it cost to obtain reasonable results?

I am not sure about that, because I trained it on celeb-a-HQ only, with many modifications during the training process. I guess about 20 epochs.

Hi, I found if I use the current master branch for training, the source face would not be swapped into target after 20 epochs using celeb-a-HQ.

As I found there are many history updates for the code, focusing on revising samp_prob, lambda_id, I just want to know some detailed information about training. For example, which commit do you used for training? And any suggestion about avoiding no-swap? Thanks.

You could also reply to my e-mail, huboqiang@gmail.com . Thanks.

I meet the same question, too. Did you solve it？And how？