You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
This repository has been archived by the owner on Nov 9, 2023. It is now read-only.
I'm using a single face set (data_src) as input for a lot of different outputs (data_dst).
Therefore, I'd like to pretrain a model on just my input faceset (data_src).
To facilitate this, I've made a faceset pack of my data_src, and copied that over the "Pretrain_celebA" set, and then pretrain using my own set.
This allows me to pretrain with my own dataset with better settings than when I'm doing a "normal" train that also includes a data_dst set as less memory is being used, so I can bump the batch size up to 7 when pretraining.
Once my pretrain using my own dataset is up to acceptable levels, I can copy it to a new model and start doing a "normal" train with both data_src and data_dst, only this time my data_src has already been pretrained / copied so starts out with a big advantage.
I'd love to be able to do "single input" training like this a bit easier, so I can train on ONLY data_src, or ONLY data_dst.
This way I can train partial input models / output models and reuse / mix and match later.
I hope you get what I mean :-)
The text was updated successfully, but these errors were encountered:
I'm using a single face set (data_src) as input for a lot of different outputs (data_dst).
Therefore, I'd like to pretrain a model on just my input faceset (data_src).
To facilitate this, I've made a faceset pack of my data_src, and copied that over the "Pretrain_celebA" set, and then pretrain using my own set.
This allows me to pretrain with my own dataset with better settings than when I'm doing a "normal" train that also includes a data_dst set as less memory is being used, so I can bump the batch size up to 7 when pretraining.
Once my pretrain using my own dataset is up to acceptable levels, I can copy it to a new model and start doing a "normal" train with both data_src and data_dst, only this time my data_src has already been pretrained / copied so starts out with a big advantage.
I'd love to be able to do "single input" training like this a bit easier, so I can train on ONLY data_src, or ONLY data_dst.
This way I can train partial input models / output models and reuse / mix and match later.
I hope you get what I mean :-)
The text was updated successfully, but these errors were encountered: