freeze_layers changed from efficientnetb4 to keras_encoder

If training is failing to start, and you are not receiving an error message telling you what to do, tell us about it here


Forum rules

Read the FAQs and search the forum before posting a new topic.

This forum is for reporting errors with the Training process. If you want to get tips, or better understand the Training process, then you should look in the Training Discussion forum.

Please mark any answers that fixed your problems so others can find the solutions.

Locked
User avatar
damingxing
Posts: 23
Joined: Mon Jun 13, 2022 8:55 pm
Has thanked: 6 times

freeze_layers changed from efficientnetb4 to keras_encoder

Post by damingxing »

Hello, I was following Phase-A (stojo preset).

In the middle of the training, I updated a few parameters in the "train.ini": allow_growth, mixed_precision, eye_multiplier, and mouth_multiplier . However, I did not change the freeze_layers from efficientnetb4 to keras_encoder. After I restarted the training, I got the log like this:

log:
07/27/2022 15:17:18 INFO Loading data, this may take a while...
07/27/2022 15:17:18 INFO Loading Model from Phaze_A plugin...
07/27/2022 15:17:19 INFO Config item: 'allow_growth' has been updated from 'False' to 'True'
07/27/2022 15:17:19 INFO Config item: 'mixed_precision' has been updated from 'False' to 'True'
07/27/2022 15:17:19 INFO Config item: 'eye_multiplier' has been updated from '3' to '25'
07/27/2022 15:17:19 INFO Config item: 'mouth_multiplier' has been updated from '2' to '10'
07/27/2022 15:17:19 INFO Config item: 'freeze_layers' has been updated from '['efficientnetb4']' to '['keras_encoder']'

I still want to use the efficientnetb4. May I know what settings might have changed the freeze_layers from efficientnetb4 to keras_encoder. In the traini.ini, enc_architecture is still set to efficientnet_b4.

Thank you.

User avatar
torzdf
Posts: 2649
Joined: Fri Jul 12, 2019 12:53 am
Answers: 159
Has thanked: 128 times
Been thanked: 623 times

Re: freeze_layers changed from efficientnetb4 to keras_encoder

Post by torzdf »

Ignore it. It will always output that. It's because the config part of Faceswap is central to the whole code and Phaze-A is just a plugin. Because there is no part of your model actually called "Keras Encoder" (in your case it is called "efficientnetb4") the code needs to switch out the name "keras encoder" for the actual encoder that was used for the model. Nothing has actually changed there.

My word is final

User avatar
damingxing
Posts: 23
Joined: Mon Jun 13, 2022 8:55 pm
Has thanked: 6 times

Re: freeze_layers changed from efficientnetb4 to keras_encoder

Post by damingxing »

thank you!!

Locked