I wanted to reply to my previous autoclip post but it's "locked". However:
Hi all,
I wanted to share my two cents on using AutoClip outside faceswap.
I have been working on other custom AI models not related to faceswap using Python/TF/Keras etc for my own personal projects.
I have noticed a dramatic decrease in training time along with much better generalization in my models using AutoClip:
https://github.com/pseeth/autoclip
From what I can see, while they may not mitigate the NaNs in FS models as well as planned, they could indeed decrease the training time and bring much faster generalization.
From my work outside FS this has led me to always use AutoClip when training new FS models.
If you want examples/logs of my personal models outside FS I can run some tests and provide the data.
AutoClip for the win.
