Mirrors
/
EveryDream2trainer
mirror of https://github.com/victorchall/EveryDream2trainer.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Update PLUGINS.md

This commit is contained in:
Victor Hall 2023-12-20 15:39:57 -05:00 committed by GitHub
parent 83623929ad
commit 1119c2130f
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 1 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
doc/PLUGINS.md
View File

@ -38,4 +38,4 @@ These runs every time the image/caption pair are loaded and passed into the trai
Could be useful for things like customized shuffling algorithms, word replacement/addition/removal, randomization of length, etc.
#### transform_pil_image(self, img:Image)
Could be useful for things like color grading, gamma adjustment, HSL modifications, etc. Note that AFTER this function runs the image is converted to numpy format and normalized (std_dev=0.5, norm=0.5), so normalization is wasted compute. From prior experimentation, all adjustments to the standard normalization are a waste of time. Gamma or curve adjustments are still potentially useful.
Could be useful for things like color grading, gamma adjustment, HSL modifications, etc. Note that AFTER this function runs the image is converted to numpy format and normalized (std_dev=0.5, norm=0.5) per the reference implementation in Stable Diffusion, so normalization is wasted compute. From prior experimentation, all adjustments to this normalization scheme degrade output of the model, thus are a waste of time and have been hardcoded. Gamma or curve adjustments are still potentially useful, as are hue and saturation changes.