.PNG Transparency and Fileword Accuracy #3467
Unanswered
iosiflazotis
asked this question in
Q&A
Replies: 0 comments
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
-
Good day. I have several questions.
Please note that my questions have hypernetworks as their context.
1.) When training a hypernetwork, is it a good idea to use images with cropped transparent backgrounds (in.png format)? I'm attempting to determine if this will help the AI learn it more effectively.
2.) Should filewords that are not fully visible in the image be omitted? For instance, if I uploaded a picture of an anime character in which half of a bed is visible in the background, and deepbooru correctly identifies it (bed). By the way, I am referring to the preprocessing stage. In fact, it was for this reason that I pondered question 1 in the first place: whether it would be preferable to simply remove the image elements I don't want the AI to learn.
3.) I have already attempted what I inquired about in number 1, which is why I want to know how it went for others. Preprocessing actually converts the transparent backgrounds of .png images to solid black, rendering them non-transparent. And based on my previous tests, this has a significant impact on the learning rate or how the system learns the images. Why is this? What does the AI see when the background is transparent as opposed to when the background is solid black?
I may have more questions, but these are the ones that immediately come to mind.
Beta Was this translation helpful? Give feedback.
All reactions