Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Some simple suggestions #1

Open
VoidxHoshi opened this issue Sep 7, 2022 · 4 comments
Open

Some simple suggestions #1

VoidxHoshi opened this issue Sep 7, 2022 · 4 comments
Labels
enhancement New feature or request good first issue Good for newcomers

Comments

@VoidxHoshi
Copy link

First of all, thank you for your work, but I do have some small suggestions.
First, there are too few tags, only 10, it would be better to show all tags that exceed a confidence threshold of 0.5.
Second, for the character judgement, it would be better to add a confidence threshold and only show the character tags if they exceed the threshold (preferably also 0.5). There is not necessarily only one character, and a character may have multiple tags (corresponding to different forms, for example, taihou_(azur_lane) and taihou_(phoenix's_spring_song)_(azur_lane) ) in danbooru.
Third, the calculation of rating is different from the original deepdanbooru, and the result is a bit strange.

@Helloyunho Helloyunho added enhancement New feature or request good first issue Good for newcomers labels Sep 8, 2022
@Helloyunho
Copy link
Member

Hi! Thank you for the suggestions.

First, there are too few tags, only 10, it would be better to show all tags that exceed a confidence threshold of 0.5.

Hmm, but then too many tags might appear and might confuse users.

Second, for the character judgement, it would be better to add a confidence threshold and only show the character tags if they exceed the threshold (preferably also 0.5).

Yeah, that would be good. @onTDB needs to check this (since the API doesn't have character confidence yet).

Third, the calculation of rating is different from the original deepdanbooru, and the result is a bit strange.

This is because we are using a modified version of deepdanbooru with some changes in its neural network.

@VoidxHoshi
Copy link
Author

Hmm, but then too many tags might appear and might confuse users.

Ten tags are too few, and there are generally not many tags that exceed the confidence threshold of 0.5. Most importantly, the 10 tags with the highest confidence levels tend to be too broad, like 1girl and solo.
Broad tags can easily take the top spot, replacing precisely defined tags.
I see this aspect seems to be the front-end code simply take a top 10 tag, should be easy to modify.

@onTDB
Copy link
Contributor

onTDB commented Sep 8, 2022

Hello, VoidxHoshi
Thank you for the suggestions.

Second, for the character judgement, it would be better to add a confidence threshold and only show the character tags if they exceed the threshold (preferably also 0.5).

That's a pretty good suggestion
Currently, we do not store the match for the character, so we need to modify the server to see how much match is out. Previously, when only a portion of the training was performed, most of the character matches were below 0.1.
So I think we have to test whether it's possible and how similar it is if possible.

@onTDB onTDB pinned this issue Sep 8, 2022
@onTDB onTDB unpinned this issue Sep 8, 2022
@Helloyunho
Copy link
Member

Helloyunho commented Sep 8, 2022

So I was thinking of the first enhancement and seems like we can first show only 10 tags and then if there're more tags users can click the Show more... button to see more tags

Check the development site and try it

onTDB added a commit to Ekonomi-moe/Eval-Server that referenced this issue Sep 17, 2022
onTDB added a commit to Ekonomi-moe/Eval-Server that referenced this issue Sep 17, 2022
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
enhancement New feature or request good first issue Good for newcomers
Projects
None yet
Development

No branches or pull requests

3 participants