Hacker News new | past | comments | ask | show | jobs | submit login

Should have noticed your comment first; I did the same. Thankfully nothing too explicit in the initial page of search results (no images, etc)



gwern keeps renaming the dataset so I don't know what to call it!

https://www.gwern.net/Danbooru2021

It is a partially (highly) NSFW dataset though, which is probably the only way to get so much accurate volunteer tagging.


I don't rename it, I release a new updated dataset. However, each dataset is a superset of the previous one, so you can always reconstruct the old one and thus 'Danbooru2018', 'Danbooru2019', 'Danbooru2020', 'Danbooru2021' etc have exact well-defined meanings that never change, and if you don't happen to have a copy of Danbooru2018 sitting around, you can just download Danbooru2021, unpack the Danbooru2018 metadata from the archive tarball & delete the post-Danbooru2018 images, and you now have a bit-for-bit copy of Danbooru2018.

Also, if you want to just discuss the general concept of boorus (there are many beyond Danbooru), it'd probably be better to invoke Safebooru https://safebooru.org/ which is what it sounds like.




Consider applying for YC's Spring batch! Applications are open till Feb 11.

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: