|
|
--- |
|
|
license: mit |
|
|
task_categories: |
|
|
- text-to-image |
|
|
- image-classification |
|
|
pretty_name: sanitized-danbooru2023-sqlite |
|
|
size_categories: |
|
|
- 100M<n<1B |
|
|
--- |
|
|
# SQLITE-DB for Danbooru 2023 (until 7110548) |
|
|
This is an cleaned-up version (almost totally recreated) of https://huggingface.co/datasets/KBlueLeaf/danbooru2023-sqlite |
|
|
|
|
|
The previous sqlite database had major defects, especially with tag ids being mismatched, which was causing data to be actually different from server. |
|
|
|
|
|
Note that minor information, such as uploader id, are not fixed. |
|
|
|
|
|
Most of the discrepancy has been detected from id 5139963-6859952. |
|
|
|
|
|
|
|
|
|
|
|
The additional scripts and example files that can be used for modify/ adding post are provided for maintenance. |
|
|
|
|
|
|
|
|
|
|
|
# view-dataset.py offers you to get the information of post: |
|
|
|
|
|
The example post which contained wrong information can be viewed like this: |
|
|
post : https://danbooru.donmai.us/posts/6719955 |
|
|
``` |
|
|
Post 6719955 |
|
|
tag_list_general: 1girl(0), long_hair(24), ribbon(36), simple_background(42), smile(43), solo(44), ahoge(71), open_mouth(90), blonde_hair(116), :d(147), red_eyes(193), blush_stickers(274), upper_body(285), neck_ribbon(376), red_ribbon(380), chibi(437), transparent_background(513), cardigan(2754), looking_afar(3998), white_cardigan(52258), sad_keanu_(meme)(136930) |
|
|
tag_list_copyright: tokino_sora_(old_design)(364562), juufuutei_raden(649634) |
|
|
tag_list_character: regloss_(hololive)(650035) |
|
|
tag_list_meta: (175) |
|
|
tag_list_artist: xt0y4x(525184) |
|
|
``` |
|
|
|
|
|
However, the actual data will show you the difference: |
|
|
``` |
|
|
{"id": 6719955, "difference": [{"tag_list_general": ["virtual_youtuber"], "tag_list_character": ["otonose_kanade"], "tag_list_artist": ["jb_jagbung"], "tag_list_copyright": ["hololive", "hololive_dev_is"]}, {"tag_list_general": ["sad_keanu_(meme)"], "tag_list_character": ["regloss_(hololive)"], "tag_list_artist": ["xt0y4x"], "tag_list_copyright": ["tokino_sora_(old_design)", "juufuutei_raden"]}]} |
|
|
``` |
|
|
|
|
|
The actual tags ids are followed: |
|
|
``` |
|
|
virtual_youtuber(136931) |
|
|
otonose_kanade(650036) |
|
|
jb_jagbung(525185) |
|
|
hololive(364563) |
|
|
hololive_dev_is(649635) |
|
|
``` |
|
|
|
|
|
There were tags added / removed from post, but other than that, there is actual shift on tags - which is not consistent over database. |
|
|
|
|
|
``` |
|
|
tag_list_character: regloss_(hololive)(650035) <-> otonose_kanade(650036) |
|
|
tag_list_artist: xt0y4x(525184) <-> jb_jagbung(525185) |
|
|
tag_list_copyright: tokino_sora_(old_design)(364562), juufuutei_raden(649634) <-> hololive(364563), hololive_dev_is(649635) |
|
|
``` |
|
|
|
|
|
The tag virtual_youtuber is the good difference that we can add to database too. |
|
|
|
|
|
# crawling code is not included. |
|
|
|
|
|
# commit_difference.py offers you to change the database's information based on **difference jsonl** files. |
|
|
|
|
|
You can prepare bunch of jsonl files, which contains the line of json which contains id-difference. |
|
|
The data **must** contain string form of data, not tag id. **The actual danbooru tag id won't match with the database, it is intended to skip bunch of tags which does not have actual post usages.** |
|
|
|
|
|
# fix-tags.py just contains the code which will reflect the actual tag usage count, to tag popularity. |
|
|
|
|
|
# add_post.py includes code to add more recent post data directly into dataset. |
|
|
It contains simple skip schema, if post already exists and it has non-empty tag list, it won't add the post. |
