This is a snapshot of training image dataset for general torso components model AA
(a part of BOORU YOLO project, **inspect README for details**) as of middle 2025.
This model define scene scale, character(-s) position and pose and can distinguish naked body.
Directory structure and label format default for [Ultralitics](https://github.com/ultralytics/ultralytics)
File naming refer to source image on imageboard and provides dispersion of similar images.
**138.816 image+labels pairs** (no background images) clustered to 10+10 uniform ZIPs.
Image longer side moslty within 800..1600px
**This is not an exact dataset for AA22 model included,** some minor cleanup and adjustments applied after.
Due to project scope and classes nature this dataset :
- focused on anime_art / CG / cartoon images, no real life
- has a substantial share of furry incl. non-humanoid creatures (dragons, pony etc)
- is certainly NSFW (but not contain straight sex scenes, gaping and bodily fluids limited)
Some SFW subset of images from here may be browsed (or even directly used) on [Ultralitics HUB](https://hub.ultralytics.com/datasets/W1NNLLAb9HH7WvWj1nwP)
Actual project description, AA model history and other models placed on github.
Source art datasets are :
- [BOORU_CHARS_2021](https://nyaa.si/view/1384820) 1.593.429 images 472 GB topic starter 1280 px
- [BOORU_CHARS_2015](https://nyaa.si/view/1468367) 463.873 images 148 GB retrospective
- [BOORU_CHARS_2022](https://nyaa.si/view/1547662) 705.467 images 191 GB newcomers
- [BOORU_CHARS_2023](https://nyaa.si/view/1740396) 1.153.513 images 302 GB last of 1280/1024/1280 px
- [BOORU_CHARS_2024](https://nyaa.si/view/1927862) 1.260.629 images 583 GB 2560/1920/2480 px
- [BOORU_CHARS_2025](https://nyaa.si/view/2004380) 896.142 images 440 GB 2560/1920/2480 px
- [BOORU_ECCHI](https://sukebei.714848.xyz/view/4284975) 515.850 erotic 115 GB archive 800..1920 px
- [BOORU_ECCHI_2025](https://sukebei.714848.xyz/view/4369314) 52.147 erotic 22 GB news 1280..2560 px

Model run over training pictures results (BTW - dataset statistics by objects).
FURRY - got from e621 and tbib (crossposts from e621), ANIME - all other imageboards.

Model classes are :
0 - head = anime pretty girl and not only
1 - bust = torso part from collarbone center to pair of covered breasts
2 - boob = bust with no bra, nipples mostly visible, generally NSFW
3 - shld = shoulder and maybe one breast viewed mostly in profile, exactly rear view excluded
4 - sideb = uncovered version of shld, with nipples or other NSFW visual marks
5 - belly = from belly button to hips half (stocking line), knees below belly, mostly covered
6 - nopan = no panty-like clothes on bikini area (regardless of censoring), evidently NSFW belly
7 - butt = buttock area visible at least partially from behind, more or less covered, standing or sitting
8 - ass = uncovered NSFW version of butt
9 - split = sitting with legs open wide (90+ degrees), typically with at least one knee above belly
10 - sprd = strongly NSFW version of split
11 - vsplt = stand split or visually similar pose
12 - vsprd = strongly NSFW version of vsplit
13 - hip = full or almost full hip(-s) side view with knee(-s) above belly, usually when sitting or lying
14 - wing = mostly dragon or pony related
15 - feral = all-four non-human torso
16 - hdrago = dragon style head
17 - hpony = pony style head
18 - hfox = cartoon fox / dog head
19 - hrabb = cartoon rabbit head or bunnygirl
20 - hcat = cartoon cat (less sharp muzzle compared to hfox) or catgirl head
21 - hbear = cartoon bear head
22 - jacko = memetic "Jack'O contest pose" with a head toward viewer
23 - jackx = jacko viewed from behind, sometimes strongly NSFW
24 - hhorse = horse head
25 - hbird = bird head
Comments - 0