moe / README.md
nebulette's picture
Update README.md
5c4a353 verified
|
Raw History Blame
1.73 kB
---
license: other
size_categories:
- 100K<n<1M
configs:
- config_name: default
data_files:
- split: train
path: metadata/train.parquet
---
# Danbooru 2026 から選んだ日本の萌えイラスト(454,978 枚)
```python
import pandas as pd, requests
from huggingface_hub import hf_hub_url
meta = pd.read_parquet("hf://datasets/swdq/danbooru2026-moe/metadata/train.parquet")
r = meta.iloc[0]
url = hf_hub_url("nyanko-devs/danbooru2026", f"original/data-{r.src_shard:04d}.tar", repo_type="dataset")
webp = requests.get(url, headers={"Range": f"bytes={r.src_offset}-{r.src_offset + r.src_length - 1}"}).content
```
## 列
| 列 | 内容 |
|---|---|
| id, md5, created_at, rating, score, up_score, down_score, fav_count | Danbooru の投稿 |
| image_width, image_height, file_ext, file_size, source, pixiv_id, parent_id, has_children | 同上(元の画像の値) |
| tag_string_general / character / copyright / artist / meta | タグ(空白区切り) |
| score_pct | 同じ rating・投稿月(日本時間)の候補の中でのスコアの順位(0〜1) |
| x_followers | 投稿の作者のうち、いちばん多い X フォロワー数 |
| reviewed | 画面で人が採用した |
| ai_refusal | 作者の X のプロフィールに AI 学習の拒否がある |
| src_shard, src_offset, src_length | 元データ nyanko-devs/danbooru2026 の `original/data-NNNN.tar` の中の位置 |
## 出典
- [ootsuka (swdq)](https://huggingface.co/datasets/swdq/danbooru2026-moe)
- 画像: [nyanko-devs/danbooru2026](https://huggingface.co/datasets/nyanko-devs/danbooru2026)(Danbooru)
- tar の中の位置: [petersunde/danbooru2026-index](https://huggingface.co/datasets/petersunde/danbooru2026-index)