- 説明:
ワンショット学習用のOmniglotデータセット。このデータセットには、50種類のアルファベットから1623種類の手書き文字が含まれています。
追加ドキュメント: Papers With Code で探す
バージョン:
-
3.0.0(デフォルト): 新しい分割 API ( https://tensorflow.org/datasets/splits )
-
ダウンロードサイズ:
17.95 MiBデータセットサイズ:
12.29 MiB自動キャッシュ(ドキュメント): はい
スプリット:
| スプリット | 例 |
|---|---|
'small1' | 2,720 |
'small2' | 3,120 |
'test' | 13,180 |
'train' | 19,280 |
- 機能構造:
FeaturesDict({
'alphabet': ClassLabel(shape=(), dtype=int64, num_classes=50),
'alphabet_char_id': int64,
'image': Image(shape=(105, 105, 3), dtype=uint8),
'label': ClassLabel(shape=(), dtype=int64, num_classes=1623),
})
- 機能ドキュメント:
| 特徴 | クラス | 形 | Dタイプ | 説明 |
|---|---|---|---|---|
| 機能辞書 | ||||
| アルファベット | クラスラベル | int64 | ||
| アルファベット文字ID | テンソル | int64 | ||
| 画像 | 画像 | (105、105、3) | uint8 | |
| ラベル | クラスラベル | int64 |
教師ありキー(
as_supervisedドキュメントを参照):('image', 'label')図( tfds.show_examples ):

- 例( tfds.as_dataframe ):
- 引用:
@article{lake2015human,
title={Human-level concept learning through probabilistic program induction},
author={Lake, Brenden M and Salakhutdinov, Ruslan and Tenenbaum, Joshua B},
journal={Science},
volume={350},
number={6266},
pages={1332--1338},
year={2015},
publisher={American Association for the Advancement of Science}
}