cifar10

CIFAR-10 データセットは、10 クラスの 60000 個の 32x32 カラー画像で構成され、クラスごとに 6000 個の画像があります。 50000 のトレーニング イメージと 10000 のテスト イメージがあります。

スプリット
'test' 10,000
'train' 50,000
  • 機能構造:
FeaturesDict({
    'id': Text(shape=(), dtype=string),
    'image': Image(shape=(32, 32, 3), dtype=uint8),
    'label': ClassLabel(shape=(), dtype=int64, num_classes=10),
})
  • 機能のドキュメント:
特徴クラスDtype説明
特徴辞書
ID文章ストリング
画像画像(32, 32, 3) uint8
ラベルクラスラベルint64

視覚化

  • 引用
@TECHREPORT{Krizhevsky09learningmultiple,
    author = {Alex Krizhevsky},
    title = {Learning multiple layers of features from tiny images},
    institution = {},
    year = {2009}
}