シファール10

  • 説明

CIFAR-10 データセットは、10 クラスの 60,000 個の 32x32 カラー画像で構成されており、クラスあたり 6,000 個の画像があります。 50000 個のトレーニング画像と 10000 個のテスト画像があります。

スプリット
'test' 10,000
'train' 50,000
  • 機能の構造:
FeaturesDict({
   
'id': Text(shape=(), dtype=string),
   
'image': Image(shape=(32, 32, 3), dtype=uint8),
   
'label': ClassLabel(shape=(), dtype=int64, num_classes=10),
})
  • 機能ドキュメント:
特徴クラスDタイプ説明
特徴辞書
ID文章
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}
}