カルテック101

  • 説明

Caltech-101 は、101 のクラスに属するオブジェクトの写真と、1 つのbackground clutterクラスで構成されています。各画像には 1 つのオブジェクトのラベルが付けられます。各クラスにはおよそ 40 ~ 800 個の画像が含まれており、合計で約 9,000 個の画像になります。画像のサイズは可変であり、一般的なエッジの長さは 200 ~ 300 ピクセルです。このバージョンにはイメージレベルのラベルのみが含まれています。元のデータセットには境界ボックスも含まれています。

スプリット
'test' 6,084
'train' 3,060
  • 機能の構造:
FeaturesDict({
   
'image': Image(shape=(None, None, 3), dtype=uint8),
   
'image/file_name': Text(shape=(), dtype=string),
   
'label': ClassLabel(shape=(), dtype=int64, num_classes=102),
})
  • 機能ドキュメント:
特徴クラスDタイプ説明
特徴辞書
画像画像(なし、なし、3) uint8
画像/ファイル名文章
ラベルクラスラベルint64

視覚化

  • 引用
@article{FeiFei2004LearningGV,
  title
={Learning Generative Visual Models from Few Training Examples: An Incremental Bayesian Approach Tested on 101 Object Categories},
  author
={Li Fei-Fei and Rob Fergus and Pietro Perona},
  journal
={Computer Vision and Pattern Recognition Workshop},
  year
={2004},
}