visual_domain_decathlon

  • 説明:

これには、PASCAL in Detail Workshop Challenge (CVPR 2017) の一部である Visual Domain Decathlon で使用される 10 個のデータセットが含まれています。この課題の目標は、非常に異なる視覚領域を代表する 10 個の画像分類問題を同時に解決することです。

ここに含まれるデータセットの一部は、TFDS の個別のデータセットとしても利用できます。ただし、画像は Visual Domain Decathlon (72 ピクセルの短いサイズになるように等方的にサイズ変更) 用に前処理されており、トレーニング/検証/テストの分割が異なる場合があることに注意してください。ここでは、競技用の公式スプリットを使用します。

@ONLINE{hakanbilensylvestrerebuffitomasjakab2017,
    author = "Hakan Bilen, Sylvestre Rebuffi, Tomas Jakab",
    title  = "Visual Domain Decathlon",
    year   = "2017",
    url    = "https://www.robots.ox.ac.uk/~vgg/decathlon/"
}

visual_domain_decathlon/aircraft (デフォルト設定)

  • 構成の説明: 「航空機」に基づくデータで、画像は等方的にサイズ変更されて 72 ピクセルの短いサイズになります。

  • ダウンロードサイズ: 409.94 MiB

  • データセットサイズ: 20.96 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
'test' 3,333
'train' 3,334
'validation' 3,333
  • 機能構造:
FeaturesDict({
    'image': Image(shape=(None, None, 3), dtype=uint8),
    'label': ClassLabel(shape=(), dtype=int64, num_classes=100),
    'name': Text(shape=(), dtype=string),
})
  • 機能のドキュメント:
特徴クラスDtype説明
特徴辞書
画像画像(なし、なし、3) uint8
ラベルクラスラベルint64
名前文章ストリング

視覚化

visual_domain_decathlon/cifar100

  • 構成の説明: 「CIFAR-100」に基づくデータで、画像は等方的にサイズ変更されて 72 ピクセルの短いサイズになります。

  • ダウンロードサイズ: 409.94 MiB

  • データセットのサイズ: 119.43 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
'test' 10,000
'train' 40,000
'validation' 10,000
  • 機能構造:
FeaturesDict({
    'image': Image(shape=(None, None, 3), dtype=uint8),
    'label': ClassLabel(shape=(), dtype=int64, num_classes=100),
    'name': Text(shape=(), dtype=string),
})
  • 機能のドキュメント:
特徴クラスDtype説明
特徴辞書
画像画像(なし、なし、3) uint8
ラベルクラスラベルint64
名前文章ストリング

視覚化

visual_domain_decathlon/daimlerpedcls

  • 構成の説明: 「Daimler Pedestrian Classification」に基づくデータ。画像は等方的にサイズ変更され、72 ピクセルの短いサイズになります。

  • ダウンロードサイズ: 409.94 MiB

  • データセットのサイズ: 68.35 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
'test' 19,600
'train' 23,520
'validation' 5,880
  • 機能構造:
FeaturesDict({
    'image': Image(shape=(None, None, 3), dtype=uint8),
    'label': ClassLabel(shape=(), dtype=int64, num_classes=2),
    'name': Text(shape=(), dtype=string),
})
  • 機能のドキュメント:
特徴クラスDtype説明
特徴辞書
画像画像(なし、なし、3) uint8
ラベルクラスラベルint64
名前文章ストリング

視覚化

visual_domain_decathlon/dtd

  • 構成の説明: 「記述可能なテクスチャ」に基づくデータ。画像は等方的にサイズ変更され、72 ピクセルの短いサイズになります。

  • ダウンロードサイズ: 409.94 MiB

  • データセットのサイズ: 13.30 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
'test' 1,880
'train' 1,880
'validation' 1,880
  • 機能構造:
FeaturesDict({
    'image': Image(shape=(None, None, 3), dtype=uint8),
    'label': ClassLabel(shape=(), dtype=int64, num_classes=47),
    'name': Text(shape=(), dtype=string),
})
  • 機能のドキュメント:
特徴クラスDtype説明
特徴辞書
画像画像(なし、なし、3) uint8
ラベルクラスラベルint64
名前文章ストリング

視覚化

visual_domain_decathlon/gtsrb

  • 構成の説明: 「ドイツの交通標識」に基づくデータで、画像は等方的にサイズ変更されて 72 ピクセルの短いサイズになります。

  • ダウンロードサイズ: 409.94 MiB

  • データセットのサイズ: 80.58 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
'test' 12,630
'train' 31,367
'validation' 7,842
  • 機能構造:
FeaturesDict({
    'image': Image(shape=(None, None, 3), dtype=uint8),
    'label': ClassLabel(shape=(), dtype=int64, num_classes=43),
    'name': Text(shape=(), dtype=string),
})
  • 機能のドキュメント:
特徴クラスDtype説明
特徴辞書
画像画像(なし、なし、3) uint8
ラベルクラスラベルint64
名前文章ストリング

視覚化

visual_domain_decathlon/imagenet12

  • 構成の説明: 「Imagenet」に基づくデータで、画像は等方的にサイズ変更されて 72 ピクセルの短いサイズになります。

  • ダウンロードサイズ: 6.11 GiB

  • データセットサイズ: 5.24 GiB

  • 自動キャッシュ(ドキュメント): いいえ

  • スプリット:

スプリット
'test' 48,238
'train' 1,232,167
'validation' 49,000
  • 機能構造:
FeaturesDict({
    'image': Image(shape=(None, None, 3), dtype=uint8),
    'label': ClassLabel(shape=(), dtype=int64, num_classes=1000),
    'name': Text(shape=(), dtype=string),
})
  • 機能のドキュメント:
特徴クラスDtype説明
特徴辞書
画像画像(なし、なし、3) uint8
ラベルクラスラベルint64
名前文章ストリング

視覚化

visual_domain_decathlon/オムニグロット

  • 構成の説明: 「Omniglot」に基づくデータで、画像は等方的にサイズ変更されて 72 ピクセルの短いサイズになります。

  • ダウンロードサイズ: 409.94 MiB

  • データセットのサイズ: 41.46 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
'test' 8,115
'train' 17,853
'validation' 6,492
  • 機能構造:
FeaturesDict({
    'image': Image(shape=(None, None, 3), dtype=uint8),
    'label': ClassLabel(shape=(), dtype=int64, num_classes=1623),
    'name': Text(shape=(), dtype=string),
})
  • 機能のドキュメント:
特徴クラスDtype説明
特徴辞書
画像画像(なし、なし、3) uint8
ラベルクラスラベルint64
名前文章ストリング

視覚化

visual_domain_decathlon/svhn

  • 構成の説明: 「Street View House Numbers」に基づくデータ。画像は等方的にサイズ変更され、72 ピクセルの短いサイズになります。

  • ダウンロードサイズ: 409.94 MiB

  • データセットのサイズ: 135.32 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
'test' 26,032
'train' 47,217
'validation' 26,040
  • 機能構造:
FeaturesDict({
    'image': Image(shape=(None, None, 3), dtype=uint8),
    'label': ClassLabel(shape=(), dtype=int64, num_classes=10),
    'name': Text(shape=(), dtype=string),
})
  • 機能のドキュメント:
特徴クラスDtype説明
特徴辞書
画像画像(なし、なし、3) uint8
ラベルクラスラベルint64
名前文章ストリング

視覚化

visual_domain_decathlon/ucf101

  • 構成の説明: 「UCF101 Dynamic Images」に基づくデータで、画像は等方的にサイズ変更されて 72 ピクセルの短いサイズになります。

  • ダウンロードサイズ: 409.94 MiB

  • データセットサイズ: 19.73 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
'test' 3,783
'train' 7,585
'validation' 1,952人
  • 機能構造:
FeaturesDict({
    'image': Image(shape=(None, None, 3), dtype=uint8),
    'label': ClassLabel(shape=(), dtype=int64, num_classes=101),
    'name': Text(shape=(), dtype=string),
})
  • 機能のドキュメント:
特徴クラスDtype説明
特徴辞書
画像画像(なし、なし、3) uint8
ラベルクラスラベルint64
名前文章ストリング

視覚化

visual_domain_decathlon/vgg-flowers

  • 構成の説明: 「VGG-Flowers」に基づくデータで、画像は等方的にサイズ変更されて 72 ピクセルの短いサイズになります。

  • ダウンロードサイズ: 409.94 MiB

  • データセットサイズ: 20.87 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
'test' 6,149
'train' 1,020
'validation' 1,020
  • 機能構造:
FeaturesDict({
    'image': Image(shape=(None, None, 3), dtype=uint8),
    'label': ClassLabel(shape=(), dtype=int64, num_classes=102),
    'name': Text(shape=(), dtype=string),
})
  • 機能のドキュメント:
特徴クラスDtype説明
特徴辞書
画像画像(なし、なし、3) uint8
ラベルクラスラベルint64
名前文章ストリング

視覚化