geirhos_conflict_stimuli

  • 説明:

「ImageNet でトレーニングされた CNN はテクスチャに偏っています。形状バイアスを増やすと、精度と堅牢性が向上します。」

データセット ソースには形状とテクスチャが一致する画像が含まれており、ここに含めていますが、元の論文のほとんどの評価では無視されていることに注意してください。

スプリット
'test' 1,280
  • 機能構造:
FeaturesDict({
    'file_name': Text(shape=(), dtype=string),
    'image': Image(shape=(None, None, 3), dtype=uint8),
    'shape_imagenet_labels': Sequence(ClassLabel(shape=(), dtype=int64, num_classes=1000)),
    'shape_label': ClassLabel(shape=(), dtype=int64, num_classes=16),
    'texture_imagenet_labels': Sequence(ClassLabel(shape=(), dtype=int64, num_classes=1000)),
    'texture_label': ClassLabel(shape=(), dtype=int64, num_classes=16),
})
  • 機能のドキュメント:
特徴クラスDtype説明
特徴辞書
ファイル名文章ストリング
画像画像(なし、なし、3) uint8
shape_imagenet_labelsシーケンス(クラスラベル) (なし、) int64
shape_labelクラスラベルint64
texture_imagenet_labelsシーケンス(クラスラベル) (なし、) int64
texture_labelクラスラベルint64

視覚化

  • 引用
@inproceedings{
  geirhos2018imagenettrained,
  title={ImageNet-trained {CNN}s are biased towards texture; increasing shape
         bias improves accuracy and robustness.},
  author={Robert Geirhos and Patricia Rubisch and Claudio Michaelis and
          Matthias Bethge and Felix A. Wichmann and Wieland Brendel},
  booktitle={International Conference on Learning Representations},
  year={2019},
  url={https://openreview.net/forum?id=Bygh9j09KX},
}