גיירוס_קונפליקט_גירוי

  • תיאור :

גירויי קונפליקט של צורה/מרקם מ"CNNs מאומנים ב-ImageNet מוטים למרקם; הגדלת הטיית הצורה משפרת את הדיוק והחוסן."

שימו לב שלמרות שמקור הנתונים מכיל תמונות בעלות צורה ומרקם תואמים ואנו כוללים אותן כאן, מתעלמים מהן עבור רוב ההערכות במאמר המקורי.

לְפַצֵל דוגמאות
'test' 1,280
  • מבנה תכונה :
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),
})
  • תיעוד תכונה :
תכונה מעמד צוּרָה Dtype תיאור
FeaturesDict
שם קובץ טֶקסט חוּט
תמונה תמונה (אין, אין, 3) uint8
רשת_תוויות_צורה Sequence (ClassLabel) (אף אחד,) int64
תווית_צורה ClassLabel int64
texture_imagenet_labels Sequence (ClassLabel) (אף אחד,) int64
texture_label ClassLabel int64

רְאִיָה

  • ציטוט :
@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},
}