- תיאור:
גירויים להתנגשות צורה/מרקם מ- "CNN מאומני ImageNet מוטים כלפי מרקם; הגדלת הטיית הצורה משפרת את הדיוק ואת החוסן."
שים לב כי למרות שמקור הנתונים מכיל תמונות בעלות צורה ומרקם תואמים ואנו כוללות אותן כאן, הן מתעלמות מרוב ההערכות בעיתון המקורי.
גרסאות:
-
1.0.0
(ברירת המחדל): הערות השחרור לא.
-
גודל ההורדה:
153.96 MiB
מערך נתונים גודל:
130.44 MiB
Auto-במטמון ( תיעוד ): רק כאשר
shuffle_files=False
(מבחן)פיצולים:
לְפַצֵל | דוגמאות |
---|---|
'test' | 1,280 |
- מאפיינים:
FeaturesDict({
'file_name': Text(shape=(), dtype=tf.string),
'image': Image(shape=(None, None, 3), dtype=tf.uint8),
'shape_imagenet_labels': Sequence(ClassLabel(shape=(), dtype=tf.int64, num_classes=1000)),
'shape_label': ClassLabel(shape=(), dtype=tf.int64, num_classes=16),
'texture_imagenet_labels': Sequence(ClassLabel(shape=(), dtype=tf.int64, num_classes=1000)),
'texture_label': ClassLabel(shape=(), dtype=tf.int64, num_classes=16),
})
מפתחות השגחה (ראה
as_supervised
doc ):('image', 'shape_label')
איור ( tfds.show_examples ):
- דוגמאות ( tfds.as_dataframe ):
- 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},
}