- Açıklama:
"ImageNet tarafından eğitilmiş CNN'ler dokuya karşı önyargılıdır; artan şekil sapması doğruluğu ve sağlamlığı artırır."
Veri kümesi kaynağının eşleşen şekil ve dokuya sahip görüntüler içermesine ve bunları buraya dahil etmemize rağmen, orijinal belgedeki çoğu değerlendirme için göz ardı edildiğini unutmayın.
Kaynak kodu:
tfds.image_classification.GeirhosConflictStimuli
sürümleri:
-
1.0.0
(varsayılan): Hayır sürüm notları.
-
İndirme boyutu:
153.96 MiB
Veri kümesi boyutu:
130.44 MiB
Otomatik önbelleğe ( dokümantasyon ): Yalnızca
shuffle_files=False
(test)Splits:
Bölmek | Örnekler |
---|---|
'test' | 1.280 |
- Özellikler:
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),
})
Denetimli tuşları (Bkz
as_supervised
doc ):('image', 'shape_label')
Şekil ( tfds.show_examples ):
- Örnekler ( tfds.as_dataframe ):
- Citation:
@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},
}