geirhos_conflict_stimuli

  • Deskripsi :

Rangsangan konflik bentuk/tekstur dari "CNN terlatih ImageNet bias terhadap tekstur; meningkatkan bias bentuk meningkatkan akurasi dan ketahanan."

Perhatikan bahwa, meskipun sumber kumpulan data berisi gambar dengan bentuk dan tekstur yang cocok dan kami sertakan di sini, mereka diabaikan untuk sebagian besar evaluasi di makalah aslinya.

Membelah Contoh
'test' 1.280
  • Struktur fitur :
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),
})
  • Dokumentasi fitur :
Fitur Kelas Membentuk Dtype Keterangan
fiturDict
nama file Teks rangkaian
gambar Gambar (Tidak ada, Tidak ada, 3) uint8
shape_imagenet_labels Urutan(Label Kelas) (Tidak ada,) int64
shape_label LabelKelas int64
texture_imagenet_labels Urutan(Label Kelas) (Tidak ada,) int64
tekstur_label LabelKelas int64

Visualisasi

  • Kutipan :
@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},
}