אוכל101

  • תיאור :

מערך נתונים זה מורכב מ-101 קטגוריות מזון, עם 101,000 תמונות. עבור כל שיעור, 250 תמונות מבחן שנבדקו ידנית מסופקות וכן 750 תמונות הדרכה. בכוונה, תמונות האימון לא נוקו, ולכן עדיין מכילות כמות מסוימת של רעש. זה מגיע בעיקר בצורה של צבעים עזים ולפעמים תוויות שגויות. כל התמונות הותאמו לאורך צד מקסימלי של 512 פיקסלים.

לְפַצֵל דוגמאות
'train' 75,750
'validation' 25,250
  • מבנה תכונה :
FeaturesDict({
    'image': Image(shape=(None, None, 3), dtype=uint8),
    'label': ClassLabel(shape=(), dtype=int64, num_classes=101),
})
  • תיעוד תכונה :
תכונה מעמד צוּרָה Dtype תיאור
FeaturesDict
תמונה תמונה (אין, אין, 3) uint8
תווית ClassLabel int64

רְאִיָה

  • ציטוט :
@inproceedings{bossard14,
  title = {Food-101 -- Mining Discriminative Components with Random Forests},
  author = {Bossard, Lukas and Guillaumin, Matthieu and Van Gool, Luc},
  booktitle = {European Conference on Computer Vision},
  year = {2014}
}