음식101

  • 설명 :

이 데이터 세트는 101,000개의 이미지가 포함된 101개의 음식 범주로 구성됩니다. 클래스별로 수동으로 검토한 테스트 이미지 250개와 훈련 이미지 750개가 제공됩니다. 일부러 훈련 이미지를 정리하지 않았으므로 여전히 약간의 노이즈가 포함되어 있습니다. 이것은 주로 강렬한 색상과 때로는 잘못된 라벨의 형태로 나타납니다. 모든 이미지는 최대 측면 길이가 512픽셀로 조정되었습니다.

나뉘다
'train' 75,750
'validation' 25,250
  • 기능 구조 :
FeaturesDict({
    'image': Image(shape=(None, None, 3), dtype=uint8),
    'label': ClassLabel(shape=(), dtype=int64, num_classes=101),
})
  • 기능 문서 :
특징 수업 모양 D타입 설명
풍모Dict
영상 영상 (없음, 없음, 3) uint8
상표 클래스 레이블 int64

심상

  • 인용 :
@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}
}