cifar100

  • คำอธิบาย :

ชุดข้อมูลนี้เหมือนกับ CIFAR-10 ยกเว้นว่ามี 100 คลาส แต่ละคลาสมี 600 รูปภาพ มีรูปภาพการฝึกอบรม 500 ภาพและรูปภาพทดสอบ 100 ภาพต่อชั้นเรียน 100 คลาสใน CIFAR-100 ถูกจัดกลุ่มเป็น 20 ซูเปอร์คลาส รูปภาพแต่ละรูปมาพร้อมกับป้ายกำกับ "ละเอียด" (คลาสที่รูปภาพนั้นอยู่) และป้ายกำกับ "หยาบ" (คลาสพิเศษที่รูปภาพนั้นอยู่)

แยก ตัวอย่าง
'test' 10,000
'train' 50,000
  • โครงสร้างคุณสมบัติ :
FeaturesDict({
    'coarse_label': ClassLabel(shape=(), dtype=int64, num_classes=20),
    'id': Text(shape=(), dtype=string),
    'image': Image(shape=(32, 32, 3), dtype=uint8),
    'label': ClassLabel(shape=(), dtype=int64, num_classes=100),
})
  • เอกสารคุณสมบัติ :
คุณสมบัติ ระดับ รูปร่าง ประเภทD คำอธิบาย
คุณสมบัติDict
หยาบ_ฉลาก ClassLabel int64
รหัส ข้อความ เชือก
ภาพ ภาพ (32, 32, 3) uint8
ฉลาก ClassLabel int64

การแสดงภาพ

  • การอ้างอิง :
@TECHREPORT{Krizhevsky09learningmultiple,
    author = {Alex Krizhevsky},
    title = {Learning multiple layers of features from tiny images},
    institution = {},
    year = {2009}
}