cifar10

ชุดข้อมูล CIFAR-10 ประกอบด้วยภาพสีขนาด 32x32 จำนวน 60,000 ภาพใน 10 คลาส โดยมี 6,000 ภาพต่อคลาส มีภาพการฝึกอบรม 50,000 ภาพและภาพการทดสอบ 10,000 ภาพ

แยก ตัวอย่าง
'test' 10,000
'train' 50,000
  • โครงสร้างคุณลักษณะ :
FeaturesDict({
    'id': Text(shape=(), dtype=string),
    'image': Image(shape=(32, 32, 3), dtype=uint8),
    'label': ClassLabel(shape=(), dtype=int64, num_classes=10),
})
  • เอกสารคุณสมบัติ :
ลักษณะเฉพาะ ระดับ รูปร่าง Dประเภท คำอธิบาย
คุณสมบัติDict
รหัส ข้อความ สตริง
ภาพ ภาพ (32, 32, 3) uint8
ฉลาก ป้ายกำกับคลาส int64

การสร้างภาพ

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