سيفار 10_n

  • الوصف :

نسخة معاد تسميتها من CIFAR-10 بها أخطاء حقيقية في التعليقات التوضيحية البشرية. لكل زوج (صورة ، ملصق) في مجموعة قطار CIFAR-10 الأصلية ، فإنه يوفر العديد من الملصقات الإضافية التي قدمها المعلقون البشريون الحقيقيون.

  • الصفحة الرئيسية : https://ucsc-real.soe.ucsc.edu:1995/Home.html/

  • كود المصدر : tfds.image_classification.cifar10_n.Cifar10N

  • إصدارات :

    • 1.0.0 : الإصدار الأولي.
    • 1.0.1 : إصلاح خطأ مطبعي في مفتاح worse_label .
    • 1.0.2 : إصلاح المراسلات بين التعليقات التوضيحية والصور.
    • 1.0.3 : الملفات الثابتة في MANUAL_DIR .
    • 1.0.4 (افتراضي): تحميل ثابت للمعلومات الجانبية.
  • حجم التحميل : 162.17 MiB

  • حجم مجموعة البيانات : 147.91 MiB

  • إرشادات التنزيل اليدوي : تتطلب مجموعة البيانات هذه تنزيل بيانات المصدر يدويًا إلى download_config.manual_dir (الإعدادات الافتراضية على ~/tensorflow_datasets/downloads/manual/ ):
    قم بتنزيل "side_info_cifar10N.csv" و "CIFAR-10_human_ordered.npy" و "image_order_c10.npy" من https://github.com/UCSC-REAL/cifar-10-100n

ثم قم بتحويل "CIFAR-10_human_ordered.npy" إلى ملف CSV "CIFAR-10_human_annotations.csv". يمكن القيام بذلك باستخدام الكود التالي:

import numpy as np
from tensorflow_datasets.core.utils.lazy_imports_utils import pandas as pd
from tensorflow_datasets.core.utils.lazy_imports_utils import tensorflow as tf

human_labels_np_path = '<local_path>/CIFAR-10_human_ordered.npy'
human_labels_csv_path = '<local_path>/CIFAR-10_human_annotations.csv'

with tf.io.gfile.GFile(human_labels_np_path, "rb") as f:
  human_annotations = np.load(f, allow_pickle=True)

df = pd.DataFrame(human_annotations[()])

with tf.io.gfile.GFile(human_labels_csv_path, "w") as f:
  df.to_csv(f, index=False)
  • التخزين المؤقت التلقائي ( الوثائق ): نعم

  • الانقسامات :

انشق، مزق أمثلة
'test' 10000
'train' 50000
  • هيكل الميزة :
FeaturesDict({
    'aggre_label': ClassLabel(shape=(), dtype=int64, num_classes=10),
    'id': Text(shape=(), dtype=string),
    'image': Image(shape=(32, 32, 3), dtype=uint8),
    'label': ClassLabel(shape=(), dtype=int64, num_classes=10),
    'random_label1': ClassLabel(shape=(), dtype=int64, num_classes=10),
    'random_label2': ClassLabel(shape=(), dtype=int64, num_classes=10),
    'random_label3': ClassLabel(shape=(), dtype=int64, num_classes=10),
    'worker1_id': int64,
    'worker1_time': float32,
    'worker2_id': int64,
    'worker2_time': float32,
    'worker3_id': int64,
    'worker3_time': float32,
    'worse_label': ClassLabel(shape=(), dtype=int64, num_classes=10),
})
  • وثائق الميزة :
ميزة فصل شكل نوع وصف
الميزات
aggre_label ClassLabel int64
هوية شخصية نص سلسلة
صورة صورة (32 ، 32 ، 3) uint8
ضع الكلمة المناسبة ClassLabel int64
التسمية_ العشوائية 1 ClassLabel int64
التسمية_ العشوائية 2 ClassLabel int64
تسمية عشوائية 3 ClassLabel int64
عامل 1_id موتر int64
عامل 1_ الوقت موتر تعويم 32
عامل 2_id موتر int64
عامل 2_ الوقت موتر تعويم 32
عامل 3_id موتر int64
عامل 3_ الوقت موتر تعويم 32
أسوأ_التصنيف ClassLabel int64

التصور

  • الاقتباس :
@inproceedings{wei2022learning,
  title={Learning with Noisy Labels Revisited: A Study Using Real-World Human
  Annotations},
  author={Jiaheng Wei and Zhaowei Zhu and Hao Cheng and Tongliang Liu and Gang
  Niu and Yang Liu},
  booktitle={International Conference on Learning Representations},
  year={2022},
  url={https://openreview.net/forum?id=TBWA6PLJZQm}
}