cifar10_n

  • توضیحات :

نسخه ای با برچسب مجدد CIFAR-10 با خطاهای حاشیه نویسی انسانی واقعی. برای هر جفت (تصویر، برچسب) در مجموعه اصلی قطار CIFAR-10، چندین برچسب اضافی ارائه شده توسط حاشیه نویسان واقعی انسان ارائه می کند.

  • صفحه اصلی : https://ucsc-real.soe.ucsc.edu:1995/Home.html/

  • کد منبع : tfds.image_classification.cifar10_n.Cifar10N

  • نسخه ها :

    • 1.0.0 : انتشار اولیه.
    • 1.0.1 : رفع اشتباه تایپی در کلید worse_label .
    • 1.0.2 : مطابقت ثابت بین حاشیه نویسی و تصاویر.
    • 1.0.3 : فایل های ثابت در MANUAL_DIR .
    • 1.0.4 (پیش فرض): بارگذاری ثابت اطلاعات جانبی.
  • حجم دانلود : 162.17 MiB

  • حجم مجموعه داده : 147.91 MiB

  • دستورالعمل‌های دانلود دستی : این مجموعه داده از شما می‌خواهد که داده‌های منبع را به صورت دستی در download_config.manual_dir (پیش‌فرض ~/tensorflow_datasets/downloads/manual/ ):
    «side_info_cifar10N.csv»، «CIFAR-10_human_ordered.npy» و «image_order_c10.npy» را از https://github.com/UCSC-REAL/cifar-10-100n دانلود کنید

سپس «CIFAR-10_human_ordered.npy» را به یک فایل CSV «CIFAR-10_human_annotations.csv» تبدیل کنید. این کار با کد زیر قابل انجام است:

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)
  • ذخیره خودکار ( اسناد ): بله

  • تقسیم ها :

شکاف مثال ها
'test' 10000
'train' 50000
  • ساختار ویژگی :
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),
})
  • مستندات ویژگی :
ویژگی کلاس شکل نوع D شرح
FeaturesDict
aggre_label ClassLabel int64
شناسه متن رشته
تصویر تصویر (32، 32، 3) uint8
برچسب ClassLabel int64
random_label1 ClassLabel int64
random_label2 ClassLabel int64
random_label3 ClassLabel int64
worker1_id تانسور int64
worker1_time تانسور float32
worker2_id تانسور int64
worker2_time تانسور float32
worker3_id تانسور int64
worker3_time تانسور float32
بدتر_برچسب ClassLabel int64

تجسم

  • نقل قول :
@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}
}