cifar10_n

  • תיאור :

גרסה עם תיוג מחדש של CIFAR-10 עם שגיאות ביאורים אנושיים אמיתיים. עבור כל זוג (תמונה, תווית) בערכת הרכבות המקורית של CIFAR-10, הוא מספק מספר תוויות נוספות שניתנו על ידי כותבים אנושיים אמיתיים.

  • דף הבית : https://ucsc-real.soe.ucsc.edu:1995/Home.html/

  • קוד מקור : tfds.image_classification.cifar10_n.Cifar10N

  • גרסאות :

    • 1.0.0 : שחרור ראשוני.
    • 1.0.1 : תוקן שגיאת הקלדה במקש worse_label .
    • 1.0.2 : התאמה קבועה בין הערות לתמונות.
    • 1.0.3 : תיקנו קבצים ב- MANUAL_DIR .
    • 1.0.4 (ברירת מחדל): טעינה קבועה של מידע צדדי.
  • גודל הורדה : 162.17 MiB

  • גודל ערכת נתונים: 147.91 MiB

  • הוראות הורדה ידניות : מערך נתונים זה מחייב אותך להוריד את נתוני המקור באופן ידני אל download_config.manual_dir (ברירת המחדל היא ~/tensorflow_datasets/downloads/manual/ ):
    הורד את 'side_info_cifar10N.csv', 'CIFAR-10_human_ordered.npy' ו-'image_order_c10.npy' מ- https://github.com/UCSC-REAL/cifar-10-100n

לאחר מכן המר את 'CIFAR-10_human_ordered.npy' לקובץ CSV 'CIFAR-10_human_annotations.csv'. ניתן לעשות זאת באמצעות הקוד הבא:

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)
  • שמור אוטומטי במטמון ( תיעוד ): כן

  • פיצולים :

לְפַצֵל דוגמאות
'test' 10,000
'train' 50,000
  • מבנה תכונה :
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),
})
  • תיעוד תכונה :
תכונה מעמד צוּרָה Dtype תיאור
FeaturesDict
aggre_label ClassLabel int64
תְעוּדַת זֶהוּת טֶקסט חוּט
תמונה תמונה (32, 32, 3) uint8
תווית ClassLabel int64
תווית_אקראית1 ClassLabel int64
תווית_אקראית2 ClassLabel int64
תווית_אקראית3 ClassLabel int64
worker1_id מוֹתֵחַ int64
worker1_time מוֹתֵחַ לצוף32
worker2_id מוֹתֵחַ int64
worker2_time מוֹתֵחַ לצוף32
worker3_id מוֹתֵחַ int64
worker3_time מוֹתֵחַ לצוף32
תווית_גרוע יותר ClassLabel int64

רְאִיָה

  • ציטוט :
@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}
}