cifar100_n

  • Deskripsi :

Versi CIFAR-100 yang diberi label ulang dengan kesalahan anotasi manusia nyata. Untuk setiap pasangan (gambar, label) dalam rangkaian kereta CIFAR-100 asli, ini memberikan label tambahan yang diberikan oleh anotator manusia asli.

Kemudian ubah 'CIFAR-100_human_ordered.npy' menjadi file CSV 'CIFAR-100_human_annotations.csv'. Ini dapat dilakukan dengan kode berikut:

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-100_human_ordered.npy'
human_labels_csv_path = '<local_path>/CIFAR-100_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)
Membelah Contoh
'test' 10.000
'train' 50.000
  • Struktur fitur :
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),
    'noise_label': ClassLabel(shape=(), dtype=int64, num_classes=100),
    'worker_id': int64,
    'worker_time': float32,
})
  • Dokumentasi fitur :
Fitur Kelas Membentuk Dtype Keterangan
fiturDict
label_kasar LabelKelas int64
Indo Teks rangkaian
gambar Gambar (32, 32, 3) uint8
label LabelKelas int64
noise_label LabelKelas int64
pekerja_id Tensor int64
waktu_pekerja Tensor float32

Visualisasi

  • Kutipan :
@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}
}