cifar10_n

  • Descrizione :

Una versione rietichettata di CIFAR-10 con veri errori di annotazione umana. Per ogni coppia (immagine, etichetta) nel treno CIFAR-10 originale, fornisce diverse etichette aggiuntive fornite da veri annotatori umani.

  • Pagina iniziale : https://ucsc-real.soe.ucsc.edu:1995/Home.html/

  • Codice sorgente : tfds.image_classification.cifar10_n.Cifar10N

  • Versioni :

    • 1.0.0 : Versione iniziale.
    • 1.0.1 : Risolto errore di battitura nella chiave worse_label .
    • 1.0.2 : Corretta corrispondenza tra annotazioni e immagini.
    • 1.0.3 : Correzione dei file in MANUAL_DIR .
    • 1.0.4 (predefinito): Corretto il caricamento delle informazioni laterali.
  • Dimensione del download : 162.17 MiB

  • Dimensione del set di dati: 147.91 MiB

  • Istruzioni per il download manuale : questo set di dati richiede di scaricare manualmente i dati di origine in download_config.manual_dir (il valore predefinito ~/tensorflow_datasets/downloads/manual/ ):
    Scarica 'side_info_cifar10N.csv', 'CIFAR-10_human_ordered.npy' e 'image_order_c10.npy' da https://github.com/UCSC-REAL/cifar-10-100n

Quindi converti "CIFAR-10_human_ordered.npy" in un file CSV "CIFAR-10_human_annotations.csv". Questo può essere fatto con il seguente codice:

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)
Diviso Esempi
'test' 10.000
'train' 50.000
  • Struttura delle caratteristiche :
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),
})
  • Documentazione delle funzionalità :
Caratteristica Classe Forma Tipo D Descrizione
CaratteristicheDict
aggre_label ClassLabel int64
id Testo corda
Immagine Immagine (32, 32, 3) uint8
etichetta ClassLabel int64
etichetta_casuale1 ClassLabel int64
etichetta_casuale2 ClassLabel int64
etichetta_casuale3 ClassLabel int64
lavoratore1_id Tensore int64
lavoratore1_ora Tensore galleggiante32
lavoratore2_id Tensore int64
lavoratore2_ora Tensore galleggiante32
lavoratore3_id Tensore int64
lavoratore3_ora Tensore galleggiante32
etichetta_peggiore ClassLabel int64

Visualizzazione

  • Citazione :
@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}
}