cifar100_n

  • विवरण :

वास्तविक मानव एनोटेशन त्रुटियों के साथ CIFAR-100 का पुनः लेबल किया गया संस्करण। मूल CIFAR-100 ट्रेन सेट में प्रत्येक जोड़ी (छवि, लेबल) के लिए, यह एक वास्तविक मानव एनोटेटर द्वारा दिया गया एक अतिरिक्त लेबल प्रदान करता है।

  • होमपेज : https://www.cs.toronto.edu/~kriz/cifar.html

  • स्रोत कोड : tfds.image_classification.cifar100_n.Cifar100N

  • संस्करण :

    • 1.0.0 : प्रारंभिक रिलीज।
    • 1.0.1 (डिफ़ॉल्ट): एनोटेशन और छवियों के बीच निश्चित पत्राचार।
  • डाउनलोड आकार : 160.71 MiB

  • डेटासेट का आकार : 136.07 MiB

  • मैन्युअल डाउनलोड निर्देश : इस डेटासेट के लिए आपको स्रोत डेटा को मैन्युअल रूप से download_config.manual_dir (डिफ़ॉल्ट रूप से ~/tensorflow_datasets/downloads/manual/ ) में डाउनलोड करना होगा:
    https://github.com/UCSC-REAL/cifar-10-100n से 'side_info_cifar100N.csv', 'CIFAR-100_human_ordered.npy' और 'image_order_c100.npy' डाउनलोड करें

फिर 'CIFAR-100_human_ordered.npy' को CSV फ़ाइल 'CIFAR-100_human_annotations.csv' में बदलें। यह निम्नलिखित कोड के साथ किया जा सकता है:

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)
विभाजित करना उदाहरण
'test' 10,000
'train' 50,000
  • फ़ीचर संरचना :
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,
})
  • फ़ीचर दस्तावेज़ीकरण :
विशेषता कक्षा आकार डीटाइप विवरण
विशेषताएं डिक्ट
मोटे_लेबल क्लासलेबल int64
पहचान मूलपाठ डोरी
छवि छवि (32, 32, 3) uint8
लेबल क्लासलेबल int64
noise_label क्लासलेबल int64
कार्यकर्ता_आईडी टेन्सर int64
कार्यकर्ता_समय टेन्सर फ्लोट32

VISUALIZATION

  • उद्धरण :
@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}
}