cifar100_n

  • বর্ণনা :

বাস্তব মানুষের টীকা ত্রুটি সহ CIFAR-100 এর একটি পুনঃলেবেলযুক্ত সংস্করণ৷ আসল CIFAR-100 ট্রেন সেটে প্রতিটি জোড়ার (ছবি, লেবেল) জন্য, এটি একজন প্রকৃত মানব টীকাকার দ্বারা প্রদত্ত একটি অতিরিক্ত লেবেল প্রদান করে।

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

  • উত্স কোড : tfds.image_classification.cifar100_n.Cifar100N

  • সংস্করণ :

    • 1.0.0 : প্রাথমিক প্রকাশ।
    • 1.0.1 (ডিফল্ট): টীকা এবং চিত্রের মধ্যে স্থির চিঠিপত্র।
  • ডাউনলোড সাইজ : 160.71 MiB

  • ডেটাসেটের আকার : 136.07 MiB

  • ম্যানুয়াল ডাউনলোডের নির্দেশাবলী : এই ডেটাসেটের জন্য আপনাকে ডাউনলোড_config.manual_dir-এ ম্যানুয়ালি উৎস ডেটা download_config.manual_dir করতে হবে ( ~/tensorflow_datasets/downloads/manual/ ডিফল্ট):
    https://github.com/UCSC-REAL/cifar-10-100n থেকে 'side_info_cifar100N.csv', 'CIFAR-100_human_ordered.npy' এবং 'image_order_c100.npy' ডাউনলোড করুন

তারপর 'CIFAR-100_human_ordered.npy' কে একটি CSV ফাইল 'CIFAR-100_human_annotations.csv'-এ রূপান্তর করুন। এটি নিম্নলিখিত কোড দিয়ে করা যেতে পারে:

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)
বিভক্ত উদাহরণ
'test' 10,000
'train' 50,000
  • বৈশিষ্ট্য গঠন :
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,
})
  • বৈশিষ্ট্য ডকুমেন্টেশন :
বৈশিষ্ট্য ক্লাস আকৃতি ডিটাইপ বর্ণনা
ফিচারসডিক্ট
মোটা_লেবেল ক্লাসলেবেল int64
আইডি পাঠ্য স্ট্রিং
ইমেজ ছবি (৩২, ৩২, ৩) uint8
লেবেল ক্লাসলেবেল int64
noise_label ক্লাসলেবেল int64
worker_id টেনসর int64
worker_time টেনসর float32

ভিজ্যুয়ালাইজেশন

  • উদ্ধৃতি :
@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}
}