To download our collected noisy labels, click the link below:
The noisy label zip file include:
(1)
CIFAR-10_human.pt (human annotated labels on CIFAR-10 train images);
(2)
CIFAR-100_human.pt (human annotated labels on CIFAR-100 train images).
import torch
# For CIFAR-10N noisy labels
noise_label = torch.load('CIFAR-10_human.pt')
clean_label = noise_label['clean_label']
worst_label = noise_label['worse_label']
aggre_label = noise_label['aggre_label']
random_label1 = noise_label['random_label1']
random_label2 = noise_label['random_label2']
random_label3 = noise_label['random_label3']
# For CIFAR-100N noisy labels
noise_label = torch.load('CIFAR-100_human.pt')
clean_label = noise_label['clean_label']
noisy_label = noise_label['noisy_label']
For more details, please refer to a starter code at Github, including:
Noisy label sets, Dataloader of CIFAR-10N, CIFAR-100N and training with Cross Entropy loss.
If you use this dataset or our reproduced results, please cite our paper below:
Learning with Noisy Labels Revisited: A Study Using Real-World Human Annotations.
Jiaheng Wei*, Zhaowei Zhu*, Hao Cheng, Tongliang Liu, Gang Niu, and Yang Liu. (*: equal contributions)
The BibTex information is detached as:
@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}
}
Please contact us via {yangliu, jiahengwei, zwzhu}@ucsc.edu, if you have any concerns regarding this dataset.