noisy-cifar-owaix2quzq 0.0.1

CIFAR datasets with label noise

Noisy CIFAR-10/CIFAR-100

Quickstart

from noisy_cifar import NoisyCIFAR10
from torchvision.datasets import CIFAR10
from torchvision.transforms import ToTensor

root = '/datasets/CIFAR-10'
train_dataset = NoisyCIFAR10(root, 'symmetric', 0.2, transform=ToTensor())
val_dataset = CIFAR10(root, train=False, transform=ToTensor())

Install

pip install noisy-cifar-owaix2quzq

Arguments

• root: path to the original dataset
• noise_type: symmetric / asymmetric / human
• noise_level:
  • (float) [0.0 ~ 1.0] for symmetric & asymmetric
  • (str) {aggregate, random1, random2, random3, worst} for human
• random_seed: default 0
• transform: default None
• target_transform: default None
• download: default False