Unknown Class Label Cleaning for Learning with Open-Set Noisy Labels

This is an official PyTorch implementation of Unknown Class Label Cleaning for Learning with Open-Set Noisy Labels. [Paper]

Requirements

• Python 3.7
• PyTorch 1.6.0
• torchvision 0.7.0
• progress
• matplotlib
• numpy

Preparation

Download TinyImageNet datasets as open-set noise.

mkdir data
cd data
wget https://www.dropbox.com/s/1zt54aawvk0245w/Imagenet_resize_full.npy
cd ..

Other datasets will be download automatically.

Usage

Train the network with CIFAR-100 as open-set noise (noise rate = 0.4):

python train.py --gpu 0 --ood cifar100 --out results/run_cifar100 --percent 0.4

The trained model and output will be saved at results/run_cifar100.

For more details and parameters, please refer to --help option.

Run the experiment with different kind of open-set noise (noise rate = 0.4):

python run_all.py

References

• Qing Yu and Kiyoharu Aizawa. "Unknown Class Label Cleaning for Learning with Open-Set Noisy Labels", in ICIP, 2020.