This code is a PyTorch implementation of the paper "RETHINKING ONE-VS-THE-REST LOSS FOR INSTANCE-
DEPENDENT COMPLEMENTARY LABEL LEARNING"

Prerequisites
-Python 3.6
-PyTorch 1.1
We provide code for generating IDCL dataset from benchmark dataset using pre-trained models. After the paper is accepted, we will provide the specific dataset and pre-trained models.

Instructions：
Run the code:
MNIST,KMNIST,FMNIST:
python -u main.py --learning_rate 1e-3 --batch_size 256  --loss 'ovr_loss' --data 'mnist' --k 3 --epochs 100 --weight_decay 1e-4 --seed 1
CIFAR-10:
python -u main_cifar10.py --learning_rate 0.1 --batch_size 128  --loss 'ovr_loss' --k 3 --epochs 200 --weight_decay 5e-4 --seed 1 --pretrain 'resnet18'
SVHN:
python -u main_svhn.py --learning_rate 5e-4 --batch_size 256 --loss 'ovr_loss' --k 3 --epochs 200 --weight_decay 1e-4 --seed 1 --pretrain 'resnet18'



