Learning with Instance-Dependent Label Noise: A Sample Sieve Approach

This code is a PyTorch implementation of our paper "Learning with Instance-Dependent Label Noise: A Sample Sieve Approach" accepted by ICLR2021.

The code is run on the Tesla V-100.

Prerequisites

Python 3.6.9

PyTorch 1.2.0

Torchvision 0.5.0

Steps on Runing CORES on CIFAR 10

Step 1:

Download the datset from http://www.cs.toronto.edu/~kriz/cifar.html Put the dataset on data/

Install theconf by pip install git+https://github.com/wbaek/theconf.git

Step 2:

Run CORES (Phase 1: Sample Sieve) on CIFAR-10 with instance 0.6 noise:

CUDA_VISIBLE_DEVICES=0 python phase1.py --loss cores --dataset cifar10 --model resnet --noise_type instance --noise_rate 0.6

Step 3:

Run CORES (Phase 2: Consistency Training) on the CIFAR-10 with instance 0.6 noise:

cd phase2
CUDA_VISIBLE_DEVICES=0,1 python phase2.py -c confs/resnet34_ins_0.6.yaml --unsupervised

Both Phase 1 and Phase 2 do not need pre-trained model.

Citation

If you find this code useful, please cite the following paper:

@article{cheng2020learning,
  title={Learning with Instance-Dependent Label Noise: A Sample Sieve Approach},
  author={Cheng, Hao and Zhu, Zhaowei and Li, Xingyu and Gong, Yifei and Sun, Xing and Liu, Yang},
  journal={arXiv preprint arXiv:2010.02347},
  year={2020}
}

References

The code of Phase 2 is based on https://github.com/ildoonet/unsupervised-data-augmentation