Skip to content

Latest commit

 

History

History
62 lines (41 loc) · 1.41 KB

README.md

File metadata and controls

62 lines (41 loc) · 1.41 KB

NGC: A Unified Framework for Learning with Open-World Noisy Data

[Paper].

Introduction

We propose a new graph-based framework, namely Noisy Graph Clean- ing (NGC), which collects clean samples by leveraging ge- ometric structure of data and model predictive confidence.

流程

Results

Hyper parameters can be found in configs folder.

Datasets

We provide all necessary datasets to reproduce results declared in paper. Please refer Baidu(code:a7d4).

Example usage

Requirements and Installation

  • Install requirements by
cd papers/ICCV2021-NGC
pip install -r requirements.txt
  • Build c++ library
cd impls
python setup.py build_ext -i
cd ..

Usage

  • Run
python run.py --config configs/cifar10_sym.py
  • Change random seed
python run.py --config config/cifar10_sym.py --work_dir work_cifar10_sym --seed 1001

Citing NGC

If you find NGC useful for your research, please consider citing the paper as follows:

@article{2021NGC,
  title={NGC: A Unified Framework for Learning with Open-World Noisy Data},
  author={ Wu, Zhi Fan  and  Wei, Tong  and  Jiang, Jianwen  and  Mao, Chaojie  and  Tang, Mingqian  and  Li, Yu Feng },
  year={2021},
}