label-noise-robust-training_kw2t's Introduction

Noise Robust Learning with Hard Example Aware for Pathological Image classification

Implementation detail for our paper "Noise Robust Learning with Hard Example Aware for Pathological Image classification", this code also includes further resaerch beyound this paper.

Citation

Please cite this paper in your publications if it helps your research:

@inproceedings{peng2020noise,
  title={Noise Robust Learning with Hard Example Aware for Pathological Image classification},
  author={Peng, Ting and Zhu, Chuang and Luo, Yihao and Liu, Jun and Wang, Ying and Jin, Mulan},
  booktitle={2020 IEEE 6th International Conference on Computer and Communications (ICCC)},
  pages={1903--1907},
  year={2020},
  organization={IEEE}
}

Dataset

DigestPath 2019: https://digestpath2019.grand-challenge.org/Dataset/

Colorectal dataset (contributed by this paper):contains 4198 microscopy images, which are distributed as follows: adenoma, polyp, adenocarcinoma, gastrointestinal stromal tumor, and neuroendocrine tumor

Envs

Pytorch 1.0
Python 3+
cuda 9.0+

Training

$ cd code/
# train label noise dataset and record training history
$ python iter_train.py --cached_data_file='pickle_data/digest_20.p'
# uncomment "detect label noise" code block in iter_train.py and apply label noise detect algorithm
$ python iter_train.py 
# label correction
$ python pre_iter.py
# train neural network on processed label noise dataset (apply different loss functions)
$ python train.py
# co-teaching training
$ python co-teaching.py

Recommend Projects