Skip to content

linzhenyuyuchen/RSNA-Intracranial-Hemorrhage-Detection

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

13 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

RSNA Intracranial Hemorrhage Detection

Overview

Identify acute intracranial hemorrhage and its subtypes

https://www.kaggle.com/c/rsna-intracranial-hemorrhage-detection

Installing useful libraries

git clone https://github.com/NVIDIA/apex && cd apex && pip install -v --no-cache-dir --global-option="--cpp_ext" --global-option="--cuda_ext" ./
pip install efficientnet_pytorch
pip install pretrainedmodels

Command

main.py:

python main.py


 --train True (default:False)

 --post_train True (default:False)
 
 --test True (default:False)

 --foldn 0 (default:0, 0,1,2,3,4)

 --tta 5 (default:5, integer above 0)

 --make_fold True (default:False, 5 folds with seed=10)

windowing.py

python windowing.py

 --mode train (default='test', 'train' or 'test')
 --input /input/csv_dicom/ (directory in which there are csv and dicom)
 --output /../../ (to which save png files)
 --p 4 (default:4,integer depending on your cpu)

analysis

  • confusion matrix

  • Receiver Operating Characteristic

  • Stacking

Run

  1. install libraries in # Libraries

  2. modify # Parameters

  3. modify # Input

  4. train : run command python main.py --train True --foldn 0

  5. test : run command python main.py --test True --foldn 0 --tta 5

  6. make folds : run command python main.py --make_fold True

Output

  1. output directory : ./output/model_name/

  2. k-fold cross validation : ./fold/

  3. submission csv file : ./submission/

Ensemble

stage2

  1. se_resnet50

Post Train

Test : Fold 0 1 2 3 4

Test time : 22 min / fold (if tta , * 5 )

  1. se_resnext50_32x4d

(Post Train)

Test : Fold 0 1 2 3 4

(Stage 2 Dataset Fold3)

Test Time : 44 min / fold (if tta , * 5 )

  1. efficientnetb0

Post Train

Test : Fold 1

  1. efficientnetb2

Post Train : 43 min

Test : Fold 2

  1. se_resnext50_32x4d (512 JPG)
    1. se_resnext50_32x4d (Stage 2 Dataset)

Timeline

October 28, 2019 - Entry deadline. You must accept the competition rules before this date in order to compete.

October 28, 2019 - Team Merger deadline. This is the last day participants may join or merge teams.

November 4, 2019 - Stage 1 ends & Model upload deadline.

November 5, 2019 - Stage 2 begins. New test set uploaded.

November 13, 2019 - Stage 2 ends & Final submission deadline.

November 25, 2019 - Solutions & Other Winners Obligations due from winners.

December 1-6, 2019 - RSNA 2019 Conference in Chicago, IL

About

Code for Kaggle Competition -- RSNA Intracranial Hemorrhage Detection

Topics

Resources

Stars

Watchers

Forks

Contributors 2

  •  
  •