Data Science Bowl

Code for 2018 data science bowl, using UNet and soft dice loss.

Because of my poor laptop, i haven't run the project so there may be many bugs.

FPN and other model will be implement, i'll run this project after next term begins.

data overview

data
├── stage1_sample_submission.csv 
├── stage1_test 
├── stage1_train 
└── stage1_train_labels.csv 

size	num
(256, 256, 3)	334
(256, 320, 3)	112
(520, 696, 3)	92
(360, 360, 3)	91
(1024, 1024, 3)	16
(512, 640, 3)	13
(603, 1272, 3)	6
(260, 347, 3)	5
(1040, 1388, 3)	1

project structure

Kaggle18
├── config.py 
├── data 
│   ├── dataset.py 
│   ├── __init__.py 
│   └── Resize.py 
├── main.py 
├── model 
│   ├── BasicModule.py 
│   ├── checkpoints 
│   ├── __init__.py 
│   └── UNet.py 
├── README.md 
└── utils 
    ├── __init__.py 
    ├── Loss.py 
    ├── saved_loss.csv 
    ├── saved_lr.csv 
    └── util.py 

Acknowledgement

thanks to Andrea for overview of data.

thanks to Yun Chen for data preprocess and net model.

thanks to Stephen Bailey for run-length encoding algorithm.