This is the training code of the 1st place solution for Kaggle HuBMAP competition.
Solution summary : https://www.kaggle.com/c/hubmap-kidney-segmentation/discussion/238198
Inference code : https://www.kaggle.com/tikutiku/hubmap-tilespadded-inference-v2?scriptVersionId=59475269
#HARDWARE: (The following specs were used to create the original solution)
Ubuntu 18.04 LTS (2TB boot disk)
Core i7-10700K
64GB memory
1 x NVIDIA GeForce RTX3090
#SOFTWARE: (python packages are detailed separately in requirements.txt)
Python 3.7.3
CUDA 11.2
cuddn 8.1.1.33
nvidia drivers v.460
#Usage
- data preparation for kaggle train data
cd src/01_data_preparation/01_01
python data_preparation_01_01.py
cd src/01_data_preparation/01_02
python data_preparation_01_02.py
- train
cd src/02_train
python train_02.py
- pseudo-label generation
cd src/03_generate_pseudo_labels/03_01_pseudo_label_kaggle_data
python pseudo_label_kaggle_data_03_01.py
cd src/03_generate_pseudo_labels/03_02_pseudo_label_dataset_a_dib
python pseudo_label_kaggle_data_03_02.py
cd src/03_generate_pseudo_labels/03_03_pseudo_label_hubmap_external
python pseudo_label_hubmap_external_03_03.py
- data preparation for pseudo-labeled data
cd src/04_data_preparation_pseudo_label/04_01_kaggle_data
python data_preparation_pseudo_label_04_01.py
cd src/04_data_preparation_pseudo_label/04_02_kaggle_data_shift
python data_preparation_pseudo_label_04_02.py
cd src/04_data_preparation_pseudo_label/04_03_dataset_a_dib
python data_preparation_pseudo_label_04_03.py
cd src/04_data_preparation_pseudo_label/04_04_dataset_a_dib_shift
python data_preparation_pseudo_label_04_04.py
cd src/04_data_preparation_pseudo_label/04_05_hubmap_external
python data_preparation_pseudo_label_04_05.py
cd src/04_data_preparation_pseudo_label/04_06_hubmap_external_shift
python data_preparation_pseudo_label_04_06.py
cd src/04_data_preparation_pseudo_label/04_07_carno_zhao_label
python data_preparation_pseudo_label_04_07.py
cd src/04_data_preparation_pseudo_label/04_08_carno_zhao_label_shift
python data_preparation_pseudo_label_04_08.py
- train with kaggle train + pseudo-labeled data
cd src/05_train_with_pseudo_labels
python train_05.py