philferriere / tf-dsb-18

Team Three_Explorers 2018 Data Science Bowl Dual U-Net Nuclei Segmentation Solution

2018 Data Science Bowl

Dual U-Net Nuclei Segmentation solution.

Data

Download and unzip in a local folder (dataset_root):

https://raw.githubusercontent.com/AakashSudhakar/2018-data-science-bowl/master/compressed_files/stage1_test.zip https://github.com/lopuhin/kaggle-dsbowl-2018-dataset-fixes/archive/master/stage1_train.zip

Trained Models

Trained models are in model_train_unet_256x256_nuclei_2px_contours_cyclic_lr.zip

Environment

Create a conda environment as defined in dlwin36.yml or requirements.txt

Notebooks

Data Preparation

Run the dataset_prep.ipynb jupyter notebook after having set dataset_root to the proper dataset folder.

Model Training & Testing

Run the model_train_unet_256x256_nuclei_2px_contours_cyclic_lr.ipynb jupyter notebook after having set dataset_root to the proper dataset folder.