kaggle salt identifcation challenge. Top 7% solution (207 ouut of 3234)

this solution is based on u-net architecture (https://arxiv.org/abs/1505.04597) with resnet backbone trained on 5 folds.

Name		Name	Last commit message	Last commit date
Latest commit History 2 Commits
configs		configs
dataset		dataset
pytorch_zoo		pytorch_zoo
README.md		README.md
__init__.py		__init__.py
config.py		config.py
eval_aggregator.py		eval_aggregator.py
evaluator.py		evaluator.py
generate_folds.py		generate_folds.py
mafat_preparation.py		mafat_preparation.py
salt_test_dataloader.ipynb		salt_test_dataloader.ipynb
salt_viz.ipynb		salt_viz.ipynb
submission.csv		submission.csv
train.py		train.py
train_plateau.py		train_plateau.py
transforms.py		transforms.py
transforms_light_aug.py		transforms_light_aug.py
utils.py		utils.py

rrkotik/k_salt_identification