solution-6

• Improved augmentation pipe
• moved from reflect padding to edge replication padding

solution-5

• refactored models
• added lovash loss
• added fine tuning
• added resize+pad option

solution-4

added k-fold validation
added averaging over k folds
decoupled network train/valid from postprocessing
added saving model predictions oof train and test
fixed pytorch memory issues

solution-3 simplified

Solution 3 simple (#50)

* initial ref

* solution 3 ported

* Update README.md

* Update README.md

* Update README.md

* dropped old paths

solution-3-cloud

• introduced option to run training on GCP using neptune.

solution-3

• updated hyperparams
• made smarter augmentations

solution-1

• simple end to end pipeline