Skip to content

Andrei-Aksionov/Unet

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

8 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Welcome to U-Net implementation in PyTorch

Implementation of U-Net architecture in PyTorch that achieves on 'Carvana' dataset accuracy of 96.84 and dice score of 92.66 after training for 3 epochs.

Install

As this project uses pyproject.toml - poetry has to be installed.

Also take a look at the required python version (described in pyproject.toml file).

In order to install all required packages run this command (when you are in the folder with pyproject.toml file).

poetry install

Usage

Parameters

All parameters like model architecture (list of features for each U-Net block) or folder paths where the data is stored can be changed in 'src/config/hyperparameters.py' file.

Data

In order to train model the first step will be to provide data for training and evaluation. As this project is configured data should be stored in 'data/raw' folder with such structure:

  • train_images
  • train_masks
  • val_images
  • val_masks

Name of the corresponding mask should be the same as image name but with '_mask' postfix. For example: in 'train_images' folder there is a file '1.jpg' and in 'train_masks' folder the corresponding masks should have name as '1_mask.gif'.

The data that was used for this project can be download from Kaggle competition.

Training the model

After data is prepared we can start training by executing this command in terminal:

python src/model/train.py

Model's checkpoint

In folder 'models' one can find checkpoint of the model with the highest accuracy score.


Additional: git pre-commit hook

In order to run black formatter and flake8 linter before each commit you need to add them into .git/hooks folder either manually or with helper script:

sh .add_git_hooks.sh`

This script will put pre-commit file into .git/hooks folder of the project and make it executable.

About

No description, website, or topics provided.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published