Semantic segmentation using UNet model on Cityscape Segmentation dataset.
Working environments:
python 3.6.9
pip 21.3.1
setuptools 59.6.0
cuda 11.0
ubuntu 18.10
With those settings, I am using
pytorch v1.7.1
to use the GPU.
Install
First clone the repository. Then, inside of the root
directory, execute :
# Minimal setup
pip install .
# or Dev setup
# pip install .[dev]
Finally, install pytorch v1.7.1
and torchvision v0.8.2
using pip
(instructions on the official website).
An example of dataset format is inside the docstring of module segmentation/data.py
You can look at example of usage inside the folder scripts/
.
Otherwise, display the helper description of the main function to get all arguments :
# from scripts/
python main.py --help