/iglovikov_segmentation

Semantic segmentation pipeline using Catalyst.

Primary LanguagePythonMIT LicenseMIT

Installation

Run pip install -r requirements.txt to install all requred libraries.

Data preparation

Cityscapes

  1. Use this guide to prepare the data.
  2. Put prepared data to the data folder or modify path to the train and test files in the config file.

Training

Almost all trainining parameters are specified in the config file. You may check examples in the configs folder.

You may star training, running:

python -m src.train <path_to_config_file>

Inference

Run:

python -m src.inference -c <path to config> \
                        -i <path to input images> \
                        -o <output_path> 
                        -w <path to pth file with weights> \
                        -b <batch size> \
                        -j <num_cpu_threads>

Contributing

To contribute follow this guide.