Use Conda or Python virtual environment to install python with 3.11 version.
conda create --name <venv_name> 'python==3.11.3'
conda activate <venv_name>
python3 -m pip install --user virtualenv
python3.11 -m venv <venv_name>
If python3.11 is not recognized, but python3 points to Python 3.11:
python3 -m venv <venv_name>
source ./<venv_name>/bin/activate # for Unix/Linux/MacOS
or
.\<venv_name>\Scripts\activate # for Windows
pip install -r requirements.txt
The following script will create an ./input
directory in the root of the repository, where it will download the necessary data for training and visualisation.
bash download_all_files.sh
The following script will train the model, create a ./experiment
directory in the root of the repository where the model weights will be saved. The loss.pdf
file with the training plot will be saved in the same directory.
python train_model.py
You can skip train step and download the test model to the ./experiment
directory using the following command:
mkdir ./experiment
curl -L $(yadisk-direct https://disk.yandex.ru/d/S5HvchhPkvUd6g) -o ./experiment/model.pth
Open visualize.ipynb
and run all cells.
For correct working, directory names must not be changed!