This repository contains the method we developped to participate to the MMSports 2022 ball 3D localization challenge. It is described in this report. The source code is available under the CeCILL 2.1 license.
Create a virtual environment:
virtualenv venv
source venv/bin/activate
Install the dependancies:
pip install -r requirements.txt
Please follow the instructions available on the challenge repository to download and generate the dataset pickle files.
To train the model, use the command line:
CUBLAS_WORKSPACE_CONFIG=:16:8 python train.py
A model is saved every 10 epochs in the output directory.
To evaluate the model, use the command line:
python eval.py dataset_pickle_file_path model_path
You can use the --visualization command line option to display images of the estimations.
The model used for our submission to the evaluation server is available in the models directory.