KaliCalib is a basketball court registration framework developped to participate to the ACM MMSports 2022 camera calibration challenge. The method is described in our paper "KaliCalib: A Framework for Basketball Court Registration" accepted as a challenge paper to MMSports 2022. You can find an ArXiv version here. This 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 prepare the dataset.
To train the model only with the challenge train dataset, you can use the train.sh script.
Evaluation on the test dataset can be run with the eval.sh script. By default, this script loads the provided model_test.pth model but this can be modified in the eval_test.yml config file.
The challenge organizers allowed to use the complete dataset (train, test and validation data) to train a model for an evaluation with the challenge data.
You can achieve this with the train_full_dataset.sh script. The evaluation on the challenge data can be run with the eval_challenge.sh script. By default, this script loads the provided model_challenge.pth model but this can be modified in the eval_challenge.yml config file.
Test model MSE on the test dataset: 107.78 mm.
Challenge model MSE on the challenge dataset: 73.16 mm.