SwimXYZ: A large-scale dataset of synthetic swimming motions and videos

What is it about?

This paper introduces SwimXYZ, a synthetic dataset of swimming motions and videos. SwimXYZ aims to help generalize the use of vision-based motion capture in swimming. It features RGB videos annotated with 2D and 3D joints, as well as swimming motions in the SMPL format. See the project page and download the data at: https://g-fiche.github.io/research-pages/swimxyz/

Featured Image

Photo by Gentrit Sylejmani on Unsplash

Photo by Gentrit Sylejmani on Unsplash

Why is it important?

SwimXYZ aims to help generalize the use of motion capture in swimming, and our experiments in swimming stroke clustering and 2D pose estimation show its potential for future applications. While videos provided by SwimXYZ could be used for training 2D and 3D pose estimation models, future works may use motions in the SMPL format for training pose and motion priors or swimming stroke classifiers. SwimXYZ was made publicly available, and we also present our methodology for creating the dataset.