Data set

Tennis actions video dataset


Aldo Hoekstra


Automatic video recognition methods have high potential value for analysis of sports such as tennis. The current state of affairs at the KNLTB is that humans manually index potentially relevant parts of a video recorded training or match. With the current huge stream of videos, there is a demand for automatic tennis video indexing.

The aim is to create a massive data set of annotated tennis videos and train a deep network to recognize specific tennis actions automatically. The KNLTB will make a large collection of 600-800 tennis videos available. We will then proceed to annotate all these videos with a vocabulary of actions and then evaluate the current state-of-the-art in video analysis. We will make all videos, all annotations, all trained models and all software available to the community. The benefit for the scientific community is a new, realistic, open, dataset with reproducible baseline results.


Gerelateerd nieuws

Toon alles

Gerelateerde data sets

Toon alles