Action detection and classification is one of the main challenges in visual content analysis and mining. Sport video analysis has been a very popular research topic, due to the variety of application areas, ranging from multimedia intelligent devices with user-tailored digests, up to analysis of athletes' performances. Datasets with sport activities are available now for benchmarking of methods. A large amount of work is also devoted to the analysis of sport gestures using motion capture systems. However, body-worn sensors and markers could disturb the natural behaviour of sports players. Furthermore, motion capture devices are not always available for potential users, be it a University Faculty or a local sport team. Coming years will build upon the basic "Sports Video Annotation: Detection of Strokes in Table Tennis" task ofered in 2019. The ultimate goal of this research is to produce automatic annotation tools for sport faculties, local clubs and associations to help coaches to better assess and advise athletes during training
Action detection and classification is one of the main challenges in
visual content analysis and mining [
Other datasets such as UCF-101 [
Because TTStroke-21 is constituted of videos with identifiable players of Table Tennis, this dataset is subject to particular conditions in order to respect the personal data and privacy of the players. These Special Conditions apply to the use of Images generated in the framework of the program Sports video annotations: classiifcation of strokes in table tennis, for the implementation of the MediaEval program. They constitute the specific usage agreement referred to in the Usage agreement for the MediaEval 2019 Research Collections, signed between the User and the University of Delft. The full and complete acceptance, without any reservation, of these Special Conditions is a mandatory prerequisite for the provision of the Images as part of the MediaEval programme. A complete reading of these conditions are necessary and engage the user, for example, to obscure the faces (blurring, black banner, etc.) before any publication and to destroy the data by October 1st 2020. 3
In MediaEval 2019, we deliver a subset of TTStroke-21 data set which has been specifically recorded in a sport faculty facility using a light-weight equipment, such as GoPro cameras. It is constituted of player-centred videos recorded in natural conditions without markers or sensors, see Fig 1. It comprises 20 table tennis strokes a. Video acquisition
b. Annotation platform c. Dataset samples classes, i.e. 8 services, 6 ofensive strokes and 6 defensive strokes. This taxonomy was designed with professional table tennis teachers.
All videos are recorded in MP4 format.
The organisation of the delivered data is as follows: • The provided dataset is split into two subsets: i)training set and ii)test set; • In each directory, there are several videos (in MP4 format) and each video may contain several actions; • Each video file is accompanied with a XML file describing the actions present in the video; • For each action there are 3 attributes: the starting frame, the ending frame, and the stroke class; • In the train set XML files, all the attributes are specified but in the test set XML files, only the starting and ending frames are specified while the stroke class attribute is purposely set to an invalid value ("Unknown") and should be updated by the participants to one of the 20 valid classes. . 4
The Sport Video Annotation task consists in assigning a label from a given taxonomy of 20 classes of Table Tennis strokes to each action delimited by starting frame and ending frame in each test video file.
Participants may submit up to four runs. For each runs, they must provide one XML file per video file, with the actions associated to the recognised stroke class. Runs may be submitted as an archive (zip or tar.gz file) with each run in a diferent directory. Participants should also indicate if any external data (other dataset, pretrained networks, ...) was used to compute their runs. The task is considered fully automatic. Once the video are provided to the system, results should be produced without any human intervention. 5
In MediaEval 2019 we propose a light-weight classification task. It consists in classification of table tennis strokes which temporal borders are supplied in the XML files accompanying each video file. Hence for each test video the participants are invited to produce an xml file in which each stroke is labelled accordingly to the given taxonomy. This means that the label "unknown" has to be replaced by the label of the stroke class which participant’s system has assigned. All submissions will be evaluated in terms of per-class accuracy(PCA) and of global accuracy (GA).The PCA is computed for each i-th class as:
PCAi = T Pi /(Nдt i ) Here T Pi is the number of True Positives, i.e. correctly labelled, by the participant’s system, strokes for the given i-th class,Nдt i is the number of recorded strokes of the i-th class in the test dataset.
GA = T P /(Nдt ) Here T P = Í T Pi is the number of correctly labelled strokes for the whole dataset, and Nдt is the number of strokes in the ground truth - the whole test set. (1) (2) 6
Participants are welcome to share their dificulties and the results even if they are negative. Better understanding of automatic classification methods are easier when all aspects of the methods are shared.
Thank you for participating at MediaEval 2019 and more specifically to our task: "Sports Video Annotation: Detection of Strokes in Table Tennis". We look forward to seeing you at the MediaEval Workshop and to discussing your results further.
We would like to thank all the players and annotators who contributed to TTStroke-21, Alain Coupet for his dedication to the project, Xavier Daverat and Chantal Durand for their help on the Particular Conditions formulation.
This work was supported by Region of Nouvelle Aquitaine grant CRISP and Bordeaux Idex Initiative. Sports Video Annotation: Detection of Strokes in Table Tennis