2020-Sports-Video-Classification-Task

Welcome

Dear participant,

Your subscribed to the task Sports Video Classification: Classification of Strokes in Table Tennis for MediaEval 2020. Thank you for your interest and we hope for your participation. The task description can be found there.
The task overview paper can be found here.
Working Notes Paper fomat here.
To have direct access to the github repository, please send us an e-mail with github username.

To have access to the data, particular conditions need to be accepted.

Please read them carefully. The user has, for example to obscure the faces (blurring, black banner, etc.) before any publication and destroy the data by October 1st 2021. The video have been partially anonymized but on some frames the face of the player might not be blurred. You will receive the login and password once you accept those particular conditions. In order to receive access to the dataset, please send us an email at mediaeval.sport.task (at) diff.u-bordeaux.fr with « I Accept the Usage Conditions » in the email body with your institutional mail address. We thank you in advance for your cooperation.

Download

On linux, in a terminal, you can use the following command to download the files (32.3GB). login and password should be replaced according to the instructions you will receive once you accepted the particular conditions. :

wget --user *login* --password "*password*" -r --no-parent
https://www.labri.fr/projet/AIV/MediaEval/2020/data/

It should create a directory “www.labri.fr/projet/AIV/MediaEval/2020/data/”, with two subdirectories: “train” and “test”. In the train directory, you should have 77 mp4 files and 77 xml files. In the test directory, you should have 28 mp4 files and 28 xml files. You can then check that files were downloaded correctly with the following commands:

cd www.labri.fr/projet/AIV/MediaEval/2020/data/
md5sum -c MD5SUMS

Data Organization

The dataset is split in two: train and test.
In each directory, there are several videos (in MP4 format). Each video may contain several actions. Each video file is accompanied with an XML file describing the actions present in the video. For each action, the starting and ending frames, and the stroke class/move attribute are specified.

There are 20 stroke classes to recognize:

In the train dataset XML files, the stroke class/move attribute is set to one of these 20 classes.
In the test dataset XML files, the stroke class/move attribute is purposely set to an invalid value (“Unknown”) and should be updated by the participants to one of the 20 valid classes.

Submission

Participants may submit up to 5 runs to mediaeval.sport.task (at) diff.u-bordeaux.fr
For each runs, they must provide one XML file per video file, with the actions associated to the recognize stroke class. Runs may be submitted as an archive (zip or tar.gz file) with each run in a different 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.

To check the xmls format submitted, save your xmls in a folder without changing the xmls from test folder and then run:

python3 verif_xml_files.py folder_with_xmls_to_submit

Submissions will be evaluated in terms of accuracy per class of a stroke and of global accuracy. Different type of confusion matrix will also be sent to you to help you work on your working note paper.

Crisp Project

Pierre-Etienne Martin, Jenny Benois-Pineau, Renaud Péteri, Julien Morlier. 2020. Fine grained sport action recognition with siamese spatio-temporal convolutional neural networks. Multimedia Tools and Applications (19 Apr 2020).

Pierre-Etienne Martin, Jenny Benois-Pineau, Renaud Péteri, and Julien Morlier. 2019. Optimal choice of motion estimation methods for fine-grained action classification with 3D convolutional networks. In ICIP 2019. IEEE,554–558.

Gül Varol, Ivan Laptev, and Cordelia Schmid. 2018. Long-Term Temporal Convolutions for Action Recognition. IEEE Trans. Pattern Anal. Mach. Intell. 40, 6 (2018), 1510–1517.

Joao Carreira and Andrew Zisserman. 2017. Quo Vadis, Action Recognition? A New Model and the Kinetics Dataset. CoRR abs/1705.07750 (2017).

Chunhui Gu, Chen Sun, Sudheendra Vijayanarasimhan, Caroline Pantofaru, David A. Ross, George Toderici, Yeqing Li, Susanna Ricco, Rahul Sukthankar, Cordelia Schmid, and Jitendra Malik. 2017. AVA: A Video Dataset of Spatio-temporally Localized Atomic Visual Actions. CoRR abs/1705.08421 (2017).

Khurram Soomro, Amir Roshan Zamir, and Mubarak Shah. 2012. UCF101: A dataset of 101 hu- man actions classes from videos in the wild. CoRR 1212.0402 (2012).

Task Organizers

You can email us directly at mediaeval.sport.task (at) diff.u-bordeaux.fr

Jenny Benois-Pineau, Univ. Bordeaux, CNRS, Bordeaux INP, LaBRI, UMR 5800, F-33400, Talence, France (jenny.benois-pineau (at) u-bordeaux.fr)
Pierre-Etienne Martin, Univ. Bordeaux, CNRS, Bordeaux INP, LaBRI, UMR 5800, F-33400, Talence, France (pierre-etienne.martin (at) u-bordeaux.fr)
Renaud Péteri, MIA, University of La Rochelle, La Rochelle, France (renaud.peteri (at) univ-lr.fr)
Boris Mansencal, Univ. Bordeaux, CNRS, Bordeaux INP, LaBRI, UMR 5800, F-33400, Talence, France (boris.mansencal (at) labri.fr)
Jordan Calandre, MIA, University of La Rochelle, La Rochelle, France
Julien Morlier, IMS, University of Bordeaux, Talence, France
Laurent Mascarilla, MIA, University of La Rochelle, La Rochelle, France

Task Schedule