%0 Conference Proceedings
%A Martin, Pierre-Etienne
%A Benois-Pineau, Jenny
%A Péteri, Renaud
%A Morlier, Julien
%+ Department of Comparative Cultural Psychology, Max Planck Institute for Evolutionary Anthropology, Max Planck Society
%T Three-stream 3D/1D CNN for fine-grained action classification and
segmentation in table tennis : 
%G eng
%U https://hdl.handle.net/21.11116/0000-000A-4B31-B
%R 10.1145/3475722.3482793
%F OTHER: hal-03353945
%D 2021
%B MMSports '21
%Z date of event: 2021-10-20 - 2021-10-24
%C Chengdu,  China (Online)
%X This paper proposes a fusion method of modalities extracted from video<br>through a three-stream network with spatio-temporal and temporal convolutions<br>for fine-grained action classification in sport. It is applied to TTStroke-21<br>dataset which consists of untrimmed videos of table tennis games. The goal is<br>to detect and classify table tennis strokes in the videos, the first step of a<br>bigger scheme aiming at giving feedback to the players for improving their<br>performance. The three modalities are raw RGB data, the computed optical flow<br>and the estimated pose of the player. The network consists of three branches<br>with attention blocks. Features are fused at the latest stage of the network<br>using bilinear layers. Compared to previous approaches, the use of three<br>modalities allows faster convergence and better performances on both tasks:<br>classification of strokes with known temporal boundaries and joint segmentation<br>and classification. The pose is also further investigated in order to offer<br>richer feedback to the athletes.<br>
%K Computer Science, Computer Vision and Pattern Recognition, cs.CV,Computer Science, Artificial Intelligence, cs.AI,Computer Science, Human-Computer Interaction, cs.HC,Computer Science, Learning, cs.LG,Computer Science, Multimedia, cs.MM
%B MMSports'21: Proceedings of the 4th International Workshop on Multimedia Content Analysis in Sports
%P 35 - 41