this? Key words Main idea / Key insights Veriﬁcation / Evaluation Discussion Things to check next Voeikov, R., Falaleev, N., Baikulov, R. (2020). TTNet: Real-time temporal and spatial video analysis of table tennis arXiv https://arxiv.org/abs/2004.09927 TTNet: Real-time temporal and spatial video analysis of table tennis Temporal-Spatial analysis on table tennis. The authors use multitask learning to predict in game events. CVPR 2020 paper. TTNet, table tennis, CV, CNN, ResNet A network comprised of two-stage ball detection, semantic segmentation and event spotting is used to ultimately predict events through the use of multitask learning Ball detection -> RMSE, ball presence i.e accuracy Event spotting -> (Smoothed )Percentage of correct Events Semantic segmentation -> Intersection over Union I wonder how accuracy/speed of ball detection changes as we increase the amount of stages (will three or four stages be better than the proposed two?). What about DenseNet or ResNeXt instead of ResNet? Ubernet, DeepBall, Integrated recognition localization and detection using convolutional networks, Convolutional neural networks based ball detection in tennis games looked interesting.