No abstract

No abstract

Background: This study presents an intelligent table tennis e-training system based on a neural network (NN) model that recognizes data from sensors built into an armband device, with the component values (performances scores) estimated through principal component analysis (PCA). Methods: Six expert male table tennis players on the National Youth Team (mean age 17.8 ± 1.2 years) and seven novice male players (mean age 20.5 ± 1.5 years) with less than 1 year of experience were recruited into the study. Three-axis peak forearm angular velocity, acceleration, and eight-channel integrated electromyographic data were used to classify both player level and stroke phase. Data were preprocessed through PCA extraction from forehand loop signals. The model was trained using 160 datasets from five experts and five novices and validated using 48 new datasets from one expert and two novices. Results: The overall model’s recognition accuracy was 89.84%, and its prediction accuracies for testing and new data were 93.75% and 85.42%, respectively. Principal components corresponding to the skills “explosive force of the forearm” and “wrist muscle control” were extracted, and their factor scores were standardized (0–100) to score the skills of the players. Assessment results indicated that expert scores generally fell between 60 and 100, whereas novice scores were less than 70. Conclusion: The developed system can provide useful information to quantify expert-novice differences in fore-hand loop skills.

Image view synthesis has seen great success in reconstructing photorealistic visuals, thanks to deep learning and various novel representations. The next key step in immersive virtual experiences is view synthesis of dynamic scenes. However, several challenges exist due to the lack of high-quality training datasets, and the additional time dimension for videos of dynamic scenes. To address this issue, we introduce a multi-view video dataset, captured with a custom 10-camera rig in 120FPS. The dataset contains 96 high-quality scenes showing various visual effects and human interactions in outdoor scenes. We develop a new algorithm, Deep 3D Mask Volume, which enables temporally-stable view extrapolation from binocular videos of dynamic scenes, captured by static cameras. Our algorithm addresses the temporal inconsistency of disocclusions by identifying the error-prone areas with a 3D mask volume, and replaces them with static background observed throughout the video. Our method enables manipulation in 3D space as opposed to simple 2D masks, We demonstrate better temporal stability than frame-by-frame static view synthesis methods, or those that use 2D masks. The resulting view synthesis videos show minimal flickering artifacts and allow for larger translational movements.

No abstract