Researchers have developed TT4D, a large-scale dataset and reconstruction pipeline for table tennis gameplay captured from monocular videos. This dataset includes over 140 hours of reconstructed gameplay with detailed annotations such as 3D ball positions, ball spin, and human meshes. The novel 'lift-first' reconstruction pipeline first estimates the 3D ball trajectory before segmenting game shots, enabling more reliable reconstruction even with occlusions and varied camera angles. AI
Summary written by gemini-2.5-flash-lite from 1 source. How we write summaries →
IMPACT Provides a new foundation for virtual replay, player analysis, and robot learning in sports.
RANK_REASON Academic paper detailing a new dataset and reconstruction pipeline for 4D table tennis gameplay. [lever_c_demoted from research: ic=1 ai=1.0]