CB-FQA: A Basketball Defensive Footwork Quality Assessment Dataset (Demo Subset)

The target domain dataset was synchronously recorded by a multi-camera optical motion capture system deployed in a standard basketball court, covering and collecting a series of continuous actions such as sliding steps, crossover steps, sudden stops and blockages in a real offensive and defensive confrontation environment. The original three-dimensional skeleton node coordinates obtained were uniformly mapped to the global world coordinate system and underwent time-domain Gaussian filtering for noise reduction processing. The quality scoring calibration of the data was independently carried out by a senior basketball coaching team with professional qualifications. The evaluation experts strictly combined the center of gravity control of the defenders, the continuity of the force application, and the effectiveness of the spatial pressure on the attackers to conduct frame-by-frame continuous quality scoring for the continuous skeleton sequences. The arithmetic average of the independent scoring by multiple experts was taken as the real label for the regression task.(Due to privacy and copyright restrictions of the athletes, the full CB-FQA dataset cannot be fully open-sourced at this time. This repository provides a demo subset (extracted dummy/sample data) to verify the feasibility of the Phy-DualGAT-ConvLSTM model code.)