Post-Processing Error Correction in Virtual Try-on Pose Detection: A Comparative Study of Multivariate Adaptive Regression Splines and Extra Trees Algorithms

Digital try-on systems essentially depend on pose detection since their accuracy defines the realism of the virtual fitting. This work presents a custom model for extra trees and Multivariate Adaptive Regression Splines-based posture correction. Built with pose key point datasets, the model searches joint position errors to raise detection accuracy. With an accuracy of 89%, the Extra Trees model performed rather poorly; the MARS model performed somewhat better with an accuracy of 90%. These results let MARS exactly fix posture estimate mistakes, so enhancing digital try-on accuracy and allowing intelligent pose refinement for special virtual try-on settings.