Temporal Refinement in Stratified Motion Diffusion: Utilizing GRU for Smoothed Full-Body Prediction

This article describes the online inference stage of the model, which predicts 3D avatars in a frame-by-frame manner using a sliding window approach