简约风科技流程图插画素材

Draw a minimal CVPR/ICLR-style diagram for CVAE-based prediction of UV contact maps from hand–object trajectories.A Trajectory Encoder extracts latent features from the input sequence.During training, a Posterior Network uses trajectory features and ground-truth UV maps to infer a latent distribution, while a Conditional Prior predicts a latent distribution from the trajectory alone.A UV Map Decoder generates a dense UV contact map from the trajectory features and sampled latent code.Show reconstruction loss and KL divergence, and distinguish training and inference paths.Use clean rectangular blocks, clear arrows, and an academic layout.