A Hierarchical Static-Dynamic Encoder-Decoder Structure for 3D Human Motion Prediction with Residual CNNs

Figure 4

The hierarchical structure based on v-CMUs. (a) Our hierarchical asymmetric structure; the encoded latest frame only passes through 2 v-CMU blocks. (b) Hierarchical symmetric structure, introduced by Liu et al. [20]. The encoded latest frame passes through 4 v-CMU blocks.

(a)

(b)