Figure 1.

Comparison of part-based methods: (a) Dividing images into rectangle parts; (b) part alignment model; (c) proposed dynamic and accurate attention parts based on body regions; and (d) explicitly and stable correspondence between parts.