Efficient Parallel Implementation of Active Appearance Model Fitting Algorithm on GPU

Figure 3

(a) The design idea of calculating the shape vector, 𝑆. Here, 𝑀 is the number of vertices in 𝑆, which equals the number of threads involved in the calculation. (b) The design idea of the remaining four subprocedures focuses on computing 𝑔_𝑠 and 𝑔_𝑚.

(a)

(b)