轻量型简约人脸线条画生成方法.

At present, portrait painting robots generally have the problems of low portrait quality, single artistic style, long rendering time and high computational complexity. Aiming at these problems, a simple facial line painting generation method based on knowledge distillation was proposed. First of all, using the large model as the teacher network, through the idea of knowledge distillation, this paper proposed a distillation loss to guide the training process of the light-weight student network, so that the light-weight model could achieve a better generation effect. Secondly, style loss was introduced to improve the continuity of generated lines. The experimental results show that this method can generate simple and beautiful face line drawings, and the robot can complete the drawing in 2 min. Compared with the teacher model, the amount of calculation is reduced by about 65.3%, and the amount of parameters is reduced by about 97.1%.This proposed method which is significantly better than the benchmark student model. [ABSTRACT FROM AUTHOR]