Attention-Based Unsupervised Sketch Colorization of Anime Avatar
Sicong Zhang, Jiting Zhou · Proceedings of the 2022 2nd International Conference on Control and Intelligent Robotics · 2022
In recent years, when it comes to the task of applying painting style to animation sketch, the method of deep learning has become a research hotspot in the field of animation sketch coloring. Although the effect of some previous coloring methods is effective, it is easy to mix colors in the facial area and dye the upper skin color of the surrounding color blocks, which affects the performance of the final effect. In the method of this paper, the attention mechanism module is added, combined with the unsupervised generated attention network with adaptive layer instance normalization, which can make the model learn the key coloring in the facial features area through the feature map, so as to reduce the surrounding color mixing and help to improve the coloring effect. At the same time, due to the use of unsupervised attention generation network, the whole process is fast and automatic. This method can be used for quick coloring of user-defined animation avatars. It is very friendly to novice users who don't know how to color at all. The final experiment also proves the superiority of this method in facial region coloring.