Sketch-based Anime Hairstyle Editing with Generative Inpainting
Shuyang Luo, Haoran Xie, Kazunori Miyata · 2021
In this work, we propose an interactive sketch-based design interface for anime hairstyle editing. The proposed system adopts the gated convolutional layer in the Generative Adversarial Networks model to achieve the generative editing of anime images. The generator network is based on an encoder-decoder framework with gated convolution to ensure the network can learn the random mask feature in image editing. The discriminator network uses the spectral-normalized architecture to preserve the stability of learning. In this study, we first collect and preprocess the dataset for data training. After that, we construct the user interface based on the pretrained model. Finally, we evaluate the system usage and user experience of the proposed editing interface.